A Review of near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010

Abstract

Muscle foods (meat and fish) are very important from the perspective of human nutrition and economic activity, both nationally and internationally. At a research and development level, major efforts continue to be focussed on improving the quantity and quality of raw and processed muscle food types available on the market and also to monitor their compliance with compositional, safety and, increasingly, provenance issues. Publications dealing with the development of near infrared (NIP) applications for the analysis of muscle foods (meat and fish) over the period 2005-2010 have been assembled and reviewed. Well-described advantages of NIR spectroscopy suit the food processing industry in terms of operating speed and possible implementation of in-line, on-line or at-line process monitoring; it also has the ability to meet consumer expectations in terms of product quality and safety assurance. These advantages allow food processors to easily monitor and manipulate processing conditions to avoid the production and release of defective products, thereby guaranteeing product quality and enhancing the possibility of repeat purchasing by customers. For public regulatory organisations which have responsibilities to both food producers and consumers, NIR technology may be able to contribute efficiently to these aims. Interrogation of NIR datasets by increasingly powerful and sophisticated chemometric techniques continues to improve calibration robustness and accuracy while the appearance of extensive suites of algorithms in commercially-available software packages helps in their deployment. The aim of this review is to provide an update on work in these areas which has been published in the period from 2005 to 2010. While targeted chiefly at researchers active in the field, it should also be of relevance to technical personnel in the meat and fish industries and to regulatory personnel.

Full text

JOURNAL
OF
NEAR
INFRARED
SPECTROSCOPY
1
ISSN: 0967-0335 © IM Publications LLP 2011
doi: 10.1255/jnirs.xxx All rights reserved
Global demand for meat as a dietary component continues to
grow, resulting in concomitant increases in livestock produc-
tion.
1
In 2005, approximately 254 million tonnes of meat
(mainly cattle, sheep, goat, pigs and poultry) were produced
globally while 140 million tonnes of fish and aquaculture
outputs were produced in 2007, up from a value of 137 million
tonnes in 2006.
2,3
Meat and fi sh, in whatever form they are
consumed, are important foodstuffs and generally considered
as luxury and nutritious products.
4
Whether for pleasure or
health benefi ts, or both, muscle food consumption (meat and
fi sh) is a cornerstone of the diet of most consumers in the
developed world whereas meeting market demand and gener-
ating profi t is the goal of suppliers: producers, processors,
distributors and retailers. Even though forecasts of growth
in the production and consumption of muscle foods suggest
only slight increases (less than 1% by volume
5
) in the near
future, development and implementation of quality assurance
measures by the food processing industry are more neces-
sary than ever.
6
Because of the well-described diffi culties in
making a confi dent selection of high quality muscle foods at
point of purchase, consumers are reported to be willing to pay
more for meat claiming to be of premium quality either on the
A review of near infrared spectroscopy in
muscle food analysis: 2005–2010
Jittima Weeranantanaphan,
a
Gerard Downey,
b
Paul Allen
b
and Da-Wen Sun
a
a
FRCFT, University College Dublin, National University of Ireland, Agriculture and Food Science Centre, Belfi eld, Dublin 4, Ireland
b
Teagasc Food Research Centre, Ashtown, Dublin 15, Ireland. E-mail: gerard.downey@teagasc.ie
Muscle foods (meat and fi sh) are very important from the perspective of human nutrition and economic activity, both nationally and
internationally. At a research and development level, major efforts continue to be focussed on improving the quantity and quality of raw
and processed muscle food types available on the market and also to monitor their compliance with compositional, safety and, increas-
ingly, provenance issues. Publications dealing with the development of near infrared (NIR) applications for the analysis of muscle foods
(meat and fi sh) over the period 2005–2010 have been assembled and reviewed. Well-described advantages of NIR spectroscopy suit
the food processing industry in terms of operating speed and possible implementation of in-line, on-line or at-line process monitoring;
it also has the ability to meet consumer expectations in terms of product quality and safety assurance. These advantages allow food
processors to easily monitor and manipulate processing conditions to avoid the production and release of defective products, thereby
guaranteeing product quality and enhancing the possibility of repeat purchasing by customers. For public regulatory organisations
which have responsibilities to both food producers and consumers, NIR technology may be able to contribute effi ciently to these aims.
Interrogation of NIR datasets by increasingly powerful and sophisticated chemometric techniques continues to improve calibration
robustness and accuracy while the appearance of extensive suites of algorithms in commercially-available software packages helps
in their deployment. The aim of this review is to provide an update on work in these areas which has been published in the period from
2005 to 2010. While targeted chiefl y at researchers active in the fi eld, it should also be of relevance to technical personnel in the meat
and fi sh industries and to regulatory personnel.
Keywords: near infrared spectroscopy, meat composition, fi sh composition, muscle food quality measurement
Introduction
J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011)
Received: 25 February 2011
■
Revised: 29 March 2011
■
Accepted: 29 March 2011
■
Publication: 2010

2 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
basis of labelling claims or by guarantees implied by the use
of quality marks.
7
Retailers require a profi t margin and facilitate this by mini-
mising purchase prices paid to suppliers while providing
appropriate product quality to consumers.
8
Consequently,
public and regulatory standards and regulations have devel-
oped over time in many countries to protect the rights of both
parties, consumers and suppliers. Consumers are assured
of the nature of purchased foods on the basis of product
descriptions: provenance labelling, nutritional claims and
quality marks. Unfair competition among suppliers through
fraudulent substitution of cheaper products or ingredients
is prohibited. Consumers demand high quality in purchased
meat products and, to ensure this, a consistent method of
quality control and process monitoring at each stage along
the supply chain needs to be established in order to ensure
compliance, to prevent economic fraud, to ensure fair busi-
ness competition among suppliers and to provide consumer
confi dence.
Relevant aspects of muscle food quality are chemical compo-
sition, physical characteristics, sensory attributes, shelf-life
and food safety (including feed regimen). Consumer satisfac-
tion with food products is largely dependent on eating quality.
9
Eating quality of muscle foods arises from chemical proper-
ties (for instance protein, lipid, water content and pH) and from
physical and/or structural properties of meat (for example,
muscle fi bre arrangement and impedance). These can vary
due to, for example, animal species, farming conditions,
feed regimen, post-slaughter handling and, sometimes, the
geographical areas in which products are produced or proc-
essed. Overall, these characteristics may be used to confi rm
wholesomeness, acceptability and to detect adulteration.
10–12
Quality control is facilitated by performing laboratory analyses.
However, traditional analytical approaches are destructive of
the sample, may involve the use of environmentally hazardous
chemical reagents, are time-consuming and often take place
far from the processing operation.
13
All these factors limit the
effectiveness of analytical tools which rely on wet chemistry.
They also prevent on-line evaluation of chemical, physical and
sensory qualities.
14
In contrast, near infrared (NIR) spectros-
copy is non-destructive, economical, simple, rapid, reliable
and environmentally safe.
To illustrate the advantageous role of NIR spectroscopy at
various points along the food supply chain, this paper provides
a review of specifi c applications dealing with meat and fi sh
quality attributes, i.e. quantitative prediction of chemical
composition, physical properties, sensory characteristics and
qualitative aspects including food authentication, food safety
and process control. This review is structured to fi rst provide
some background knowledge of product nature, associated
quality aspects, relevant laws and regulations together with
the principles of NIR spectroscopy and data handling before
discussion of reported applications. Coverage of these reports
is restricted to published accounts publicly available since
2005. For coverage of earlier publications, the interested
reader is referred to other reports.
15–18
Variety of muscle foods
Food from animal sources appears in various forms such as
red and white meat, seafood, processed meat and fi sh and
also as ingredients of other co-products. Beef, lamb, veal
and pork are examples of red meat while chicken and turkey
fall into the white meat category. Smoked and cured meats
(such as ham, bacon, sausages and salami), tinned meat and
hamburgers are all examples of processed products.
19
Seafood
includes fi sh, crustaceans and molluscs.
20
Conventionally, fi sh
is filleted, skinned and trimmed to produce fish fillets for
commercial distribution.
21
Fish can be processed by curing,
salting, smoking, fermenting, drying and irradiation.
22,23
In
addition to the meat species generally consumed by humans,
i.e. pigs, cattle, sheep, goats and poultry, there is a broad
category including many other species which is referred to as
game meat; most of these are regionally produced. Meat from
donkeys, buffalos, camels, yaks, llamas, North American bison,
deer, horses, mules, rabbits, kangaroos, ostriches, domestic
fowl, geese, alligators, invertebrates, crocodiles, emus, ante-
lopes, frogs, termites, elephants, dogs, squirrels, hippopotami,
snakes, locusts, octopi and some insects are examples of such
meats.
24–26
Certain species are mainly produced for export; for
example, 80% of the venison produced in Australia is exported
to European countries and southeast Asia.
27
Quality defi nition
The concept of meat quality refers both to product description
and perception;
28
how quality is defi ned is dependent on who
defi nes it. Participants in the food supply chain (producers,
processors, distributors and consumers) each have different
roles and, hence, defi ne quality differently. Quality attributes
include carcass composition and conformation, palatability,
freshness, health concerns and safety from food-borne
illness.
8,23
In addition, animal welfare issues, specifi c raw
material selection and the way certain meats are produced
(such as free-range or intensively farmed) together with
environmental impact from the production process are also
values of importance to certain segments of the consumer
market.
28
Essential meat quality considerations are listed in
Table 1; for the majority of consumers, eating quality is the
main quality criterion with tenderness being the attribute of
greatest importance.
8,28
Before purchase, customers look
for clues indicating that meat will be tender after cooking.
Meat appearance therefore plays a critical role in purchasing
behaviour; colour and the presence of intramuscular fat
(marbling) are signifi cant attributes.
29,30
Consumers asso-
ciate colour with freshness and the extent of visible fat with
health issues although this latter is dependent on regional
preferences and perceptions.
30
Freshness and safety are of
high importance in seafood products.
31
Freshness can be
defi ned on the basis of different criteria but some relevant
factors include time after the fi sh were caught and delivered
to food retailers, how fi sh were processed and fi nally their
appearance, odour, flavour and texture.
31
Any associated
quality indicator which consumers trust to guarantee specifi c
muscle food quality attributes such as tenderness, species,

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 3
organic or free-range production increases its perceived
value and, if such expectations are fulfi lled, increases the
chance of repeat purchases.
Factors affecting quality of meat and fi sh
The complexity of muscle tissue means that meat quality is
highly dependent on many factors involving both pre-slaughter
and post-mortem conditions. With regard to pre-slaughter
factors, differences in breed, sex, age, weight, environment
and animal nutrition are important. After slaughtering, it
is a requirement to store meat at low temperature in order
to retard the growth of bacteria. Careful control of the
temperature fall in the fi rst 24 h will optimise the proteolytic
mechanisms which convert rigor muscle into tender meat.
This process may continue for at least 14 days and possibly
up to 28 days during which time fl avour development will
also occur.
32
This process is called meat ageing or matura-
tion and can alter meat quality to achieve the desired fi nal
quality. Intramuscular fat is the meat component that has
the greatest infl uence on eating quality (Table 2).
33
In fi shery
products, quality can be altered in a similar fashion involving
both pre- and post-mortem factors. Fish is a highly perish-
able food; muscle fl esh quickly deteriorates after catching
due to unstable chemical species naturally present in aquatic
animals.
21,23
Chemical processes, including proteolysis and
lipid oxidation, are responsible for muscle deterioration,
rendering the tissue prone to microbial contamination and
loss of freshness.
23
A number of adenosine triphosphate
(ATP) breakdown products, accumulated products from
microbial activity (volatile basic nitrogen (TVBN) compounds,
trimethylamine (TMA) and volatile sulphur compounds such
as H
2
S, CH
3
SH and (CH
3
)
2
S) can function as indicators of
freshness and storage life.
34
Consequently, these compo-
nents must be reliably measured.
Regulatory quality guarantees in the meat
and fi shery industries
The establishment of regulatory organisations covering
different aspects of quality such as food safety and hygiene,
animal ethical treatment and welfare, eating quality, authen-
tication and traceability has taken place to protect consumer
rights and ethical food processors from food adulteration
by those wishing to gain unfair economic advantage.
8,35,36
Setting standards and classification systems for meat
carcasses and cutting enable quality screening, especially
when trading either nationally and especially internation-
ally. These procedures are referred to as carcass classifi ca-
tion and meat grading. Carcass features, such as weight,
shape (conformation), fat cover, estimated yield, lean colour,
fat colour and marbling, are useful in ranking carcasses
into hierarchies of quality which may be different among
geographic regions and different pricing values can arise
Yield and gross composition Quantity of saleable product
Ratio of fat to lean
Muscle size and shape
Appearance and technological characteristics Fat texture and colour
Amount of marbling in lean (intramuscular fat)
Colour and water-holding capacity of lean
Chemical composition of lean
Palatability Texture and tenderness
Juiciness
Flavour
Wholesomeness Nutritional quality
Chemical safety
Microbiological safety
Ethical quality Acceptable husbandry of animals
Provenance
Authenticity
Table 1. Essential considerations of meat quality and indicative elements.
19
Parameter Fat Moisture Protein Tenderness Juiciness
Moisture−0.72***————
Protein −0.24** −0.11 — — —
Tenderness 0.16 −0.06 0.02 — —
Juiciness 0.11 −0.04 −0.06 0.62*** —
Overall appraisal 0.30*** −0.16 −0.01 0.74*** 0.60***
**P < 0.01; ***P < 0.001.
Table 2. Relationship (correlation coeffi cients) between chemical composition and eating quality of meat.
33

4 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
from meat grading.
37
A list of relevant issues with respect
to meat and fi sh quality is given in Table 3. Offi cial regula-
tory bodies overseeing quality control and trade in meat
and fi sh products are responsible for ensuring compliance
at different levels, domestic, interstate and international;
some of these are listed in Table 4.
Issues Examples
Food hygiene and safety Mandatory hygiene and food safety practice aimed at ensuring products
are safe from chemical, biological and physical hazards and contaminants:
HACCP and GMP
Control of veterinary medicine residues and pesticides in foods of animal
origin.
System covering the control of bovine animals in relation to BSE
Animal welfare and ethical issues
Requirement for the stunning of animals before slaughter
Intensive fi sh farming and environmental impact
Eco-labelling on fi shery products
Eating quality
Compulsory carcass classifi cation for standardising pricing systems
Quality label for fresh beef
Classifi cation of fi sh myosystem based on protein and fat content
Authentication
Meat origin: sex, meat cuts, breed, feed intake, slaughter age, wild vs farmed
meat, organic vs conventional meat, geographical origin
Meat substitution: meat species and fat and protein ingredients
Meat processing treatment: irradiation, fresh vs frozen-thawed meat, meat
preparation
Non-meat ingredient addition: additives and water
Fish species identifi cation
Classifi cation of wild and farmed fi sh
Table 3. Regulatory issues for livestock, meat and fi sh quality.
29,33–40
Country/Region Regulation body Offi cial website
EU
European Commission—Agriculture and Rural
Development
http://ec.europa.eu/agriculture/envir/index_en.htm
Australia
Australian Government—Department of
Agriculture, Fisheries and Forestry: Meat, Wool
and Dairy
http://www.daff.gov.au/agriculture-food/meat-wool-
dairy/red-meat-livestock
New Zealand
New Zealand Food Safety Authority—comply to the
Animal Products Act 1999
http://www.nzfsa.govt.nz/animalproducts/index.htm
USA
USDA: United States Department of Agriculture
FDA: U.S. Food and Drug Administration—Federal
Meat Inspection Act
http://www.fsis.usda.gov/Home/index.asp
http://www.fda.gov/RegulatoryInformation/
Legislation/ucm148693.htm
Canada
Canadian Food Inspection Agency—Food and
Animals
http://www.inspection.gc.ca/english/toce.shtml
China
Ministry of Agriculture, PRC
State Food and Drug Administration, PRC
http://english.agri.gov.cn/ http://eng.sfda.gov.cn/
eng/
India
Ministry of Food Processing Industries—Poultry
industry, Meat and meat products and Marine
products sectors
http://mofpi.nic.in/default.aspx
South Korea
Ministry for Food, Agriculture, Forestry and
Fisheries—Livestock Bureau
http://english.mifaff.go.kr/USR/WPGE0201/m_413/
DTL.jsp
Japan
Ministry of Agriculture, Forestry and Fisheries
Agriculture and Livestock Industries Corporation
http://www.maff.go.jp/e/index.html
http://www.alic.go.jp/english/what.html
International International Standards Organization (ISO)
Codex Alimentarius Commission (Codex)
International HACCP Alliance
Offi ce International des Épizooties (OIE)
World Trade Organisation (WTO)
http://www.iso.org/iso/home.html
http://www.codexalimentarius.net/web/index_en.jsp
http://www.haccpalliance.org/sub/index.html
http://www.oie.int/eng/en_index.htm
http://www.wto.org
Table 4. Offi cial regulatory bodies establishing regulations for meat and seafood industries

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 5
Near infrared spectroscopy
On the basis of the chemical constituents and physical prop-
erties of muscle foods, NIR technology coupled with multi-
variate data analysis, enables classifi cation of food variety
and traceability in production history. They can also quantify
quality factors, such as chemical and physical properties and
palatability, in a wide variety of muscle food products. Different
quality attributes and characteristics of meat and seafood
products can be explained on the basis of chemical compo-
sition and chemical changes which may arise from farming
practices, intermediate handling or processing treatments or
a combination of these.
4,16
In NIR spectroscopic applications
in muscle foods, visible wavelength ranges are often included
in addition to the NIR range (780–2500 nm) due usually to the
presence of pigments in the raw material. Thus, the spectral
range of applications to be discussed in this review will include
the visible and NIR regions.
Advantages of near infrared spectroscopy in
muscle food quality control
Despite the complexity of muscles in animal tissues, product
quality must be guaranteed in order to satisfy consumer expec-
tations if a food business is to be developed and sustained.
However, routine laboratory analyses to measure quality are
destructive, time-, labour- and chemical-consuming and
sometimes involve the use of toxic materials. Product sensory
assessment by trained human panels is difficult to imple-
ment effi ciently, requiring a dedicated testing location and the
training and availability of skilful testers which cost money
and time.
21
In this section, the possibility of using NIR spec-
troscopy to solve these diffi culties is illustrated.
The operating speed of NIR spectroscopy is a major advan-
tage of the technique which does not normally require any
sample preparation. Preliminary studies and actual imple-
mentations of NIR methods have been reported for the deter-
mination of product variety and various quality determinants of
meat and fi sh for at-line, on-line, in-line and off-line applica-
tions, thereby facilitating control (including regulatory compli-
ance
13
) and monitoring at the production line, in commercial or
trading environments. By quantitative prediction of the chem-
ical constituents in meat and fi sh muscles, NIR allows meat
processors to adjust and optimise their production control,
the quality of meat raw materials in processed meat and the
management of low-grade meat, while the information on
fi nal meat composition is useful to enable meat producers
to adjust and optimise feeding regimens and dietary supple-
mentation.
13,41
Traceability and authentication issues such as
characterisation of muscle foods grown intensively or exten-
sively can be confi rmed allowing for different pricing levels.
42
All in all, the application of NIR allows protective action which
is benefi cial to all parties in terms of food safety and nutrient
quantifi cation, manipulation of different meat grades, recipe
optimisation and adjustment of animal nutrition.
43
Increased consumer awareness of and demand for safe
and high-quality food has driven technological developments
throughout the food chain towards tighter controls which inev-
itably impose additional costs. Trading diffi culties may arise
between developing and developed countries as a result of the
increasingly sophisticated quality standards required by the
latter and the inability of the former to implement the required
testing procedures in a time- and cost-effective manner. This
may be a particular problem for fi shery exporters in developing
countries.
44
To harmonise this global trade issue, two agree-
ments have been established by the World Trade Organisation
(WTO); these are the Agreement on the Application of Sanitary
and Phytosanitary (SPS) Measures and the Agreement on
Technical Barriers to Trade (TBT).
45,46
Near-infrared tech-
nology may have a signifi cant role to play in this regard as
an effective and appropriate tool for harmonising technical
trading issues due to its simplicity, economic feasibility and
reliability.
Potential sources of variation in near infra-
red analysis of muscle foods
The major sources of variation in NIR spectra of muscle foods
arise from their solid and highly heterogeneous nature. As a
result of their fi brous and layered construction, the collection
of high-quality spectra is more diffi cult than for liquid biolog-
ical samples.
47–49
Scattering effects arise chiefl y from textural
inhomogeneity as a result of variation in protein fi bre arrange-
ment and the presence of intramuscular fat and connective
tissue. Samples can be presented to NIR sensors in several
forms, i.e. as on-carcass muscles, intact fresh meat cuts,
ground and/or homogenised material, freeze-dried samples or
liquid extracts. NIR spectra can be collected in several modes
(transmittance, refl ectance or transfl ectance) and by different
optical arrangements (angular optics, integrating spheres
and interactance fi bre optic probes). Inconsistent conditions of
ambient light and relative humidity may also affect the quality
of NIR spectra collected at different times.
50,51
Near infrared spectral data pre-processing
In NIR spectroscopy, according to the Beer–Lambert (or Beer’s)
law, spectral linearity with respect to chemical concentrations
and qualitative characteristics of analytes of interest should be
achieved to obtain meaningful data for multivariate analysis.
52
The physical structure of samples, such as particle size and
porosity, may result in non-linearity and low signal-to-noise
ratios
53
in spectral datasets collected from muscle foods.
Spectral pre-treatments to correct non-linearity are well-
established; these include multiplicative signal correction
(MSC), standard normal variate (SNV), normalisation, baseline
correction, spectral derivatives and many others including the
representative layer theory or Dahm equation.
47,48,52
A sample
is modelled as a series of plane parallel layers made up from
particulate units and each represents the sample as a whole,
hence it is independent of sample thickness. Through this
approach, analyte concentrations in heterogeneous systems
can be evaluated as well as providing a way to understand
optical phenomena in diffusing materials and to explain spec-
tral observations qualitatively.
47,48

6 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Measurement of chemical
composition by near infrared
spectroscopy
Accurate analysis of meat composition is important because
of its relationship to both overall quality and specifi c charac-
teristics such as eating quality, wholesomeness and impact on
consumer health. Many feasibility studies on the application of
NIR spectroscopy to the measurement of chemical composi-
tion in meat, seafood and their derived products have been
reported over the past fi ve years [Tables 5(a)–(f)]; a selection of
these studies is discussed below for individual meat compo-
nents. Unless otherwise stated, all quantitative measure-
ments are quoted on a fresh matter (FM) basis.
Protein
Protein is an important functional and nutritional compo-
nent of meat and meat products and several authors have
reported the development of predictive NIR models for this
constituent [Tables 5(a)–(e)]. It is often diffi cult to compare
the reported errors for protein measurement given that they
are expressed as a percentage of fresh or dry matter and
have been determined using either cross-validation or a
separate validation sample set. In some cases, only calibra-
tion errors are reported. For example, in the case of beef
[Table 5(a)], reported error values were 0.48% FM (RMSECV),
2.03% dry matter (DM; RMSECV), 1.02% FM (SEP) and 1.24%
(RMSEC).
In the case of pork and pork products, a limited number of
publications have reported protein measurements [Table 5(b)].
For intact pork sausage, a SEP value of 1.08% in combination
with a coeffi cient of determination of cross-validation (r
2
cv
)
equal to 0.95 was reported for Iberian pork.
63
When minced or
homogenised pork sausage was analysed in the wavelength
range 515–1650 nm, prediction errors of 0.83% (r
2
cv
= 0.93)
for minced material
62
and 0.87% (r
2
cv
= 0.93)
62
and 0.95%
(r
2
cv
= 0.96)
63
for homogenised samples were reported.
Only a single publication reported the prediction of crude
protein in lamb (19-month-old Merino sheep). Using freeze-
dried Longissimus dorsi muscle and the wavelength range
400–1900 nm, a standard error of prediction equal to 0.92%
(r
2
cv
= 1.0) was reported.
72
Crude protein prediction errors for
ground poultry carcasses of 1.83% (r
2
= 0.96) and 2.01% DM
(r
2
cv
= 0.85) for broilers have been published.
73,74
A RMSECV of
0.74% (r
2
cv
= 0.91) was quoted for ground, freeze-dried chicken
breast muscle while the equivalent fi gures for freeze-dried,
minced ostrich meat and homogenised guinea fowl muscle
were 0.64% (r
2
= 0.94)
46
and 1.96% (r
2
cv
= 0.76),
80
respectively.
NIR measurement of crude protein content has been
reported in surimi, a traditional Japanese fish-derived
product [Table 5(f)].
77
Measurements were performed on
intact surimi or fish gel in transmittance mode over the
900–1100 nm region producing a coeffi cient of determination
of cross-validation (r
2
cv
) equal to 0.96 and a prediction error
(RMSECV) of 0.13%. On the basis of a range in constituent
values of 11.98–16.17 g 100 g
−1
, the associated RPD value
(ratio of prediction error to constituent range) was high (10.4),
indicating a calibration that should be suitable for commer-
cial deployment.
Fat
The amount and distribution of intramuscular fat (IMF) may
play an important role in the eating quality (Table 2) of muscle
foods as it may affect characteristics such as fl avour, juiciness,
texture and appearance.
81
Prediction of IMF content in different meat products has
varied from poor to good. In most cases, published r
2
values
ranged from 0.80 to 0.98, but one study of pork meat by Hoving-
Bolink et al.
60
involving three different muscles [Table 5(b)]
produced a coeffi cient of determination in cross-validation
(r
2
cv
) of 0.35 with an associated RMSECV equal to 3.6 g kg
−1
.
This study was performed at a commercial slaughterhouse
and had the aim of developing a model for on-line applica-
tion to meat one day after slaughter. It involved a multi-linear
regression (MLR) technique for calibration development using
absorbance values at 1210 nm as an initial input variable due
to the signifi cant absorbance by C–H bands (CH
2
vibrations)
at this wavelength. Poor predictive performance in this appli-
cation may have arisen from the use of a small sampling
area
(1 mm
2
) when collecting spectra with a diffuse refl ection
probe (Zeiss MCS 511/512 instrument); in the same study, a
higher coeffi cient of determination (0.70) was obtained using a
benchtop NIR instrument which also scanned a larger surface
area (50 mm
2
). This observation illustrates the critical impor-
tance of spectral collection strategies in maximising analytical
prediction accuracy.
In another pork study,
r
2
cv
values between 0.40 and 0.58
with associated prediction errors (SEPs) of 0.37–0.40% were
reported for a single (Longissimus dorsi) muscle.
61
Pork
samples were placed in a sample holder of 5 cm × 6 cm
(30 cm
2
sample area) and analysed using a Foss NIRSystems
6500 spectrophotometer. The rather poor result from this
study demonstrates a major problem in trying to predict
IMF content. Despite efforts to deliberately increase the
range in IMF content in the sample set by using boars, gilts
and barrows from different crossbreeds, the absolute range
(0.1–4.3%) was still rather narrow. Even using animals of
different ages, which would not refl ect commercial prac-
tice, would not generate a large range in IMF as D’Souza et
al.
82
demonstrated no signifi cant difference (P > 0.05) in IMF
levels in the Longissimus dorsi muscle between animals of
different ages.
One fish study involving the measurement of fat and
astaxanthin in farmed Atlantic salmon [Table 5(f)] demon-
strated a non-invasive analysis,
78
which maximised hygiene
maintenance in food production due to the avoidance of direct
contact between the scanning device and food sample. Three
different sample presentations were compared, whole live
fi sh, gutted whole fi sh and fi llets. A promising result (r
2
= 0.88;
RMSEP = 1.02%) for fat analysis was reported for live fish,

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 7
Study
purposes
Sample preparation
Quality parameters (n)
(measurement ranges)
Measurement mode;
regression method
Wavelength
range (nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To predict chemi-
cal composition of
raw beef
Beef:
Friesian;
Hereford
(6 grades of Chilean
system)
Longissimus tho-
racis
et lumborum;
Supraspinosus;
Semitendinosus
Minced
Dry matter
(72)
;(21.5–26.8% FM) R; PLS 400–2500 1
st
Deriv +SNV-D RMSECV = 0.58% FM; r
2
cv
= 0.77
54
Crude protein
(72)
(18.3–22.6% FM) R; PLS 400–2500 2
nd
Deriv RMSECV = 0.48% FM; r
2
cv
= 0.82
Ether extract
(72)
; (0.47–6.10% FM) R; PLS 400–2500 2
nd
Deriv RMSECV = 0.4 4% FM; r
2
cv
= 0.82
Ash
(72)
; (0.93–1.2% FM) R; PLS 400–2500 Smooth +SNV-D RMSECV = 0.03% FM; r
2
cv
= 0.66
Collagen
(72)
; (0.31–1.9% FM) R; PLS 400–2500 2
nd
Deriv RMSECV = 0.30% FM; r
2
cv
= 0.18
To predict intra-
muscular fat
of beef by NIRS
calibrated against
Folch extraction
method
Beef:
Bulls;
Steers
Longissimus dorsi Minced
IMF
(33)
; (0.88–8.5% FM) R; PLS 400–2500 SNV-D + 1
st
Deriv RMSECV = 0.39% FM; r
2
cv
= 0.94; RPD = 4.1 55
IMF
(34)
; (0.88–8.5% FM) R; PLS 1100–2500 SNV-D + 1
st
Deriv RMSECV = 0.44% FM; r
2
cv
= 0.93; RPD = 3.8
To predict chemi-
cal parameters
in oxen meat
samples guaran-
teed under quality
marks
Beef:
Mountain oxen
Longissimus
thoracis
Homogenised
Crude protein
(53)
; (588.7–851.0 g kg
–1
DM)
R; PL S 1100–250 0 MSC + 2
nd
Deriv
RMSECV = 20.3 3 g kg
–1
DM; R
2
= 0.874; RPD = 2.56
36
Myoglobin
(46)
;(17.7–37.0 g kg
–1
DM) R; PLS 1100–2500 MSC + 2
nd
Deriv
RMSECV = 3.45 g kg
–1
DM; R
2
= 0.4 40; RPD = 1.09
Collagen
(47)
; (5.7–21.3 g kg
–1
DM) R; PLS 1100–2500 2
nd
Deriv
RMSECV = 3.82 g kg
–1
DM; R
2
= 0.472; RPD = 1.26
Ether extract
(53)
;(92.2–359.8 g kg
–1
DM) R; PLS 1100–2500 MSC +2
nd
Deriv
RMSECV = 16.22 g kg
–1
DM; R
2
= 0.924; RPD = 3.32
Gross energy
(51)
; (24.0–28.7 MJ kg
–1
DM)
R; PL S 1100–250 0 MSC + 2
nd
Deriv
RMSECV = 0 . 2 9 M J k g
–1
DM; R
2
= 0.941; RPD = 3.31
Dry matter
(46)
;(271.0–339.1 g kg
–1
FM) R; PLS 1100–2500 None
RMSECV = 6.75 g kg
–1
FM; R
2
= 0.874; RPD = 2.28
Ash
(53)
;(31.7–57.7 g kg
–1
DM) R: PLS 1100–2500 None
RMSECV = 5.15 g kg
–1
FM; R
2
= 0.168; RPD = 1.04
To predict chemi-
cal characteris-
tics, instrumental
texture and sen-
sory properties
of beef
Beef:
SEUROP classifi cation
involving 12 different
breeds and cross-
breeds
Longissimus
thoracis
Homogenised
IMF
(190a)
; (0.33–2.20%FM) R; PLS 1108–2492.8 SNV-D + 1
st
Deriv
SEP = 0.490%FM; r
2
= 0.759; RPD = 1.00
33
Moisture
(190a)
;(73.3–77.9%FM) R; PLS 1500–2460.8 1
st
Deriv
SEP = 0.369%FM; r
2
= 0.717; RPD = 1.87
Protein
(190a)
;(17.1–23.8%FM) R; PCR 1108–2492.8 MSC + 1
st
Deriv
SEP = 1.019%FM; r
2
= 0.158; RPD = 1.09
Myoglobin
(190a)
;(2.6–5.1 mg g
–1
)R; PLS408–2492.81
st
Deriv
SEP = 0.260 mg g
–1
; r
2
= 0.914; RPD = 2.38
To predict fatty
acid content
Beef:
Asturiana de los Valles;
Asturiana de la
Montaña
Longissimus
thoracis
Ground meat
SFA
(93)
;(0.19–2.31 g 100g
–1
)T; PLS850–10502
nd
Deriv + SNV-D RMSECV = 0.182 g 10 0g
–1
; r
2
cv
= 0.8 37
56
BFA
(95)
;(3.4–26.6 mg 100g
–1
)T; PLS850–10502
nd
Deriv + SNV-D RMSECV = 2.907 mg 100g
–1
; r
2
cv
= 0.701
MUFA
(92)
;(0.1–1.8 g 100g
–1
)T; PLS850–10502
nd
Deriv + SNV-D RMSECV = 0.140 g 100g
–1
; r
2
cv
= 0.852
PUFA
(93)
;(0.2–0.4 g 100g
–1
) T; PLS 850–1050 None RMSECV = 0.033 g 100g
–1
; r
2
cv
= 0.244
CLA
(96)
;(0.5–10.97 mg 100g
–1
) T; PLS 850–1050 None RMSECV = 1.613 mg 100g
–1
; r
2
cv
= 0.586
Table 5(a). Applications of near infrared spectroscopy in analysis of muscle food composition—beef and beef products.

8 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Table 5(a) (cocntinued). Applications of near infrared spectroscopy in analysis of muscle food composition—beef and beef products.
Study
purposes
Sample preparation
Quality parameters (n)
(measurement ranges)
Measurement mode;
regression method
Wavelength
range (nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To predict
cholesterol,
fat, calories
and moisture in
raw and cooked
ground beef
patties
Beef Mixtures of peeled
knuckles and lean
Fresh beef
patties
Cholesterol
(96)
;(143.9–262.8 mg g
–1
) R; PLS 350–1075 Deriv spectra SEP = 13.3–15.0 mg g
–1
; r
2
= 0.73– 0.8 0; RPD = 2.0–2.2
57
Fat
(96)
;(14.6–83.7%DM) R; PLS 350–1075 Deriv spectra SEP = 4.1–4.5%DM; r
2
= 0.94–0.96; RPD = 4.1–4.8
Calorie
(96)
;(5.9–8.8 Kcal g
–1
) R; PLS 350–1075 Deriv spectra SEP = 0.16–0.19 Kcal g
–1
; r
2
= 0.94– 0.96; RPD = 4.2–4.9
Moisture
(96)
;(32.5–73.3%DM) R; PLS 350–1075 Deriv spectra SEP = 2.5%DM; r
2
= 0.94– 0.95; RPD = 4.2–4.3
Cooked beef
patties
Cholesterol
(96)
;(173.1–276.4 mg g
–1
) R; PLS 350–1075 Deriv spectra SEP = 5.8–6.2 mg g
–1
; r
2
= 0.76–0.79; RPD = 2.1–2.2
Fat
(96)
;(15.0–51.0%DM) R; PLS 350–1075 Deriv spectra SEP = 3.2–3.5%DM; r
2
= 0.85– 0.87; RPD = 2.6–2.9
Calorie
(96)
;(5.9–7.4 Kcal g
–1
) R; PLS 350–1075 Deriv spectra SEP = 0.06 –0.07 Kcal g
–1
; r
2
= 0.86– 0.87; RPD = 2.7–
3.0
Moisture
(96)
;(27.0–55.9%DM) R; PLS 350–1075 Deriv spectra SEP = 1.6 –1.7%DM ; r
2
= 0.70–0.72; RPD = 1.9–2.0
To measure meat
quality indicator
at pre- and post-
rigor stages
Beef:
Hereford
Longissimus
lumborum
Intact Glycogen
(67)
;(0.0–18.7 mg g
–1
)R; PLS400–1700SNV+GLSRMSEP = 2.7 mg g
–1
; r
2
= 0.72; SEL = 0.4 58
To develop an NIR
prediction model
from samples
collected at meat
packer after 48
hours ageing**
Beef
Tenderloin;
Ribeye; Topside;
Shin; Striploin
NA
Moisture
(114a)
;(NA) R; PLS 950–1650 MSC, SNV + 1
st
Deriv
RMSEC = 0.31; r
2
= 0.90
b
59
IMF
(114a)
;(NA) R; PLS 950–1650 MSC, SNV + 1
st
Deriv
RMSEC = 0.22; r
2
= 0.85
b
Protein
(114a)
;(NA) R; PLS 950–1650 MSC, SNV + 1
st
Deriv
RMSEC = 1.24; r
2
s
= 0.87
b
(n
) Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language.
N/A: not available

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 9
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength range
(nm)
Spectral
pre–treatment
Calibration performance Ref.
Animal
breeds/variety
Selected
muscles
Sample
presentation
To develop an in–line
prediction model of IMF
in relation to techno-
logical quality of pork
Pork: Yorkshire and
Piétrain crossbreds
Longissimus thoracis;
Longissimus lumborum;
Semimembranosus
Intact carcass IMF
(102)
;(5.1–25.7 g kg
–1
)R; MLR 1210 2
nd
Deriv
RMSECV = 3.6 g kg
–1
; r
2
cv
= 0.35
60
To predict IMF and other
technological proper-
ties of pork
Pork: Crossbreed Dutch
landr ace + Finnish lan-
drace and Yorkshire
Longissimus Intact IMF
(115–169)
;(0.1–4.3%) R; PLS 800–2500 2
nd
Deriv SEP = 0.37–0.40%; r
2
cv
= 0.4 0–
0.58
b
61
To optimise prediction
models for composition
with different sample
presentations for qual-
ity control of typical sal-
chichón pork sausage
Pork sausages: Iberian;
Standard
NA Minced (M);
Homogenised (H)
IMF
(80)
;(8–31.7%) R; PCR and PLS 515–1650 Deriv, MSC + SNV–D SEP (M) = 1.38%,
SEP(H) = 0.94%; r
2
cv
(M) = 0.98,
r
2
cv
(H) = 0.99
62
Moisture
(80)
;(50.2–68.4%) R; PLS 515–1650 Deriv, MSC + SNV–D SEP (M) = 1%, SEP(H) = 0.76%;
r
2
cv
(M) = 0.98, r
2
cv
(H) = 0.98
Protein
(80)
;(12.7–20.5%) R; PLS 515–1650 Deriv, MSC + SNV–D SEP (M) = 0.83%,
SEP(H) = 0.87%; r
2
cv
(M) = 0.93,
r
2
cv
(H) = 0.93
To optimise NIR predic-
tion models for compo-
sition analysis of pork
sausages for possible
implementation in vari-
ous control check points
during manufacturing
Pork: Iberian; Standard
NA
Intact (I);
Homogenised (H)
IMF
(80)
;(11.7–43.2%) R; PCR and PLS 515–1650 Deriv, MSC + SNV–D
SEP (I) = 1.47%, SEP(H) = 0.71%;
r
2
cv
(I) = 0.98, r
2
cv
(H) = 1.0
63
Moisture
(80)
;(29.5–45.2%) R; PLS 515–1650 Deriv, MSC + SNV–D
SEP (I) = 0.97%, SEP(H) = 0.41%;
r
2
cv
(I) = 0.92, r
2
cv
(H) = 0.99
Protein
(80)
;(20.1–36.1%) R; PLS 515–1650 Deriv, MSC + SNV–D SEP (I) = 1.0 8%, SEP(H) = 0.95%;
r
2
cv
(I) = 0.95, r
2
cv
(H) = 0.96
To measure fatty acid
composition of pig fat
raw material for proc-
essed meat control
Pork Outer layer of backfat Homogenised
Total SFA
(155)
;(34.5–45.9%) NA; PLS 1041.67–2380.95 MSC + 1
st
Deriv SEP = 0.9%; r
2
cv
= 0.98
64
Total MUFA
(155)
;(40.5–53.6%)
NA; PLS 1041.67–2380.95 Normalisation + 1
st
Deriv
SEP = 1.6%; r
2
cv
= 0.88
Total PUFA
(155)
;(7.0–20.9%)
NA; PLS 1041.67–2380.95 Normalisation + 1
st
Deriv
SEP = 4.7%; r
2
cv
= 0.96
Total SFC
(80)
;(14.7–80.3%)
at
10
°
C; 20
°
C
NA; PLS 1041.67–2380.95 Normalisation + 1
st
Deriv
SEP = 2.9% and 3.2%;
r
2
cv
= 0.94
at 10°C
and 0.96
at 20°C
Table 5(b). Applications of near infrared spectroscopy in analysis of muscle food composition—pork and pork products.

10 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Study purpose Sample description Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength range
(nm)
Spectral
pre–treatment
Calibration performance
Ref.
Animal
breeds/variety
Selected
muscles
Sample
presentation
To determine four main
fatty acids in live Iberian
pigs and on carcasses
in slaughterhouse for
on–site control and
traceability of specifi c
feeding system of
Iberian pork
Pork: Iberian; Duroc
Jersey crossbreeds
NA
In vivo
Palmitic acid—C16:0
(52)
;(17.8–
25.5%)
R; PLS 450–2300 SNV–D + 1
st
Deriv RMSECV = 1.24%; r
2
cv
= 0.74;
RPD = 2.0
65
Stearic acid—C18:0
(52)
;(6.9–12.5%)
R; PLS
450–2 300 SNV–D + 1
st
Deriv, RMSECV = 0.67%; r
2
cv
= 0.72;
RPD = 1.9
Oleic acid—C18:1
(52)
;(46.7–
59.1%)
R; PLS
450–2 300 SNV–D + 1
st
Deriv RMSECV = 1.42%; r
2
cv
= 0.77;
RPD = 2.1
Linoleic acid—C18:2
(52)
;(6.5–10.2%)
R; PLS
450–2 300 SNV–D + 2
nd
Deriv RMSECV = 0.36%; r
2
cv
= 0.60;
RPD = 1.6
On carcass
Palmitic acid—C16:0
(52)
;(17.8–25.5%)
R; PLS
450–2000 SNV–D and 1
st
Deriv RMSECV = 0.82%; r
2
cv
= 0.87;
RPD = 2.8
Stearic acid—C18:0
(52)
;(6.9–12.5%)
R; PLS
450–2000 SNV–D and 1
st
Deriv, RMSECV = 0.94%; r
2
cv
= 0.4 6;
RPD = 1.4
Oleic acid—C18:1
(52)
;(46.7–
59.1%)
R; PLS
450–2000 SNV–D and 1
st
Deriv RMSECV = 1.48%; r
2
cv
= 0.80;
RPD = 2.3
Linoleic acid—C18:2
(52)
;(6.5–10.2%)
R; PLS
1100–2300 SNV–D and 2
nd
Deriv RMSECV = 0.55%; r
2
cv
= 0.31;
RPD = 1.2
To study potential
at–line prediction of
hydroxyproline in sau-
sages
Pork: Iberian; Non
Iberian
NA Homogenised Hydroxyproline
(70)
;(0.13–0.74 g
100 g
–1
)
R; PLS 1100–2000 MSC, SNV–D, and 1
st
Deriv
SEP = 0.05 g 100g
–1
; R
2
= 0.6 4
b
;
RPD = 2.2
66
To evaluate NIR spec-
troscopy of fermented
sausages using on–
contact and remote
probes
Pork Shoulder and bellies Homogenised Moisture
(101)
;(16.8–66.1%) R; PLS
On–contact probe
1332.25–2327.26 1
st
Deriv and MSC RMSEP = 0.68%; r
2
cv
= 0.99;
RPD = 19.8
67
R; PLS
Remote probe
833.63–2175.00
1
st
Deriv and MSC
RMSEP = 0.62%; r
2
cv
= 0.99;
RPD = 21.6
Aw
(101)
;(0.8–1.08) R; PLS
On–contact probe
1332.25–2327.26 1
st
Deriv and VN RMSEP = 0.01; r
2
cv
= 0.99;
RPD = 9.22
R; PLS
Remote probe
1332.96–2175.00 1
st
Deriv and MSC RMSEP = 0.01; r
2
cv
= 0.99;
RPD = 8.26
Sodium chloride (NaCl)
(101)
;(1.1–3.5%)
R; PLS
On–contact probe
1332.25–2175.95 1
st
Deriv and VN RMSEP = 0.12%; r
2
cv
= 0.94;
RPD = 6.19
R; PLS
Remote probe
833.63–2175.00 1
st
Deriv and MSC RMSEP = 0.12%; r
2
cv
= 0.97;
RPD = 6.23
Table 5(b) (continued). Applications of near infrared spectroscopy in analysis of muscle food composition—pork and pork products.

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 11
Study purpose Sample description Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength range
(nm)
Spectral
pre–treatment
Calibration performance
Ref.
Animal
breeds/variety
Selected
muscles
Sample
presentation
To evaluate fatty acids
of pork raw material in
Iberian dry–cured
sausages
Iberian dry–cured
sausages of two
recipes: salchichón,
chorizo
NA Minced sausages
C12:0
(86)
;(0.06–0.10%) R; PLS 400–2498 SNV–D + 2
nd
Deriv RMSECV = 0.01%; r
2
= 0.03;
RPD = 1.00
68
Myristic—C14:0
(86)
;(1.22–
1.78%)
R; PLS 400–2498 SNV–D + 1
st
Deriv RMSECV = 0.07%; r
2
= 0.63;
RPD = 1.63
Palmitic—C16:0
(86)
;(22.83–
28.00%)
R; PLS
400–2498
SN V–D + 1
st
Derive RMSECV = 0.58%; r
2
= 0.8 4;
RPD = 2.4 8
Palmitoleic—C16:1
(86)
;(2.3–
3.7%)
R; PLS
400–2498
SN V–D + 2
nd
Deriv RMSECV = 0.26%; r
2
= 0.41;
RPD = 1.30
C17:0
(86)
;(0.13–0.35%)
R; PLS
400–2498
SN V–D + 2
nd
Deriv RMSECV = 0.04%; r
2
= 0.04;
RPD = 1.00
C17:1
(86)
;(0.15–0.33%)
R; PLS
400–2498
SN V–D + 1
st
Deriv RMSECV = 0.0 4%; r
2
= 0.03;
RPD = 1.03
Stearic—C18:0
(86)
;(10.6–
14.8%)
R; PLS
400–2498
SN V–D + 2
nd
Deriv RMSECV = 0.55%; r
2
= 0.78;
RPD = 2.07
Oleic—C18:1
(86)
;(43.0–52.6%) R; PLS
400–2498
SN V–D + 2
nd
Deriv RMSECV = 1.51%; r
2
= 0.58;
RPD = 1.54
Linoleic—C18:2
(86)
;(4.5–10.3%)
R; PLS 400–2498 SNV–D + 2
nd
Deriv RMSECV = 0.86%; r
2
= 0.56;
RPD = 1.50
α –linoleic—C18:3
(86)
;(0.4–
1.1%)
R; PLS
400–2498
SN V–D + 2
nd
Deriv RMSECV = 0.16%; r
2
= 0.56;
RPD = 1.50
C20:0
(86)
;(0.2–0.3%) R; PLS 4 00–2498 SNV–D + 2
nd
Deriv RMSECV = 0.02 %; r
2
= 0.02;
RPD = 1.00
Icosaenoic—C20:1
(86)
;(0.39–
1.09%)
R; PLS 400–2498 SNV–D + 1
st
Deriv RMSECV = 0.17%; r
2
= 0.07;
RPD = 1.03
SFA
(86)
;(35.7–44.8%) R; PLS
400–2498
SN V–D + 2
nd
Deriv RMSECV = 0.98%; r
2
= 0.86;
RPD = 2.63
MUFA
(86)
;(46.9–56.8%) R; PL S 40 0–2498 SN V–D + 2
nd
Deriv RMSECV = 1.47%; r
2
= 0.53;
RPD = 1.4 5
PUFA
(86)
;(4.9–11.2%) R; PL S 400 –2498 SN V–D + 2
nd
Deriv RMSECV = 0.88%; r
2
= 0.61;
RPD = 1.58
To evaluate the use
of TOP algorithm to
improve robustness of
NIRS equations for the
prediction of fatty acids
in Iberian pig fats
Pork: lberian
Subcutaneous fat Melted fat
Palmitic acid—C16:0
(188);(17.20–60 %)
R; PLS 1100–2500 SNV + 2nd Deriv SEP = 1.95%
69
R; PLS + TOP 1100–2500 SNV + 2nd Deriv SEP = 0.34%
Stearic acid—C18:0 (188)
;(7.20–14.50 %)
R; PLS 1100–250 0 SN V + 2nd Der iv SEP = 0.47 %
R; PLS + TOP 1100–2500 SNV + 2nd Deriv SEP = 0.31 %
Oleic acid—C18:1 (188)
;(44.50–60.30 %)
R; PLS 1100–250 0 SN V + 1st Deriv SEP = 0.53 %
R; PLS + TOP 1100–2500 SNV + 1st Deriv SEP = 0.39%
Linoleic acid—C18:2 (188)
;(6.00–14.00 %)
R; PLS 1100–250 0 SN V + 1st Deriv SEP = 4.45 %
R; PLS + TOP 1100–2500 SNV + 1st Deriv SEP = 0.58 %
Table 5(b) (continued). Applications of near infrared spectroscopy in analysis of muscle food composition—pork and pork products.

12 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Study purpose Sample description Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength range
(nm)
Spectral
pre–treatment
Calibration performance
Ref.
Anima
breeds/variety
Selected
muscles
Sample
presentation
Use of a fi bre optics
probe on subcutaneous
fat to predict fatty acid
composition of pigs fed
with different diets
Pork: Piétr ain × (Large
White × Landrace)
cross breed
Subcutaneous fat Intact Myristic—C14:0
(115)
;(0.96–2.38
%)
R; PLS
1124–2 381 EMS C, 1st + 2nd
Deriv
RMSEP = 0.2 %
c,d
; r
2
= 0.75
c
;
0.72
d
70
Palmitic—C16:0
(115)
;(16.13–
23.20%)
R; PLS
1124–2 381 EMS C, 1st + 2nd
Deriv
RMSEP = 1.4 %
c, d
; r
2
= 0.37
c
;
0.41
d
Stearic—C18:0
(115)
;(6.93–
15.60%)
R; PLS
1124–2 381 EMS C, 1st + 2nd
Deriv
RMSEP = 1.0%
c
; 0.9%
d
;
r
2
= 0.80
c
; 0.84
d
Oleic—C18:1 n-9
(115)
;(28.51–
41.63 %)
R; PLS
1124–2 381 EMS C, 1st + 2nd
Deriv
RMSEP = 1.1%
c
; 1.2%
d
;
r
2
= 0.93
c
; 0.92
d
Asclepic—C18:1 n-7
(115)
;(1.69–3.23 %)
R; PLS 1124–2381 EMSC, 1st + 2nd
Deriv
RMSEP = 0.3%
c
; 0.2%
d
;
r
2
= 0.41
c
; 0.62
d
Linoleic—C18:2 n-6
(115)
;(16.15–26.07 %)
R; PLS 1124–2381 EMSC, 1st + 2nd
Deriv
RMSEP = 1.2%
c
; 1.0%
d
;
r
2
= 0.71
c
; 0.80
d
Linolenic—C18:3 n-3
(115)
;(1.14–1.92 %)
R; PLS 1124–2381 EMSC, 1st + 2nd
Deriv
RMSEP = 0.1%
c
; 0.1%
d
;
r
2
= 0.51
c
; 0.54
d
CLA9_11 (115) ;(0.0–1.99 %) R; PLS 1124–2381 EMSC, 1st + 2nd
Deriv
RMSEP = 0.2%
c
; 0.3%
d
;
R
p
= 0.90
c
; 0.86
d
CLA10_12 (115) ;(0.0–1.24 %) R; PLS 1124–2381 EMSC, 1st + 2nd
Deriv
RMSEP = 0.1%
c
; 0.2%
d
;
r
2
= 0.87
c
; 0.87
d
SFA (115) ;(24.77–41.04 %) R; PLS 1124–2381 EMSC, 1st + 2nd
Deriv
RMSEP = 1.7%
c
; 1.9%
d
;
r
2
= 0.81
c
; 0.77
d
MUFA (115) ;(32.99–48.91 %) R; PLS 1124–2381 EMSC, 1st + 2nd
Deriv
RMSEP = 1.2%
c,d
; r
2
= 0.94
c
;
0.93
d
PUFA (115)
(18.57–32.16 %)
R; PLS 1124–2381 EMSC, 1st + 2nd
Deriv
RMSEP = 1.6%
c
; 1.4%
d
;
r
2
= 0.73
c
; 0.79
d
(
n
)
Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
c
Longitudinal cuts;
d
Transversal cuts;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was
written in another language.
Table 5(b) (continued). Applications of near infrared spectroscopy in analysis of muscle food composition—pork and pork products.

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 13
Study
purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength ranges
(nm)
Spectral
pre-treatment
Calibration
performance
Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To investigate
spectral regions
to explain sensory
evaluation of lamb
Lamb: Texel;
Scottish blackface
Longissimus thoracis Intact
IMF
(231)
;(0.3–4.6%FM) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv
RMSECV = 0.409;
r
2
cv
= 0.794
71
Water
(231)
;(72.0–78.6%FM) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv
RMSECV = 0.687;
r
2
cv
= 0.59 2
To develop NIRS
calibrations for
the proximate and
mineral composition
using freeze-dried
samples
Lamb: 19-month
old Merino sheep
Longissimus dorsi Freeze-dried ground
Ash
(128)
;(2.09–5.17%) R; PLS 1100–2500 2
nd
Deriv SEP = 0.15%; r
2
cv
= 0.94
b
72
Dry matter
(131)
;(90.55–95.92%) R; PLS 1100–2500 2
nd
Deriv SEP = 0.38%; r
2
cv
= 0.92
b
Crude protein
(118)
;(52.94–86.95%) R; PLS 1100–2500 2
nd
Deriv SEP = 0.92%; r
2
cv
= 1.00
b
Fat
(120)
(7.30–51.80%) R; PLS 1100–2500 2
nd
Deriv SEP = 0.43%; r
2
cv
= 1.00
b
Merino crossbreed
s
Semimembranosus
Freeze-dried ground
K
(49)
(7400–11500 mg kg
–1
) R; PLS 1100–2500 2
nd
Deriv SEP = 600 mg kg
–1
;
r
2
cv
= 0.74
b
P
(51)
(5000–10600 mg kg
–1
) R; PLS 1100–2500 1
st
Deriv SEP = 9 00 mg kg
–1
;
r
2
cv
= 0.77
b
Na
(48)
(831–1629 mg kg
–1
)
R; PLS 1100–2500 Normalised SEP = 77.89 mg kg
–1
;
r
2
cv
= 0.79
b
Mg
(52)
(500–700 mg kg
–1
) R; PLS 1100–2500 1
st
Deriv SEP = 4 0 m g k g
–1
;
r
2
cv
= 0.85
b
Cu
(52)
(0.57–2.09 mg kg
–1
) R; PLS 1100–2500 1
st
Deriv SEP = 0 . 1 4 m g k g
–1
;
r
2
cv
= 0.22
b
Fe
(52)
(26.20–58.40 mg kg
–1
) R; PLS 1100–2500 Normalisation SEP = 3 . 1 5 m g k g
–1
;r
2
cv
= 0.77
b
Zn
(50)
(45.90–72.30 mg kg
–1
) R; PLS 1100–2500 Normalisation SEP = 3.59 mg kg
–1
;r
2
cv
= 0.74
b
B
(44)
(0.24–0.90 mg kg
–1
) R; PLS 1100–2500 2
nd
Deriv SEP = 0 . 12 m g k g
–1
;
r
2
cv
= 0.15
b
Mn
(40)
(0.24–0.46 mg kg
–1
) R; PLS 1100–2500 Normalisation SEP = 0 . 0 4 m g k g
–1
;
r
2
cv
= 0.08
b
Ca
(51)
(100–300 mg kg
–1
) R; PLS 1100–2500 2
nd
Deriv SEP = 5 0 m g k g
–1
;r
2
cv
= 0.24
b
Al
(51)
(2.72–8.31 mg kg
–1
) R; PLS 1100–2500 1
st
Deriv SEP = 0 . 8 6 m g k g
–1
;
r
2
cv
= 0.07
b
Protein
(103)
36.20–76.09%DM) R; PLS NA NA SEP = 1.98%; R
2
= 0.96
Fat
(103)
(7.50–55.03%DM) R; PLS NA NA SEP = 1.07%; R
2
= 0.99
Calcium
(103)
(0.99–4.41%DM) R; PLS NA NA SEP = 0.3 0%; R
2
= 0.90
Phosphorous
(103)
(0.60–2.28%DM) R; PLS NA NA SEP = 0.11%; R
2
= 0.91
(n)
Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language; NA: not avail-
able
Table 5(c). Applications of near infrared spectroscopy in analysis of muscle food composition—lamb and sheep meat products.

14 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength ranges
(nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To measure
organic and inor-
ganic composition
of whole broiler
carcass**
Broiler NA Ground freeze-dried
sample of whole
carcasses
Dry matter
(103)
(26.41–
43.47%DM)
R; PLS NA NA SEP = 1.83%; R
2
= 0.82 73
To predict chemi-
cal components
of dried chicken
carcass in accord-
ance with different
growing systems
Broiler chicken: Fast
growing; Laying hen;
Slow growing
Whole carcass
excluding gut contents
Dry ground chicken
carcasses
Crude protein
(243)
(484.7–
667.4 g kg
–1
DM)
R; PLS 400–2498 SNV-D + 2
nd
Deriv RMSECV = 20.12 g kg
–1
DM r
2
cv
= 0.85;
RMSECV⋅SD
–1
= 0.44
74
Fat
(241)
(151.5– 34 6.6 g kg
–1
DM) R; PLS 400–2498 SNV-D + 2
nd
Deriv RMSECV = 17.23 g kg
–1
DM ;
r
2
cv
= 0.93; ;RMSECV⋅SD
–1
= 0.26
Ash
(237)
(76.7–110.8 g kg
–1
DM) R; PLS 400–2498 SNV-D + 2
nd
Deriv RMSECV = 7.95 g kg
–1
DM ; r
2
cv
= 0.65;
;RMSECV⋅SD
–1
= 0.57
To investigate
chemical composi-
tions of chicken
from laying hens
enriched with dif-
ferent sources of
omega-3 FA sup-
plements
Chicken: Laying hens
fed with four differ-
ent diets—Control
diet; Extruded lin-
seed; Ground linseed
and n-3 of marine
origin
Breast meat
(pectoralis superfi cialis)
Ground freeze-dried
samples
Dry matter
(68)
(91.8–94.8 g 100g
–
1
DM)
R; PL S 1100–2498 DT + 1
st
Deriv
RMSECV = 0.19 g 100g
–1
DM; r
2
cv
= 0.91
75
Crude protein
(70)
(83.0–
93.5 g 100 g
–1
DM)
R; PL S 1100–2498 SN V-D + 1
st
Deriv RMSECV = 0.74 g 100g
–1
DM;
r
2
cv
= 0.91
Fat
(69)
(1.9–11.8 g 100g
–1
DM ) R; PLS 1100–2498 DT + 1
st
Deriv RMSECV = 0.24 g 100g
–1
DM;
r
2
cv
= 0.99
Ash
(69
(4.0–7.5 g 100g
–1
DM ) R; PLS 1100–2498 SNV-D + 1
st
Deriv RMSECV = 0.65 g 100g
–1
DM;
r
2
cv
= -0.004
Cholesterol
(67)
(185 –272 mg 100g
–1
DM)
R; PL S 1100–2498 SN V-D + 2
nd
Deriv RMSECV = 0.14 mg 100g
–1
DM;
r
2
cv
= 0.3 4
TBARS
(35)
(0.11–0.49 μg) R; PLS 1100–2498 MSC +
2
Deriv RMSECV = 0.08 μg ; r
2
cv
= 0.53
C20:2 n-6
(28)
(0.00–0.03 g 100g
–
1
DM)
R; PL S 1100–2498 SN V-D + 1
st
Deriv RMSECV = 0.003 g 10 0g
–1
DM;
r
2
cv
= 0.67
C20:3 n-6
(35)
(0.00–0.07 g 100g
–
1
DM)
R; PL S 1100–2498 SN V-D + 1
st
Deriv RMSECV = 0.011 g 100g
–1
DM ;
r
2
cv
= 0.57
C20:5 n-3
(43)
(0.00–0.12 g 100g
–
1
DM)
R; PL S 1100–2498 SN V-D + 1
st
Deriv RMSECV = 0.011 g 100g
–1
DM;
r
2
cv
= 0.71
To predict major
nutritional profi les
of ostrich meat
Ostrich Tenderloin
(Ambiens); Big drum
(Iliofibularis); Fan fi llet
(Gastrocnemius)
Freeze-dried minced
meat
Ash
(141)
(4.31–6.07%) R; PLS 1100–2500 Normalisation SEP = 0.23%; r
2
= 0.50
b
; SEL = 0.05%
51
Dry matter
(142)
(94.07–100.42%) R; PLS 1100–2500 2
nd
Deriv SEP = 0.72%; r
2
= 0.71
b
; SEL = 0.27%
Crude protein
(150)
(84.33–94.63%) R; PLS 1100–2500 2
nd
Deriv SEP = 0.64%; r
2
= 0.94
b
; SEL = 0.47%
Fat
(153)
(1.41–8.98%) R; PLS 1100–2500 2
nd
Deriv SEP = 0.18%; r
2
= 0.98
a
; SEL = 0.29%
Table 5(d). Applications of near infrared spectroscopy in analysis of muscle food composition—poultry and bird meat products.

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 15
raising the possibility of using NIR analysis for on-site quality
screening during fi sh production.
Fatty acids
Recent applications of near infrared spectroscopy to animal-
derived foods have placed more focus on the assessment of
fatty acid (FA) profi les, especially in meat products marketed
on the basis of quality marks or standards. Knowledge of
fatty acid classes and their relative ratios in muscle foods
is useful in identifying specifi c animal genotypes produced
under defi ned feeding regimes and in providing feedback on
feed formulations.
83
NIR prediction models for fatty acids have
commonly been constructed using two approaches, a quanti-
tative model of individual fatty acid constituents or a model to
predict fatty acids belonging to the same class, for instance
total saturated fatty acids (SFA), total monounsaturated fatty
acids (MUFA) and total polyunsaturated fatty acids (PUFA).
In many cases, subcutaneous fat and backfat were used as
samples for ensuring the traceability of feeding regime and
animal husbandry, grass or grain feed, organic or conventional
housing.
Overall published accuracies for prediction of fatty acid
content by NIR spectroscopy (homogenised or intact fresh
samples) have produced coeffi cients of determination ranging
from 0.03 to 0.98 with relative prediction errors between
0.01% and 4.45%. Much lower prediction errors (≤0.01%)
were achieved with freeze-dried samples although such
models are not convenient to deploy industrially. Predictive
accuracy was generally improved when fatty acids were
combined into classes based on degree of saturation
56,68,70,76
although this was not always the case. Poor prediction was
normally due to low concentrations of individual fatty acids
present in the meat, a narrow range of reference values and
difficulties arising from the very similar spectral profiles
exhibited by fatty acids in general.
84
One successful applica-
tion involved certifi ed Iberian ham, a luxury product much
prized in Europe. Iberian pigs are required to be raised
under a specifi c montanera diet, rich in acorns, for 4 years
to reach a high degree of marbling; ham produced from such
animals may use a Protected Designation of Origin (PDO)
label.
65,85
An on-site NIR application was examined by Pérez-
Marín et al.
65
[Table 5(b)] which involved analysis of skin to
predict palmitic acid (C16:0), stearic acid (C18:0), oleic acid
(C18:1) and linoleic acid (C18:2) concentrations as indica-
tors of dietary history. The results of off-line (450–2300 nm)
and in-line applications (1100–2300 nm) were compared.
Models of moderate accuracy (RMSECV = 0.36% to 1.42%;
r
2
cv
= 0.60 to 0.77) were found in the in vivo study whereas
better performance in predicting palmitic and oleic acids
(RMSECV = 0.82% and 1.48%; r
2
cv
= 0.87 and 0.80, respectively)
was found for the in-line application. Rabbit meat [Table
5(e)] is considered to have a healthy fat profi le
78
and one
study attempted to discriminate between meat from rabbits
reared under conventional and organic systems. Coeffi cients
of determination in cross-validation were much improved
by combining fatty acids into subgroups of SFA (r
2
cv
= 0.85,
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength ranges
(nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To evaluate chemi-
cal components
supporting quali-
tative analysis of
food authentica-
tion regarding
free-range Guinea
fowl under EC
Regulations
2092/91 and
1804/99
Guinea fowl:
Numida meleagris
Breast
(pectoralis superficialis);
Thigh
(biceps femoris, liotibial)
Homogenised
Ash
(60)
(1.32–0.94%) R; MLR 1445–2348 NA RMSECV = 0.34%; r
2
cv
= 0.4 6;
SEL = 0.0 95%
42
Fat
(68)
(5.08–0.76%) R; MLR 1445–2348 NA RMSECV = 0.22%; r
2
cv
= 0.8 3;
SEL = 0.12%
Protein
(59)
(26.85–20.56%) R; MLR 1445–2348 NA RMSECV = 1.96%; r
2
cv
= 0.76;
SEL = 0.35%
Dry Matter
(57)
(24.03–28.24%) R; MLR 1445–2348 NA RMSECV = 1.76%; r
2
cv
= 0.54;
SEL = 0.61%
(
n
) Number of samples in calibration set;
a
the total number of samples if (n) was not reported in original reference ;
b
calculated from correlation coeffi cient (r) reported in original references.
f
Mean ± SD; Deriv = derivatives; ** The original article was written in another language;
NA: not available.
Table 5(d) (continued). Applications of near infrared spectroscopy in analysis of muscle food composition—poultry and bird meat products.

16 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength ranges
(nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To provide a fatty
acid profi le in
authentication of
rabbit meat raised
from conventional
system and organic
farm
Rabbit meat:
Conventional system;
Organic system
Hind legs Ground meat Myristic—C14:0
(103)
(1.66–
3.12%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 0.26%; r
2
cv
= 0.21;
RPD = 1.12
76
Palmitic—C16:0
(102)
(22.85–
34.76%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 1.21%; r
2
cv
= 0.8 3;
RPD = 2.46
Palmitoleic—C16:1
(100)
(0.91–6.83%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 0.64%; r
2
cv
= 0.77;
RPD = 2.08
Stearic—C18:0
(96)
(5.03–9.74%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 0.63%; r
2
cv
= 0.50;
RPD = 1.41
Oleic—C18:1 n-9
(99)
(18.52–
30.18%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 1.26%; r
2
cv
= 0.8 4;
RPD = 2.49
Vaccenic—C18:1 n-7
(102)
(0.96–1.73%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 0.15%; r
2
cv
= 0.3 3;
RPD = 1.21
Linoleic—C18:2
(100)
(14.99–
41.19%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 2.08%; r
2
cv
= 0.91;
RPD = 3.29
α-linoleic—C18:3
(101)
(1.82–
4.72%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 0.47%%; r
2
cv
= 0.59;
RPD = 1.55
Icosaenoic—C20:1
(92)
(0.19–
0.53%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 0.07%; r
2
cv
= 0.08;
RPD = 1.04
Eicosadienoic—C20:2 n-6
(97)
(0.23–0.63%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 0.08%; r
2
cv
= 0.23;
RPD = 1.14
Eicosatrienoic—C20:3 n-6
(93)
(0.15–0.47%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 0.04%; r
2
cv
= 0.54;
RPD = 1.49
Arachidonic—C20:4 n-6
(101)
(0.65–3.17%)
R; PLS 110 0–2498 Deriv + scatter correction RMSECV = 0.31%; r
2
cv
= 0.63;
RPD = 1.63
SFA
(99)
(30.26–46.03%) R; PLS 1100–2498 Deriv + scatter correction RMSECV = 1.4 3%; r
2
cv
= 0.85;
RPD = 2.60
MUFA
(99)
(20.81–37.21%) R; PL S 1100 –2498 Deriv + scatter correction RMSECV = 1.81%; r
2
cv
= 0.8 3;
RPD = 2.46
PUFA
(98)
(20.11–4 6.78%) R; PLS 1100 –2498 Deriv + scatter correc tion RMSECV = 2.03%; r
2
cv
= 0.93;
RPD = 3.6 5
n–6
(100)
(17.17–42.26%) R; PLS 1100–2498 Deriv + scatter correction RMSECV = 2.17%; r
2
cv
= 0.91;
RPD = 3.27
(n)
Number of samples in calibration set; a The total number of samples if (n) was not reported in original reference; b calculated from correlation coeffi cient (r) reported in original references; f Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language.
Table 5(e). Applications of near infrared spectroscopy in analysis of muscle food composition—other meat species

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 17
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength ranges
(nm)
Spectral
pre-treatment
Calibration
performance
Ref
Animal breeds/
variety
Selected
muscles
Sample
presentation
To determine water
and protein contents
of surimi for product
formulation
Surimi:
Theragra
chalcogramma
Surimi:
Theragra chalcogramma
Intact
Crude protein
(110)
(11.98–
16.17 g 100g
-1
)
T; PLS 900–1100 MSC + 2
nd
Deriv RMSECV = 0.13 g 100g
–1
;
r
2
cv
= 0.96
b
; RPD = 10.38
77
Water content
(110)
(74.17–
83.11 g 100g
-1
)
T; PLS 900–1100 2
nd
Deriv RMSECV = 0.3 8 g 100g
–1
R
2
= 0.96
;
RPD = 7.63
To predict fat and
pigment contents
using a remote NIR
interactance probe on
whole salmon with a
comparison to analy-
sis performance on
fi sh fi llets
Farmed Atlantic
salmon: Salmo salar
Live fi sh Intact Fat
(30)
(9.0–19.5%) I; PLS 760 –1040 SN V + Normalisation
RMSEP = 1.02 % ;r
2
= 0.88
78
Gutted whole fi sh Intact
Fat
(76)
(1.3–19.5%) I; PLS 760–1040 SNV + Normalisation RMSEP = 0.6 8-1.25 %;
r
2
= 0.74–0.81
b
Astaxanthin pigment
(46)
(0.3–6.1 mg kg
–1
)
I; PLS 4 49–744 SNV + Nor malisation RMSEP = 0 . 8 8 m g k g
–1
;
r
2
= 0.72
b
Intact fi llets Intact
Fat
(30)
(1.3–19.5%) I; PLS 760–1040 SNV + Normalisation RMSEP = 1.58 % ;
r
2
= 0.69
b
Astaxanthin pigment
(76a)
(3,6–7.7 mg kg
–1
)
I; PLS 4 49–744 SNV + Nor malisation RMSEP = 0 . 4 2 m g k g
–1
;
r
2
= 0.85
b
To measure moisture
and water activity in
relation to discrimi-
nant analysis of fresh
and frozen-thawed
sole**
Sole: Solea valgaris — Minced fresh muscle
Moisture
(76)
(78.8 ± 1.27 %)
f
R; PLS 1100–2500 SNV-D RMSECV = 0.72 %;
r
2
cv
= 0.67
79
A
w
(96)
(0.985 ± 0.01)
f
R; PLS 1100–2500 SNV-D RMSECV = 0.004;
r
2
cv
= 0.69
(n)
Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language.
Table 5(f). Applications of near infrared spectroscopy in analysis of muscle food composition—seafood and fi sh-based products.

18 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
RMSECV = 1.43%), MUFA (r
2
cv
= 0.83, RMSECV = 1.81%), PUFA
(r
2
cv
= 0.93, RMSECV = 2.03%) and omega-6 or n-6 (r
2
cv
= 0.91,
RMSECV = 2.17%).
Water
Water accounts for approximately 75% of meat composition.
NIR applications to measure water content have generally
been successful due primarily to the high absorbance of water
in this spectral region. Control of water content is impor-
tant in meat preservation processes such as curing, smoking
and drying
86
and reports have been published which studied
minced or homogenised fresh, cooked or lyophilised sample
material. Collell et al.
67
reported a good model for meas-
uring water content in homogenised fermented pork sausage
(r
2
= 0.99; RMSECV = 0.62%) and for water activity (r
2
= 0.99;
RMSECV = 0.007) using a NIR remote probe to monitor the
drying of fermented pork sausages. For the studies summa-
rised in Tables 5(a)–(f), low prediction accuracies for water
content have arisen because of either a narrow range of refer-
ence values or poor accuracy in the reference methods.
84
In the seafood industry, water content is an important
quality indicator.
87
Structural changes in water molecules
are observable in NIR spectra; using the interaction between
water and other chemical species can be useful in monitoring
physiochemical quality traits of fi shery products including the
discrimination between fresh and frozen-then-thawed fi sh. In
2008, Fasolato et al.
79
predicted water content in sole with r
2
cv
equal to 0.67 and an associated RMSECV of 0.72%. In contrast,
a higher prediction accuracy (R
2
= 0.96; RMSECV = 0.38 g 100 g
−1
)
was obtained by Uddin et al.
77
in a study of a surimi product.
The ratios (RPD) of standard deviation to standard error of
validation, a measure of the prediction accuracy, were 1.76 in
Fasolato et al. and 7.63 in Uddin et al.
Other chemical components
A range of other molecular constituents of muscle foods has
been studied by NIR. Some are of interest because of human
health concerns (cholesterol) while others are related to
sensory properties, for example, collagen and texture.
In 2009, Bajwa et al.
57
demonstrated that cholesterol content
and total calories in raw beef could be predicted by visible–
NIR spectroscopy in the 350–1075 nm region. In this experi-
ment, prediction accuracies for NIR-based models of raw
beef patties (r
2
= 0.80 and 0.96 for cholesterol content and total
calories, respectively, with associated SEP values of 15.0 mg g
−1
and 0.19 Kcal g
−1
) were compared to those for the cooked coun-
terparts (r
2
= 0.79 and 0.87 for cholesterol content and total
calories respectively with associated SEP fi gures of 6.2 mg g
−1
and 0.07 Kcal g
−1
). In contrast, a lower prediction accuracy
for the PLS regression model (1100–2498 nm) for the meas-
urement of cholesterol content in chicken meat (r
2
cv
= 0.34;
RMSECV = 0.14 mg 100 g
−1
DM) was reported by Berzaghi et al.
75
despite the fact that ground, freeze-dried samples were used.
To help predict beef tenderness, Prieto et al.
36
studied
collagen content prediction and reported a r
2
value of 0.47
with a corresponding RMSECV equal to 3.82 g kg
−1
DM, quite a
limited accuracy which may be explained, in part, by a masking
effect of the dominant fibrous protein in muscle tissue.
4
Furthermore, hydroxyproline in pork (an indicator for collagen)
was predicted by NIR spectroscopy; values of R
2
equal to 0.64
and SEP of 0.05% (w/w)
were reported.
66
This study suggests
that better prediction accuracies may be possible if a wider
range of reference values could be obtained, i.e. greater than
0.13 g 100 g
−1
to 0.74 g 100 g
−1
.
Other minor components studied include mineral and ash
contents. The merit in developing NIR models to predict total
ash contents is questionable given that mineral compounds
generally have no signal in the NIR spectra range. Seemingly
accurate calibrations are often based on secondary absorp-
tion by associated compounds and to the extent that these
are always present in the sample matrix in the same relative
concentrations, such calibrations may have very limited use.
Reported prediction errors varied between 0.15% and 1.7%
DM. In a study of the mineral content of freeze-dried mutton,
it was claimed that major cations (Fe, Na, K, P, Mg and Zn)
could be predicted with r
2
cv
values between 0.74 and 0.79
and corresponding SEP fi gures of 3.15 mg kg
−1
to 900 mg kg
−1
.
Prediction of other minor elements was not successful. The
stated basis for this investigation was that minerals exist in
association with organic acids,
99
thereby rendering them indi-
rectly detectable.
Measurement of physical
properties
Mechanical tenderness, water-holding capacity, pH and
colour are quality indicators of physical properties in muscle
foods. Potential applications of NIR spectroscopy in this area
are summarised in Tables 6(a)–(e) and are discussed in this
section on the basis of each individual quality attribute.
Measurement of mechanical tenderness
Tenderness is of critical concern to consumers of muscle
foods and signifi cantly infl uences re-purchasing behaviour.
100
In addition to sensory analysis, meat tenderness can be
mechanically determined through measurement of the shear
force (newton or kilogram unit) necessary to cut sampled
portions of, generally, cooked muscle; Warner–Bratzler shear
force (WBSF) and sliced shear force (SSF) are the two most
common measurements reported.
101
Accuracy of NIR tenderness measurement in meat is
generally quite poor with low correlations between measured
tenderness and NIR predicted values being encountered in
many studies. Coefficient of determination values greater
than 0.60 have rarely been obtained while broad ranges of
prediction errors were reported: 0.81–59 N. However, Ripoll et
al.
33
predicted WBSF values in beef with a r
2
value of 0.74 and
a corresponding SEP of 10.4 kg. Samples used in this study
were claimed to have a higher coeffi cient of variation (%CV)
than others.
33
Xia et al.
102
investigated correlation between

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 19
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement
mode; regression
method
Wavelength
ranges (nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To determine physical
properties of beef from
NIR spectra collected
after different ageing
periods
Beef: Young
Maronesa bulls
Longissimus
thoracis
et lumborum
Intact
L* colour at 0 min PM
(108)
(27.62–42.70) R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 1.16 ; r
2
cv
= 0.80; RPD = 2.22
88
L* colour at 60 min PM
(109)
(28.89–
43.78)
R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 1.36 ; r
2
cv
= 0.75; RPD = 2.07
a* colour at 0 min PM
(104)
(12.88–21.90) R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 1.09 ; r
2
cv
= 0.23; RPD = 1.14
a* colour at 60 min PM
(100)
(15.23–
26.40)
R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 1.28 ; r
2
cv
= 0.29; RPD = 0.90
b* colour at 0 min PM
(109)
(0.88–5.59) R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 0.75 ; r
2
cv
= 0.27; RPD = 1.17
b* colour at 60 min PM
(99)
(2.96–11.37) R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 0.99 ; r
2
cv
= 0.4 6; RPD = 1.37
pH at 24 hrs PM
(27)
(5.50–6.67) R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 0.10 ; r
2
cv
= 0.91; RPD = 3.17
Sarcomere length
(30)
(1.51–1.84 μm) R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 010 μm ; r
2
cv
= 0.02; RPD = 0.8 4
WBSF
(112)
(3.85–19.88 kg) R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 2.67 kg ; r
2
cv
= 0.53; RPD = 1.46
Cook loss
(99)
(2.91–16.81%) R; PLS 400–2498 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 0.08 %; r
2
cv
= 0.02; RPD = 1.01
To determine beef
tenderness and WHC
for the improvement of
prediction accuracy
Beef:: SEUROP
classifi cation
consisting of 12
different breeds and
cross-breeds
Longissimus
thoracis
Homogenised
WHC
(190a)
(21.17–29.17 %) R; PLS 408–2492.8 1
st
Deriv SEP = 1.3 38% ; r
2
= 0.892; RPD = 1.76
33
WBSF
(190a)
(2.60–10.00 kg) R; PLS 1108 –2492.8 SNV-D + 1
st
Deriv SEP = 1.05 8 kg ; r
2
= 0.743; RPD = 1.4 4
Stress at 20%
(190a)
(3.54–17.08 N cm
–2
) R; PCR 1108–2492.8 SNV-D + 1
st
Deriv SEP = 2.6 68 N cm
-2
; r
2
= 0.365; RPD = 1.31
Stress at 80%
(190a)
(20.47–72.00
N cm
–2
)
R; PCR 1108 –2492.8 SNV-D + 1
st
Deriv SEP = 6.238 N cm
-2
; r
2
= 0.4 32; RPD = 1.71
To estimate physical
parameters of meat
from adult steers (oxen)
under quality mark and
young cattle
Beef: Adult steers Longissimus
thoracis
Homogenised
L* c olo ur
(50)
(32.13–40.68) R; PLS 1100–2500 MSC and 1
st
Deriv RMSECV = 1.50; r
2
cv
= 0.585 ;RPD = 1.24;
SEL = 1.57
89
a* colour
(51)
(17.52–23.67) R; PLS 1100–2500 MSC and 2
nd
Deriv RMSECV = 1.58; r
2
cv
= 0.008 ;RPD = 0.98;
SEL = 1.74
b* colour
(52)
(3.01–9.98) R; PLS 1100–2500 MSC and 1
st
Deriv RMSECV = 1.46; r
2
cv
= 0.3 45 ;RPD = 1.16;
SEL = 1.13
Drip loss
(53)
(1.25–2.99 %) R; PLS 1100–2500 MSC and 2
nd
Deriv RMSECV = 0.3 6 %; r
2
cv
= 0.258 ;RPD = 1.04;
SEL = 0.24
Press loss
(50)
(19.01–28.70 %) R; PLS 1100–2500 MSC and 1
st
Deriv RMSECV = 2.08 %; r
2
cv
= 0.476 ;RPD = 1.11;
SEL = 2.34
Cook loss
(48)
(20.90–27.58 %) R; PLS 1100–2500 None RMSECV = 1.61 %; r
2
cv
= 0.138;RPD = 1.03;
SEL = no report
WBSF
(49)
(43.35–90.52 N) R; PLS 1100–2500 MSC RMSECV = 10.00 N; r
2
cv
= 0.4 48 ;RPD = 1.18;
SEL = 14.56 N
pH
(50)
(5.52–5.88) R; PLS 1100–2500 2
nd
Deriv RMSECV = 0.06; r
2
cv
= 0.410 ;RPD = 1.12;
SEL = 0.02
Table 6(a). Applications of near infrared spectroscopy in measurement of muscle food properties—beef and beef products

20 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement
mode; regression
method
Wavelength
ranges (nm)
Spectral
Pre-treatment
Calibration performance
Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To estimate physical
parameters of meat
from adult steers (oxen)
under quality mark and
young cattle
Young cattle Longissimus
thoracis
Homogenised
L* c olo ur
(60)
(30.97–48.02) R; PLS 1100–2500 1
st
Deriv RMSECV = 1.56; r
2
cv
= 0.869 ;RPD = 2.17;
SEL = 1.66
89
a* colour
(62)
(11.05–18.25) R; PLS 1100–2500 1
st
Deriv RMSECV = 1.15; r
2
cv
= 0.707 ;RPD = 1.58;
SEL = 1.06
b* colour
(61)
(-1.36–10.51) R; PLS 1100–2500 None RMSECV = 1.08; r
2
cv
= 0.901 ;RPD = 2.51;
SEL = 0.88
Drip loss
(65)
(1.61–4.30 %)
R; PL S 1100–2500 MSC + 2
nd
Deriv
RMSECV = 0.55; r
2
cv
= 0.195 ;RPD = 1.02;
SEL = 0.41
Press loss
(65)
(12.97–31.14 %) R; PLS 1100–2500 MSC + 1
st
Deriv RMSECV = 2.51 ; r
2
cv
= 0.576 ;RPD = 1.3 0;
SEL = 2.41
Cook loss
(60)
(21.15–30.45 %) R; PLS 1100–2500 MSC + 2
nd
Deriv RMSECV = 2.4 5; r
2c
v
= 0.001 ;RPD = 0.97;
SEL = no report
WBSF
(61)
(49.33–129.45 N) R; PLS 1100–2500 2
nd
Deriv RMSECV = 15.89; r
2
cv
= 0.167 ;RPD = 1.07;
SEL = 18.20 N
pH
(61)
(5.41–5.89) R; PLS 1100–2500 None RMSECV = 0.08; r
2
cv
= 0.472 ;RPD = 1.26;
SEL = 0.023
To predict technological
attributes in beef for on-
line NIRS implementa-
tion in the abattoir
Beef:
cross-bred sired
by Aberdeen Angus
or Limousin sire
breeds
Longissimus
thoracis
Intact on
carcass
muscle
L* colour
(178)
(31.49–53.93) R; PLS 350–1800 None RMSECV = 0.96 ;r
2
cv
= 0.8 3; RPD = 2.47
107
a* colour
(176)
(16.71–28.89) R; PLS 350–1800 None RMSECV = 0.95 ;r
2
cv
= 0.76; RPD = 2.02
b* colour
(171)
(3.40–15.40) R; PLS 350–1800 None RMSECV = 0.69; r
2
cv
= 0.8 4; RPD = 2.4 8
Cook loss
(130)
(16.26–29.84 %) R; PLS 350–1800 None RMSECV = 2.35 % ;r
2
cv
= 0.23; RPD = 1.14
Volodkevitch shear force
(172)
(16.95–
98.33 N)
R; PLS 350–1800 MSC + 1
st
Deriv RMSECV = 12.70 N ;r
2
cv
= 0.21; RPD = 1.11
SSF at 3 days post-mortem
(176)
(89.28
– 85.75 N)
R; PLS 350–1800 MSC + 2
nd
Deriv RMSECV = 55.76 N r
2
cv
= 0.31; RPD = 1.25
SSF at 14 days PM
(176)
(61.43–273.28
N)
R; PLS 350–1800 None RMSECV = 28.49 N ;r
2
cv
= 0.23; RPD = 1.14
To measure meat qual-
ity indicators pre- and
post-rigor
Beef: Hereford Longissimus
lumborum
Intact
pH
(530)
(5.15–7.17) R; PLS 400–1700 SNV+GLS RMSEP = 0.20 ; r
2
= 0.8 3
58
WBSF
(257)
(19–265 N) R; PLS 400–1700 GLS RMSEP = 28 N ; r
2
= 0.58
WHC
(301)
(0.4–25.6 cm
2
g
-1
)R; PLS400–1400GLS RMSEP = 2.8 cm
2
g
-1
; r
2
= 0.67
Table 6(a) (continued). Applications of near infrared spectroscopy in measurement of muscle food properties—beef and beef products

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 21
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement
mode; regression
method
Wavelength
ranges (nm)
Spectral
pre-treatment
Calibration
performance
Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To determine the effi -
ciency of NIRS measure-
ment of beef tenderness
compared to measure-
ments made using vis-
ible spectra
Beef: Heifer
Striploins Intact WBSF
(87)
(29.4–75.5 N)
R;
Linear regression
375–700 NA RMSEP = 0.75 –0.76 N ; r
2
= 0.22 – 0.23
90
R;
Linear regression
705–1100 NA RMSEP = 0.781- 0.784 N ; r
2
= 0.14
R;
Linear regression
375–1100 NA RMSEP = 0.761-0.767 N ; r
2
= 0.19-0.20
R; PCR 375–700 NA RMSEP = 0.814 N ; r
2
= 0.07
R; PCR 705–1100 NA RMSEP = 0.788 N ; r
2
= 0.13
R; PCR 375–1100 NA RMSEP = 0.781 N ; r
2
= 0.15
To develop an NIR pre-
diction model from col-
lected samples at meat
packer after 48 hours
ageing**
Beef Tenderloin;
Ribeye;
Topside;
Shin;
Striploin
NA
pH
(114a)
(NA) R; PLS 950–1650 MSC, SNV + 1
st
Deriv RMSEC = 0.75 ; r
2
= 0.38
b
59
L* - c o lo u r
(114a)
(NA) R; PLS 950–1650 MSC, SNV + 1
st
Deriv RMSEC = 1.78 ; r
2
= 0.67
b
a*-colour
(114a)
(NA) R; PLS 950–1650 MSC, SNV + 1
st
Deriv RMSEC = 1.4 0 ; r
2
= 0.75
b
b*-colour
(114a)
(NA) R; PLS 950–1650 MSC, SNV + 1
st
Deriv RMSEC = 1.76 ; r
2
= 0.57
b
WBSF
(114a)
(NA) R; PLS 950–1650 MSC, SNV + 1
st
Deriv RMSEC = 1.07 ; r
2
= 0.23
b
To predict mechanical
tenderness of beef by
VIS-NIRS with reference
values obtained from
different methods
Beef: Select car-
casses;
Low Choice; Top
Choice;USDA Prime
Ribeye rolls
Longissimus
muscle
Intact
MORSpf (40a) (20.52 – 24.85 N) R; PLS 400–2498
None RMSEP = 4.15 N; r
2
= 0.47 ; RPD = 1.13 ;
Robust = 1.08
91
2nd Deriv RMSEP = 3.15 N; r
2
= 0.74 ; RPD = 1.50;
Robust = 1.33
MORSta (40a) (246.34–281.28 N mm) R; PLS
400–2498
None RMSEP = 41.9 N mm; r
2
= 0.43 ; RPD = 1.11;
Robust = 1.03
2nd Deriv RMSEP = 33.9 N mm; r
2
= 0.69 ; RPD = 1.3 8;
Robust = 1.46
WBSF (40a) (3.1–3.78 N) R; PLS
400–2498
None RMSEP = 0.73 kg; r
2
= 0.31; RPD = 1.05;
Robust = 1.06
2nd Deriv RMSEP = 0.6 5 kg; r
2
= 0.53 ; RPD = 1.18;
Robust = 1.41
(n)
Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language;
NA: not available
Table 6(a) (continued). Applications of near infrared spectroscopy in measurement of muscle food properties—beef and beef products

22 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Table 6(b). Applications of near infrared spectroscopy in measurement of muscle food properties —pork and pork products.
Study purpose
Sample descripƟ on
Quality parameters
(n)
(measurement ranges)
Measurement
mode; regression
method
Wavelength ranges
(nm)
Spectral
pre-treatment
CalibraƟ on performance Ref.
Animal breeds/
variety
Selected muscles
Sample pres-
entation
To determine physical
parameters of pork in-line
at the slaughter house
Pork: Yorkshire and
Piétrain crossbreds
Longissimus tho-
racis;
Longissimus
lumborum;
Semimembra-nosus
Intact
carcass
a* colour
(102)
(NA) R; PLS2 400–750 2
nd
Deriv RMSECV = 1.0; r
2
cv = 0.31
60
Drip loss
(102)
(0.6–8.6%)
R; PLS2 1400 2
nd
Deriv RMSECV = 0.004% ; r
2
cv = out of calculation
R; PLS2 1400 nm with other
parameters
2
nd
Deriv RMSECV = 1.6% r
2
cv = 0.06
To predict IMF content
and compare the infl u-
ence of muscle types to
predictability of NIRS-
based models
Pork: Crossbreeds
with Duroc
Longissimus dorsi;
Intact IMF
(45)
(0.88–5.82%FM) R; PLS 400–2500 SNV-D + 1
st
Deriv RMSECV = 0.26%FM ; r
2
cv = 0.8 8; RPD = 2.9
55
IMF
(43)
(0.88–5.82%FM) R; PLS 1100–2500 SNV-D + 1
st
Deriv RMSECV = 0.28%FM ; r
2
cv = 0.87; RPD = 2.9
Semitendinosus Intact
IMF
(46)
(2.03–8.48%FM) R; PLS 400–2500 SNV-D + 1
st
Deriv RMSECV = 0.49%FM ; r
2
cv = 0.91; RPD = 3.3
IMF
(46)
(2.03–8.48%FM) R; PLS 1100–2500 SNV-D + 1
st
Deriv RMSECV = 0.49%FM ; r
2
cv = 0.91; RPD = 3.3
Longissimus dorsi
and Semitendinosus
Intact
IMF
(90)
(0.88–8.48%FM) R; PLS 400–2500 SNV-D + 1
st
Deriv RMSECV = 0.3 4%FM ; r
2
cv = 0.97; RPD = 5.3
IMF
(92)
(0.88–8.48%FM) R; PLS 1100–2500 SNV-D + 1
st
Deriv RMSECV = 0.41%FM ; r
2
cv = 0.95; RPD = 4.6
To evaluate WHC and other
physical parameters of
pork
Pork: crossbreed
Dutch lan-
drace + Finnish lan-
drace and Yorkshire
Longissimus Intact
L* c olo ur
(119–165)
(43.7–59.5) R; PLS 400–800 None SEP = 1.25–1.64 ; r
2
= 0.60– 0.79
b
61
a* colour
(116–165)
(13.0–17.6) R; PLS 400–1100 1
st
Deriv + SN V-D SEP = 0.67–0.74 ; r
2
= 0.4 4–0.52
b
b* colour
(117–168)
(4.3–8.1) R; PLS 400–800 None SEP = 0.42–0.52 ; r
2
= 0.45 –0.71
b
Drip loss
(117–165)
(2.5–11.9 %) R; PLS 400–800 None SEP = 1.14–1.42% ; r
2
= 0.31– 0.35
b
pH
(117–166)
(5.25–5.71) R; PLS 400–1100 2
nd
Deriv SEP = 0.047–0.071; r
2
= 0.4 0–0.71
b
To measure technological
traits of pig meat
Pork: Large
white; Landrace;
Hampshire;
Piétrain; Duroc
Longissimus dorsi
Intact
pH at 24 h PM
(146)
(5.21–6.20) R; PLS 400–1100 SNV-D + 1
st
Deriv SEP = 0.08; r
2
= 0.54 ;RPD = 1.5; SEL = 0.05
92
L* c olo ur
(148)
(41.99–59.42) R; PLS 400–1100 SNV-D + 1
st
Deriv SEP = 2.0; r
2
= 0.47 ;RPD = 1.5; SEL = 3.1
a* colour
(146)
(3.51–17.02) R; PLS 400–1100 SNV-D + 1
st
Deriv SEP = 0.8; r
2
= 0.75 ;RPD = 2.2; SEL = 1.0
b* colour
(148)
(0.65–9.62) R; PLS 400–1100 SNV-D + 1
st
Deriv SEP = 0.7; r
2
= 0.59 ;RPD = 1.8; SEL = 0.9
EZ drip loss at 24 h PM
(144)
(0.10–16.15 % )
R; PL S 400–1100 SNV-D + 1
st
Deriv SEP = 1.4%; r
2
= 0.53 ;RPD = 1.9; SEL = 1.1
EZ drip loss at 48 h PM
(146)
(0.23–18.07 % )
R; PL S 400–1100 SNV-D + 1
st
Deriv SEP = 1.7%; r
2
= 0.56 ;RPD = 1.7; SEL = 1.3
To predict pork drip loss Pork
Longissimus dorsi
Minced Drip loss
(135)
(6.3 ± 3.1 %)
f
R; CP-ANN 400–2500 Normalisation RMSEP = 2.6%; r
2
= 0.28
93
R; BP-ANN 400–2500 Normalisation RMSEP = 2.5%; r
2
= 0.36
To determine WHC in fresh
pork
**
Pork Pork loins Intact
Drip loss
(106a)
(NA)
R; PLS NA Smoothing + 2
nd
Deriv
R
2
= 0.55
b
94
Press loss
(106a)
(NA) R; PLS NA Smoothing + 2
nd
Deriv
R
2
= 0.62
b
To determine pH of fresh
pork on moving conveyor
**
Pork
Longissimus dorsi
Fresh meat
on moving
conveyor
pH
(NA)
(NA) R; PLS 510–980 MSC + 1
st
Deriv RMSEP = 0.0 45; r
2
= 0. 8 6 95

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 23
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement
mode; regres-
sion method
Wavelength
ranges (nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To evaluate mechanical ten-
derness of sheep meat from
different diet controls
Lamb: Commercial
lamb; Nutrient sup-
plemented lamb
Longissimus
thoracis et
lumborum
Intact WBSF
(260)
(25–95 N) I; PLS 800–1000 2
nd
Deriv RMSEP = 12.2 N; r
2
= 0.85 97
To observe pH of lamb in
relation to eating quality
Lamb: Texel; Scottish
blackface
Longissimus
thoracis
Intact
pH at 45 min PM
(231)
(5.89–7.02) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 0.260; r
2
cv
= 0.025
71
pH at 3 hrs PM
(231)
(5.51–6.88) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 0.214; r
2
cv
= 0.174
pH at 24 hrs PM
(231)
(5.31–6.38) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv RMSECV = 0.157; r
2
cv
= 0.188
Table 6(c). Applications of near infrared spectroscopy in measurement of muscle food properties—lamb and sheep meat.
Table 6(b) (continued). Applications of near infrared spectroscopy in measurement of muscle food properties —pork and pork products.
Study purpose
Sample descripƟ on
Quality parameters
(n)
(measurement ranges)
Measurement
mode; regression
method
Wavelength ranges
(nm)
Spectral
pre-treatment
CalibraƟ on performance Ref.
Animal breeds/
variety
Selected muscles
Sample pres-
entation
To predict pork WHC
using different refer-
ence measurements
Pork: The
progeny of
Landrace×Large
white dams;
Piétrain sires
Longissimus dorsi
Intact
EZ drip loss
(143–146)
(4.4 ± 1.7%)
f
R; PLS
400–250 0 Deriv + SNV-D SEP = 1.32 %; r
2
= 0.28; SEL = 1.1
96
1100–2500 Der iv + SN V-D SEP = 1.3 3%; r
2
= 0.27; SEL = 1.1
400–1100 Deriv + SN V-D SEP = 1.20%; r
2
= 0.4 4; SEL= 1.1
Cook loss
(145–146)
(31.2 ± 3.3%)
f
R; PLS
400–250 0 Deriv + SNV-D SEP = 2.41%; r
2
= 0.26; SEL = 1.8
1100–2500 Der iv + SN V-D SEP = 2.4 0%; r
2
= 0.26; SEL = 1.8
400–1100 Deriv + SN V-D SEP = 2.21%; r
2
= 0.39; SEL = 1.8
Centrifuge force
(139–143)
(12.0 ± 3.1%)
f
R; PLS
400–250 0 Deriv + SNV-D SEP = 2.16%; r
2
= 0.31; SEL = 2.0
1100–2500 Der iv + SN V-D SEP = 2.14%; r
2
= 0.30; SEL = 2.0
400–1100 Deriv + SN V-D SEP = 2.25%; r
2
= 0.31; SEL = 2.0
Tray drip loss
(144–146)
(3.2 ± 1.3%)
f
R; PLS
400–250 0 Deriv + SNV-D SEP = 1.13%; r
2
= 0.22; SEL = 0.4
1100–2500 Der iv + SN V-D SEP = 1.12%; r
2
= 0.23; SEL = 0.4
400–1100 Deriv + SN V-D SEP = 1.08%; r
2
= 0.28; SEL = 0.4
(
n
) Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language.

24 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength
ranges (nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected muscles
Sample
presentation
To predict pH
in chicken sup-
plemented with
omega-3 fatty
acid from different
sources
Chicken from
laying hens fed
with 4 different
diets – Control
diet; Extruded
linseed; Ground
linseed and n-3 of
marine origin
Breast meat (pec-
toralis superficialis)
Ground freeze-
dried samples
pH at 24hrs PM
(71)
(5.57–5.85
)
R; PL S 1100–2498 DT + 1
st
Deriv RMSECV = 0.04; r
2
cv
= 0.4 0
75
pH at 7 days PM
(69)
(5.51–5.86) R; PLS 1100–2498 SNV-D + 2
nd
Deriv RMSECV = 0.05; r
2
cv
= 0.41
To evaluate WHC
for food authenti-
cation of Spanish
free-range Guinea
fowl from different
housing
ystems
Guinea fowl:
Numida meleagris
Breast (pectoralis
superficialis);
Thigh (biceps femo-
ris, liotibial)
Homogenised WHC
(57)
(9.45–18.14 %) R; MLR 1445–2348 NA RMSECV = 4.1711 %; r
2
cv
= 0.392;
SEL = 0.589
42
To examine
technological traits
of broiler meat from
raw and cooked
meat
Mixed-gender of
broiler meat
Breast muscle Intact raw
(raw)
meat and cooked
(cooked)
meat
pH
(96)
(5.56–6.35) R; PLS 400–1850 Mean centering, MSC + 2
nd
Deriv
RMSEP = 0.15; r
2
= 0.91 98
L* colour
(96)
(41.16–55.04) R; PLS 400–1850 Mean centering, MSC + 2
nd
Deriv
RMSEP = 1.21 ; r
2
= 0.94
a* colour
(96)
(1.60–6.70) R; PLS 400–1850 Mean centering, MSC + 2
nd
Deriv
RMSEP = 0.87 ; r
2
= 0.38
b* colour
(96)
(0.32–7.04) R; PLS 400–1850 Mean centering, MSC + 2
nd
Deriv
RMSEP = 0.95 ; r
2
= 0.80
WBSF
(96)
(2.74–17.32 kg) R; PLS 400–1850 Mean centering, MSC + 2
nd
Deriv
RMSEP = 2.65 kg
(raw)
; 2.55 kg
(cooked)
;
r
2
= 0.29
(raw)
; 0.68
(cooked)
Table 6(d). Applications of near infrared spectroscopy in measurement of muscle food properties—poultry and bird species.

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 25
scattering coeffi cients of fresh bovine muscles in the visible–
NIR region (450–950 nm) and measured WBSF values. They
reported that optical scattering coeffi cients were highly corre-
lated with measured WBSF values in cooked beef samples
(P < 0.0001), suggesting a possible strategy for predicting
meat tenderness by NIR spectroscopy. Another study of lamb
tenderness included lamb samples collected after different
post-mortem storage times (0, 8, 24 and 72 h) revealing an
r
2
of 0.85 and associated RMSEP equal to 12.2 N.
97
Samples
collected at different storage times formed separate clusters
along the PLS regression line in this study, possibly suggesting
that the model recognised ageing effects rather than actual
tenderness; such an effect would pose difficulties in the
accurate prediction of new samples.
97
In 2010, Yancey et al.
91
demonstrated another way to optimise accuracy in the predic-
tion of meat tenderness. These authors compared predic-
tion accuracies of NIR models computed using two types of
reference values—WBSF and Meullenet–Owens razor shear
(MORS)—in four beef cutting grades – Select, Low choice, Top
choice and USDA Prime. Prediction accuracy from this study
was better when reference values from MORS peak force and
2
nd
derivative spectra (r
2
= 0.74; RMSEP = 3.15 N) were used;
these parameters were compared to a model based on WBSF
reference values (r
2
= 0.53; RMSEP = 6.4 N).
91
These observa-
tions highlight the importance of choosing the appropriate
reference methods for NIR spectroscopic modelling of meat
tenderness.
Water-holding capacity
According to Hamm (1986) in Warriss,
24
water-holding capacity
(WHC) is “the ability of meat to hold its own or added water
during the application of any force” which will have an effect
on product appearance. WHC can be expressed in various
ways including percentage cook loss, drip loss, press loss,
tray drip loss, centrifuge force, the EZ drip loss and express-
ible drip. If WHC is too low, a drip or exudate of juice secreted
from muscles is noticeable in packaging or in retail display,
resulting in poor product appearance at the point of purchase.
Poor WHC results in reduced water retention with concomitant
weight loss and diminished production yields for the producer
or retailer. Negative sensory consequences for the consumer
after cooking meat of low WHC include reduction in product
juiciness, a dry mouthfeel and lack of succulence.
24
There is a
clear analytical need for a simple, rapid and accurate method
for WHC prediction at various time points along the meat
distribution chain.
Table 6(a) summarises the prediction accuracy of WHC
calibrations developed for beef products using a number of
different sample processing steps (intact or homogenised
muscle) over the entire visible–NIR spectral range (400–
2500 nm) or sub-sets thereof. In the case of intact muscle,
reported models (PLS, PCR or linear regression) produced
lowest RMSEP fi gures of 1.34%
33
and 2.8 cm
2
g
−1
. Rosenvold
et at.
58
reported a calibration (400–1700 nm range ) for WHC
prediction with r
2
= 0.67 and RMSEP = 2.8 cm
2
g
−1
using NIR
reflectance spectra of intact beef with a generalised least
Study purpose
Sample description
Quality parameters
(n)
(measurement ranges)
Measurement mode;
regression method
Wavelength
ranges
(nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To measure physical
characteristics
explaining discriminant
analysis of fresh and
frozen-thawed sole**
Sole:
(Solea valgaris)
NA Minced fresh
muscle
Expressible drips
(108)
(13.4 ± 6.36%)
f
R; PLS 1100–2500 SNV-D RMSECV = 4.26%; r
2
cv
= 0.55 79
(n)
Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original
article was written in another language.
Table 6(e). Applications of near infrared spectroscopy in measurement of muscle food properties—seafood.

26 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
squares (GLS) mathematical pre-treatment of spectra. Second,
Ripoll et al.,
33
using the 1108–2492.8 nm wavelength range,
achieved a higher prediction accuracy (r
2
= 0.89, SEP = 1.34%)
although there was some suggestion of over-fi tting of the PLS
regression model in this case. Interestingly, both studies were
performed using identical reference methods, filter paper
press or percentage press loss. Theoretically, WHC is related
to the content or structure of chemical components in meat,
especially protein;
24,103
nonetheless, no significant correla-
tion between spectral absorbances and WHC was evident in
the work of Prieto et al.
89
(Figure 1). This may be due to the
complex nature of muscle proteins. Proteins in meat comprise
three groups: myofi brillar, sarcoplasmic and stroma protein.
Myofi brillar protein is most abundant (60% w/w) followed by
sarcoplasmic protein (20% w/w); both are water-soluble while
water-insoluble stroma protein is found at approximately 10%
w/w concentrations as a building unit of connective tissue.
24
It
has been reported, in relation to pork muscle, that percentage
drip loss was more highly correlated to sarcoplasmic protein
content than that of the other protein species;
104
however, it
is not possible to isolate the spectral signature of any one
protein class to establish correlations with physical properties
such as WHC.
Effect of pH
Ultimate pH in post-rigor, aged meat is the pH at 24 h post-
slaughter and is one of the well-established parameters
affecting meat quality, appearance and texture in particular.
Ultimate pH is signifi cantly related to DFD (dark, fi rm, dry)
and PSE (pale, soft, exudative) meats.
105
According to Savenije
et al.,
61
it was possible to predict pH values in 84% of pork
Longissimus dorsi samples to within 0.1 unit by NIR spectros-
copy. Acidity changes affect the internal structure of muscle
tissue which impact on light refl ection and scatter; meat with
a high pH appears darker and has overall stronger absorption
bands than meat at a lower pH.
106
The r
2
cv
and r
2
values for reported pH prediction models
have varied from 0.03 to 0.91 with associated prediction
errors ranging between 0.1 and 0.3 [Table 6(a)]. Andrés et al.
88
studied pH prediction by NIR refl ectance spectroscopy (400–
2498 nm; PLS modelling) using only a small number (n = 30)
of intact muscle samples of Maronesa bulls; the reported
prediction accuracy of pH values after 24 h post-mortem (PM)
using a cross-validation model was quite good (r
2
cv
= 0.91;
RMSECV = 0.10; range = 5.50–6.67). In other research using
Hereford beef, Rosenvold et al.
58
also obtained a low predic-
tion error (RMSEP = 0.2) with a good r
2
(0.83) for pH values
recorded from 1 to 90 h after slaughter using PLS regression
on spectral data between 400 nm and 1700 nm. In addition,
Liao et al.
95
studied the same application in fresh pork on
a conveyor belt moving at a velocity of 0.25 m s
−1
, an experi-
mental arrangement which was operationally convenient
and hygienic; spectra (510–980 nm) were used to develop a
PLSR calibration using the Kennard–Stone algorithm to select
samples for inclusion in calibration and prediction sets. Owing
to the distance between the light source and the sample in
this work, light scatter and atmospheric interference are of
concern; multiplicative scatter correction (MSC) together with
a 1
st
derivative pre-treatment were employed to correct for
scattering effects resulting in a model with RMSEP = 0.045 and
r
2
equal to 0.86 in this study.
Colour values: L*, a* and b*
Predictive performance of NIR models for colour measure-
ment in red meat—beef, pork and lamb—has ranged from
fair to good. Colour parameters are conventionally expressed
as L* (lightness), a* (red–green) and b* (yellow–blue) values.
NIR prediction accuracy (RMSECV) of meat L* values ranged
from 1.0 to1.56 units (r
2
cv
values of 0.70 to 0.87). Savenije et
al.
61
achieved an r
2
value of 0.80 and an associated SEP equal
to 1.64 in predicting L* of intact pork Longissimus dorsi muscle
(range of 43.7–59.5 units) using raw visible–NIR spectra
Figure 1. Correlation coeffi cient (r) of measured values from different WHC methods and second-derivative spectral data of oxen and
young cattle meat samples. Overall, r < 0.5 showing no meaningful correlation between absorption wavelength and reference values.
(Reprinted with permission from Meat Sci. 79, 692 (2008) Prieto et al.
89
).

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 27
between 400 nm and 800 nm. Better performance (r
2
cv
= 0.87;
RMSECV = 1.56) was claimed by Prieto et al.
89
for homogenised
Longissimus thoracis muscle from young cattle using a PLS
regression model based on absorbance in the NIR region
(1100–2500 nm). However, as part of the same study, a much
poorer coeffi cient of determination (r
2
cv
= 0.6) was found for
the same muscles from adult steers, although the prediction
error (RMSECV = 1.50) was similar. The main reason was likely
to have been the narrower range of L* units in the muscles
from the adult cattle.
NIR prediction errors for a* values ranged from 0.7 to 1.6
with associated r
2
cv
values of 0.008 to 0.76; in the case of b*,
prediction errors between 0.5 to 1.5 with r
2
cv
values between
0.3 and 0.9 were reported. One study by Prieto et al.
89
was
aimed at developing NIR spectroscopy as a tool for authenti-
cation of a premium oxen meat and commercial, young cattle
meat. In both cases, homogenised meat samples were studied
using near infrared spectra covering the 1100–2498 nm range.
Better model performance was reported for young cattle beef
(r
2
cv
of 0.71, RMSECV of 1.15 for a* value; r
2
cv
of 0.90, RMSECV of
1.08 for b* value) than for oxen meat (r
2
cv
and RMSECV of 0.008
and 1.58 for a* value respectively; r
2
cv
of 0.345, RMSECV of 1.46
for b* value). It was claimed by the authors that this difference
in predictive performance arose from the wider measurement
ranges in young cattle than in adult steers. Later, Prieto et
al.
109
conducted another similar experiment using portable
NIR equipment with the goal of developing an on-line applica-
tion. Visible–NIR spectra (350 nm to 1800 nm) were collected
from animal carcasses of Aberdeen Angus cross-bred cattle
(194 heifers and steers; M. longissimus thoracis) and scanned
at quartering, 48 h post-mortem. Good prediction models
were obtained for a* value (r
2
cv
= 0.76; RMSECV = 0.95) and b*
value (r
2
cv
= 0.84; RMSECV = 0.90) without any spectral pre-
treatments; the authors claimed these results were good
enough for screening purposes.
Sensory evaluation
The most valued eating qualities of meat are tenderness,
juiciness and fl avour;
30
freshness is the most highly desired
property expected in fi sh products.
31
Sensory characteristics
are dictated by the presence of a range of physiochemical
compounds and their interactions in muscle tissue; thus, it
is a priori possible to use absorption spectra in the NIR or
visible–NIR regions to predict eating quality of muscle foods.
103
Reports of such studies by NIR spectroscopy are summarised
in Tables 7(a)–(d) In these reports, determination coeffi cients
ranged from 0.1 to 0.98 with r
2
cv
values lower than 0.70 being
observed in most cases; low prediction errors (0.4 to 1.18;
RMSECV or RMSEP), however, were often found.
Performance of NIR prediction in these applications was
noticeably dependent on the range of values for sensory scores
and their accuracy. Poor performance due to limited ranges in
sensory scores can be seen in Andrés et al.
71
and Prieto et
al.
107
; in both studies, r
2
cv
values only ranged between 0.1 and
0.5 with associated prediction errors (RMSECV) between 0.4
and 0.8. On the other hand, using homogenised Longissimus
thoracis muscle, Ripoll et al.
33
achieved a r
2
value of 0.98
(SEP = 0.35; RPD = 3.8) for beef tenderness over a range of
reference values of 2.0–7.2 but lower prediction accuracies in
measuring meat juiciness (r
2
= 0.6; SEP = 0.6; range of 2.7–6.6;
RPD = 1.54) and overall appraisal (r
2
= 0.6; SEP = 0.4; range
of 3.0–6.1; RPD = 1.58). As mentioned in previous sections,
models reported by these authors may be slightly over-fi tted;
however, variation in the reference values between these
three parameters differed from tenderness (32.9% CV), juici-
ness (20% CV) and overall appraisal (14.9% CV) and this may
be expected to play a part in the different levels of accuracy
obtained for these quality parameters. In 2010, Yancey et al.
91
obtained r
2
cv
values of 0.70 for both tenderness and overall
impression of intact Longissimus muscle in beef with associ-
ated RMSECV values of 0.4 and 0.6, in spite of a much narrower
range in reference sensory scores (5.99 to 6.56 for tenderness
and 6.37 to 6.94 for overall impression) than reported in Ripoll
et al.
33
An additional difference between the studies of Yancey
et al. and Ripoll et al. lay in the spectral range used; in the
former, this included the visible range, while in the latter it was
restricted to the 1108–2492 nm range. In a processed meat
product, García-Rey et al.
108
determined pastiness texture in
dry cured ham (intact Biceps femoris muscle) using visible–
NIR reflectance spectroscopy (400–2200 nm); good calibra-
tions (r
2
cv
= 0.74; RMSECV = 1.18; range of 1–10) were obtained.
Two other studies have been reported dealing with the devel-
opment of models to predict sensory characteristics of intact
lamb Longissimus thoracis muscle and both raw and cooked
broiler breast muscles.
98
Model predictive capabilities were
not signifi cantly different from those described above for beef.
All in all, these studies have demonstrated the potential of
NIR spectroscopy for development of prediction models for
sensory analysis of meats in a variety of forms and using a
number of different wavelength ranges. The chief limitation in
this subject area is the complexity of sensory evaluation based
on individual, human subjects as well as the normal variations
within samples.
Classifi cation and identifi cation
of quality traits in muscle foods
Many consumers place particular emphasis on non-
compositional aspects of muscle foods which are related to
quality; such aspects include the geographical origin of raw
material, production methods, rearing or feeding systems.
30
As a result, consumers are willing to pay a price premium for
such products and the potential for economic fraud therefore
exists. Adequate analytical methods are required by regulatory
authorities
109
for the effective control of such fraud. NIR spec-
troscopy has previously been applied to authenticity questions
in muscle foods and applications in this area have continued
to appear; recent applications in discriminant analysis of meat

28 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Study purpose
Sample description
Quality parameters
(n)
(Measurement ranges)
Measurement mode;
regression method
Wavelength
ranges
(nm)
Spectral
pre-treatment
Calibration performance Ref.
Sample
description
Sample
description
Sample
description
To evaluate the
potential of NIR
spectroscopy in
measuring sensory
quality of beef
Beef: SEUROP clas-
sifi cation consisting of
12 different breeds and
cross-breeds
Longissimus
thoracis
Homogenised Tenderness
(190a)
(2.0–7.2) R; PLS 1108–2492.8 SNV-D + 1
st
Deriv SEP = 0.353; r
2
= 0.981; RPD = 3.82 33
Juiciness
(190a)
(2.7–6.6) R; PLS 1108–2492.8 SNV-D + 1
st
Deriv SEP = 0.609; r
2
= 0.58 8; RPD = 1.5 4
Overall appraisal
(190a)
(3.0–6.1)
R; PLS 1108–2492.8 1
st
Deriv SEP = 0.44 3; r
2
= 0.589; RPD = 1.5 8
To predict sensory
characteristics of
beef using on-line
VIS-NIR spectros-
copy conducted in an
abattoir
Beef: Cross-bred sire
by Aberdeen Angus or
Limousin sire breeds
Longissimus
thoracis
Intact carcass
muscle
Tenderness
(173)
(3.00–6.70) R; PLS 350–1800 MSC + 1
st
Deriv RMSECV = 0.60; r
2
cv
= 0.16; RPD = 1.09
107
Juiciness
(174)
(3.80–5.90) R; PLS 350–1800 MSC + 1
st
Deriv RMSECV = 0. 41; r
2
cv
= 0.13; RPD = 1.07
Flavour
(181)
(2.57–5.70) R; PLS 350–1800 1
st
Deriv RMSECV = 0.42; r
2
cv
= 0.4 0; RPD = 1.2 8
Abnormal Flavour
(172)
(2.00–4.00)
R; PLS 350–1800 2
nd
Deriv RMSECV = 0.37; r
2
cv
= 0.13; RPD = 1.07
Overall liking
(178)
(3.10–
5.70)
R; PLS 350–1800 None
RMSECV = 0.38; r
2
cv
= 0.20; RPD = 1.12
To evaluate con-
sumer feedback
on eating quality of
different USDA beef
grades
Beef: From different
USDA grades—Select;
Low; Choice; Top Choice
and USDA Prime
Longissimus
muscle
Intact
Tenderness
(40a)
(5.99–6.56) R; PLS
400–2498
None RMSEP = 0.71; r
2
cv
= 0.4 0;RPD = 1.10; Robust = 1.03 91
2
nd
Deriv RMSEP = 0.57; r
2
cv
= 0.70;RPD = 1.3 8; Robust = 1.46
Overall impression
(40a)
(6.37–6.94)
R; PLS
400–2498
None RMSEP = 0.52; r
2
cv
= 0.33;RPD = 1.06; Robust = 1.06
2
nd
Deriv RMSEP = 0.39; r
2
cv
= 0.70;RPD = 1.4 0; Robust = 1.56
Table 7(a). Applications of near infrared spectroscopy in muscle food sensory quality—beef and beef products.
Study purpose
Sample description
Quality parameters
(n)
(Measurement ranges)
Measurement
mode; regression
method
Wavelength
ranges (nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal breeds/
variety
Selected
muscle
Sample
presentation
To quantify dry-cured
ham texture and colour
score
Dry cured ham:
Cross breeds of
Duroc, Landrace,
and Large white
Biceps femoris Intact
Pastiness texture
(103)
(1–10) R; PLS 400–2200 1
st
Deriv RMSECV = 1.18; r
2
cv
= 0.74
b
107
Colour
(107)
(4–10) R; PLS 400–2200 1
st
Deriv RMSECV = 1.11; r
2
cv
= 0.4 8
b
(n)
Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language.
Table 7(b). Application of near infrarared spectroscopy in muscle food sensory quality—pork and processed pork meat.

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 29
Study purpose
Sample description
Quality parameters
(n)
(Measurement ranges)
Measurement mode;
regression method
Wavelength
ranges (nm)
Spectral
pre-treatment
Calibration performance Ref.
Animal
breeds/ variety
Selected
muscle
Sample
presentation
To examine sensory
evaluation in lamb in corpo-
ration with physiochemical
attributes
Lamb: Texel; Scotish
blackface
Longissimus
thoracis
Intact
Texture
(232)
(2.5–7.0) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv
RMSECV = 0.846; r
2
cv
= 0.125
71
Juiciness
(232)
(3.3–6.3) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv
RMSECV = 0.44 0; r
2
cv
= 0.295
Flavour
(232)
(2.6–5.4) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv
RMSECV = 0.4 65 ; r
2
cv
= 0.271
Abnormal Flavour
(232)
(1.6–4.0) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv
RMSECV = 0.436; r
2
cv
= 0.04 8
Overall liking
(232)
(2.9–5.8) R; PLS 400–1900 SNV-D, MSC, 1
st
+ 2
nd
Deriv
RMSECV = 0.4 81; r
2
cv
= 0.243
Table 7(c). Application of near infrared spectroscopy in muscle food sensory quality—sheep meat products.
Study purpose
Sample description
Quality parameters
(n)
(Measurement ranges)
Measurement
mode and regres-
sion method
Wavelength
ranges (nm)
Spectral
pre-treatment
Calibration performance
Ref.
Animal
breeds/ variety
Selected
muscle
Sample
presentation
To examine sensory
attributes of broiler
meat from NIR spec-
tra of raw and cooked
meat
Mixed-gender
broiler meat
Breast muscle
Raw broiler
breasts
Flavour profi les
(96)
(0.07–6.44) R; PLS 400–1850 NA RMSEP = 0.45–1.20; r
2
= 0.05–0.62
98
Texture profi les
(96)
(1.20–9.30) R; PLS 400–1850 NA RMSEP = 0.79–2.04; r
2
= 0.15–0.63
Aftertaste characteristics
(96)
(0.00–6.00)
R; PLS 400–1850 NA RMSEP = 0.95–1.20; r
2
= 0.09–0.79
Cooked
broiler
breasts
Flavour profi les
(96)
(0.07–6.44) R; PLS 400–1850 NA RMSEP = 0.48 –1.10; r
2
= 0.12–0.77
Texture profi les
(96)
(1.20–9.30) R; PLS 400–1850 NA RMSEP = 0.72–1.87; r
2
= 0.08–0.83
Aftertaste characteristics
(96)
(0.00–6.00)
R PLS 400–1850 NA RMSEP = 0.97–1.20; r
2
= 0.14–0.60
(
n)
Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language;
NA: not available.
Table 7(d). Application of near infrared spectroscopy in muscle food sensory quality—poultry.

30 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
and fish quality are summarised in Tables 8(a)–(d). These
reports are considered below under the headings (a) trace-
ability of animal husbandry and feeding regimen, (b) classifi -
cation based on intrinsic quality parameters, (c) discrimina-
tion on the basis of eating quality and preparation processes
and (d) detection of adulteration and low-value ingredient
substitution.
Traceability of animal husbandry and feeding
regimen
In the prediction of the rearing history of livestock, animal
adipose tissues were normally utilised, although some studies
have used meat and fi sh fl esh; reported correct classifi cation
rates have ranged from 74% to 100%. Xiccato et al.
110
investi-
gated the classifi cation of sea bass fi llets from four different
rearing systems: extensive farming, semi-intensive farming,
intensive farming and sea-cage rearing. Using visible–NIR
refl ectance spectra and a class-modelling approach [soft inde-
pendent modelling of class analogy, (SIMCA)], correct classifi -
cation rates of only 37–65% were achieved. When the study was
extended to include the analysis of fi llets in freeze-dried form,
better results (74–83% correct classifi cation) were obtained.
In another study,
111
ground beef from premium oxen was 100%
correctly discriminated from similar material from commer-
cial, young cattle using partial least squares– discriminant
analysis (PLS-DA). This is an important application because
oxen require a long growth period to reach adulthood (up to
four years) and must be fi nished in an extensive rearing envi-
ronment to achieve the high degree of marbling and strong
taste which are sought after by consumers in European coun-
tries in particular.
111
Young cattle are raised for no more than
14 months, contain lower fat thickness and degree of marbling
and, hence, command a lower market price.
24
The ability to
discriminate between these two beef types after slaughter is
important for fair trading. In a similar application, Pla et al.
76
examined discrimination between ground hind leg meat from
rabbits raised on conventional and organic farms. Correct
classifi cation rates of up to 98% were achieved using PLS-DA,
albeit in a study using low sample numbers (n = 26).
Classifi cation based on intrinsic quality
factors
Intrinsic quality factors of muscle foods involve intrinsic char-
acteristics of animals, for example, species, breed, genotype
and muscle type or cuts. Meat derived from different species,
cuts or grades can be more valuable to certain groups of
consumers than others;
54,60
it is therefore a matter of some
importance to develop classification methods to assist in
meat and fi sh quality control. Classifi cation of pig meat from
different breeds—Iberian and Durocis is one such example.
The Iberian pig (Sus mediterraneus) has a gene potentially
giving rise to an animal with a high amount of adipose tissue.
They are naturally reared in La Dehesa and undergo a fi nal
fattening in an extensive pasture system rich in acorn – the
so-called ‘montanera’ in Spanish. Increasing demand for dry-
cured products from Iberian pigs has resulted in a strategy to
cross-breed Iberian and Duroc pig genotypes to achieve more
piglets per sow with higher carcass weight and hence greater
fatness. However, the cross-bred pork is still leaner, with less
intramuscular fat, poorer colour, odour, texture, a saltier taste
and lower product stability due to a higher content of polyun-
saturated fatty acids than the pure Iberian breed.
121
To sustain
Iberian characteristics, Spanish legislation (BOE, 2001) only
permits Iberian–Duroc cross-breeding if the maternal line
is pure Iberian.
115
Using visible–NIR spectroscopy for the
authentication of Iberian products assured by quality marks
has been reported in two recent studies [Table 8(b)]—del
Moral et al.
115
and Guillen et al.
116
—using samples of intact
meat. Classifi cation models from both studies were developed
using non-linear mathematical techniques, artifi cial neural
networks (ANN) and support vector machines (SVM). In del
Moral et al.,
98
Iberian pork was discriminated from Duroc
at correct classifi cation rates of 96.6% to 100% using three
wavelengths (variables) which were selected by mutual infor-
mation ranking (MI) as inputs to a radial basis function neural
network (RBFNN). Meanwhile, Guillen et al.
117
performed a
similar study in which the RBFNN and SVM methods were
compared; correct classifi cation rates of between 98.9% and
99.8% were achieved, with both mathematical techniques
producing similar results.
Classifi cation based on eating quality and
preparation processes
Reported applications of NIR spectroscopy in classifying eating
quality traits have focused on texture and appearance of muscle
foods [Table 8(c)]; discrimination of fresh muscle foods from
frozen-then-thawed fl esh is an example of a preparation process.
Eating quality classes will be discussed fi rst in this section.
Visible–NIR spectroscopy (400–1850 nm) for classifying intact
chicken breast meat on the basis of tenderness was studied by
Liu et al.
98
Using a Warner–Bratzler shear value cut-off of 7.5 kg
to separate tender from tough broiler meat, a correct classifi ca-
tion rate of 82% for tender meat and 57.1% for tough meat was
produced by a SIMCA model; PLS models of the same dataset
produced correct classifi cation rates of 76.5% and 71.4%. When
tenderness and toughness categories for the same sample
collection were defi ned differently (tender WBSF ≤ 6.5 kg; tough
WBSF > 9.0 kg) (83%) the classifi cation performance declined,
irrespective of modelling technique. This illustrates a diffi-
culty which arises in modelling measurements which are, to
a large extent, empirical. In another case involving spectral
collection from intact slices of a processed meat (dry cured
ham; Biceps femoris muscle), a k-nearest neighbours (KNN)
method was applied to visible–NIR spectral data (400–2200 nm)
for the purpose of classifying texture (pastiness and normal)
and colour (defective and normal). With regard to texture, ham
samples of normal and pasty texture were 93.3% and 79.0%
correctly classifi ed respectively; for segregation on the basis of
colour, normal and defective hams were correctly classifi ed at
84.4% and 73.5% rates.
108
Intentional sale of fl esh from frozen-then-thawed meat or
fi sh as fresh material is a form of fraud and there may also be

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 31
Study purposes
Sample description
Classifi cation
(n)
Measurement mode;
regression method
Wavelength
ranges (nm)
Spectra
pre-treatment
% Correct
classifi cation
(Performance)
Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To identify rearing
system of European sea bass
Sea bass
(Dicentrarchus labrax
L.): caught in 4 rear-
ing systems
Fillets
Fresh minced
fi llets
Extensive ponds
(48a)
R; SIMCA 1100–2500 2
nd
Deriv 65%
110
Semi-intensive ponds
(56a)
R; SIMCA 1100–2500 2
nd
Deriv 58%
Intensive tanks
(27a)
R; SIMCA 1100–2500 2
nd
Deriv 37%
Sea-cages
(70a)
R; SIMCA 1100–2500 2
nd
Deriv 45%
Freeze-dried
minced fi llets
Extensive ponds
(48a)
R; SIMCA 1100–2500 2
nd
Deriv 83%
Semi-intensive ponds
(56a)
R; SIMCA 1100–2500 2
nd
Deriv 80%
Intensive tanks
(27a)
R; SIMCA 1100–2500 2
nd
Deriv 74%
Sea-cages
(70a)
R; SIMCA 1100–2500 2
nd
Deriv 83%
To classify chicken enriched by
supplemented omega-3 fatty acid
from different diets
Chicken breast from
laying hens fed four
different diets
Breast meat:
(pectoralis
superfi cialis)
Ground freeze-
dried samples
Control diet
(18)
R; PLS-DA 1100 –2498 SN V-D + smoothing 94.4%
75
n-3 of marine origin
(18)
R; PL S-DA 1100–2498 SN V-D + smoothing 8 8.89%
Ground linseed
(18)
R; PLS-DA 1100–2498 SNV-D + smoothing 66.67%
Extruded linseed
(18)
R; PLS-DA 1100 –2498 SN V-D + smoothing 8 3.33%
To classify suckling lamb fed with
different milk diets
Lamb: Churra-breed
suckling lambs
Perirenal fat Extracted liquid fat
Lamb fed with natural milk
(50)
Transfl ectance; PLS-DA 1100–2500 2
nd
Deriv 100%
112
Lamb fed with milk replacer
(48)
Transfl ectance; PLS-DA 1100–2500 2
nd
Deriv 100%
To classify rabbit meat raised
from conventional and organic
farms
Rabbit meat Hind legs Ground meat Different types of farming systems:
Conventional feeding system
(26)
;
Organic system
(26)
R; PLS-DA 1100 –2498 SN V-D + 1
st
Deriv 98% 76
To classify sheep meat from dif-
ferent feeding systems
Lamb Perirenal fat Intact Pasture-fed
(120)
; Stall-fed
(139)
R; PLS-DA 400–2500 SNV-D + 1
st
Deriv 97.7%
113
R; PLS-DA 400–700 None 94.9%
To classify adult steer beef from
commercial young cattle beef
Beef: Adult steers;
Young cattle
Longissimus
thoracis
Homogenised Adult steers
(53)
; Young cattle
(67)
R; PLS-DA 1100–2500 2
nd
Deriv 100% 111
To classify pork meat from differ-
ent feeding systems
Pork: Ilberian Subcutaneous
fat
Melted fat
High oleic-formulated feed
(12)
T; LDA 1100–2500 Mean centred 100%
1114
Standard feed
(24)
T; LDA 1100–2500 Mean centred 83.3%
Montanera
(22)
T; LDA 1100–2500 Mean centred 100%
Table 8(a). Application of near infrared spectroscopy in qualitative analysis of muscle food—traceability of animal husbandry and feeding regimen.

32 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Study purposes
Sample description
Classifi cation
(n)
Measurement
mode; regression
method
Wavelength
ranges (nm)
Spectral
pre-treatment
% Correct classifi ca-
tion (performance)
Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To classify chicken
carcasses from different
genotypes
Chicken: Fast
growing broiler;
Laying hen; Slow
growing broiler
Whole carcass
excluding gut
contents
Dry, ground chicken
carcasses
Different genotypes: Fast
growing broiler
(50)
; Laying hen
(144)
; Slow growing broiler
(64)
R; PLS-DA 40 0–2498 MSC + 2
nd
Deriv 96.5% 74
To classify pork meat
based on breeds
Pork: Duroc; Iberian M. masseter Intact Two swine breeds: Duroc pork
(15)
; Iberian pork
(15)
R; BFNN 350–2500
1 wavelength input selected with
MI ranking
70-90%
115
1 wavelength input selected with
VIS/NIR characterisation
90–93.33%
3 wavelength input selected with
MI ranking
96.60-100%
3 wavelength input selected with
VIS/NIR characterisation
93.33–96.67%
10 wavelength input selected with
MI ranking
76.67–93.30%
10 wavelength input selected with
VIS/NIR characterisation
73.33–93.33%
To classify Iberian pork
from White pork
Pork: Ilberian pork,
White pork
Masseter Intact Ilberian pork; White pork R; RBFNN 350–2500 1 variable input 99.77% 116
R; SVM 350–2500 1 variable input 98.89%
R; RBFNN 350–2500 2 variable input 99.77%
R; SVM 350–2500 2 variable input 99.61%
R; RBFNN 350–2500 3 variable input 99.61%
R; SVM 350–2500 3 variable input 99.77%
(n)
Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language.
Table 8(b). Application of near infrared spectroscopy in qualitative analysis of muscle foods—instrinsic quality factors and quality attributes.

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 33
Study purposes
Sample description
Classifi cation
(n)
Measurement mode;
regression method
Wavelength
ranges (nm)
Spectral
pre-treatment
% Correct classifi cation
performance)
Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To classify tender and tough
broiler meat based on WBSF
values
Broiler: Mixed-
gender of broiler
meat
Breasts Intact
Tender (WBSF < 7.5 kg)
(66)
R; PLS 400–1850 Mean centring and MSC 76.5%
98
R; SIMCA 400–1850 Mean centring and MSC 82.4%
Tough (WBSF > 7.5 kg)
(30)
R; PLS 400–1850 Mean centring and MSC 71.4%
R; SIMCA 400–1850 Mean centring and MSC 57.1%
Tender (WBSF < 6.5 kg)
(56)
R; PLS 400–1850 Mean centring and MSC 63.3%
R; SIMCA 400–1850 Mean centring and MSC 83.3%
Tough (WBSF > 9.0 kg)
(18)
R; PLS 400–1850 Mean centring and MSC 11.1%
R; SIMCA 400–1850 Mean centring and MSC 44.4%
To classify texture and colour
defects in dry cured ham
Dry cured ham: Cross
breeds of Duroc,
Landrace and Large
White
Biceps femoris Intact slices
Pastiness texture
(38)
R; KNN 400–2200 1
st
Deriv 79.0%
108
Normal texture
(75)
R; KNN 400–2200 1
st
Deriv 93.3%
Defective colour
(49)
R; KNN 400–2200 1
st
Deriv 73.5%
Normal colour
(64)
R; KNN 400–2200 1
st
Deriv 84.4%
To discriminate between
fresh and frozen-thawed Red
sea bream fi sh
Red sea bream
(Pagrus major)
A location behind
the dorsal fi n;
midway on the
epaxial part
Intact
Fresh fi sh
(54)
I; SIMCA 400–1100 None 63%
117
I; LDA 400–1100 None 100%
I; SIMCA 400–1100 MSC 63%
I; LDA 400–1100 MSC 79%
Frozen-thawed fi sh
(54)
I; SIMCA 400–1100 None 84%
I; LDA 400–1100 None 100%
I; SIMCA 400–1100 MSC 84%
I; LDA 400–1100 MSC 84%
To discriminate fresh from
frozen-thawed sole**
Sole (Solea
vulgaris)
NA Minced fresh
muscle
Fresh sole
(106)
R; PLS 1100–2500 SNV-D 98%
79
R; SVM 1100–2500 SNV-D 93%
Frozen sole
(35)
R; PLS 1100–2500 SNV-D 80%
R; SVM 1100–2500 SNV-D 83%
(n) Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references; f Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language;
NA: not available.
Table 8(c). Applications of near infrared spectroscopy in qualitative analysis of muscle foods—eating quality and preparation processes.

34 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
Study purposes
Sample description
Classifi cation
(n)
Measurement
mode; regression
method
Wavelength
ranges (nm)
Spectral
pre-treatment
% Correct classifi cation
(performance)
Ref.
Animal breeds/
variety
Selected
muscles
Sample pres-
entation
To discriminate pure
Iberian sausages from
mixture group and quan-
tify levels of mixtures
Dry-cured sau-
sages: Ilberian
(I); Standard
Landrace (S)
NA
Homogenised
fresh meat
(n = 60)
100% I; Mixture; 100% S R; PLS-DA 400–1700 SNV-D, MSC,
1
st
+ 2
nd
Deriv
87%–98 % (calibration);
60% (validation)
118
100I; 75I/25S; 50I/50S; 25I/75S;
100S
R; PLS 400–1700 SNV-D, MSC,
1
st
+ 2
nd
Deriv
RMSECV = 4.7 %; r
2
cv
= 0.98
Homogenised
dry-cured
sausages
(n = 60)
100%I; Mixture; 100% S R; PLS-DA 400–1700 SNV-D, MSC,
1
st
+ 2
nd
Deriv
80%–92% (calibration)
60%–80% (validation)
100I; 75I/25S; 50I/50S; 25I/75S;
100S
R; PLS 400–1700 SNV-D, MSC,
1
st
+ 2
nd
Deriv
RMSECV = 5.9 %; r
2
cv
= 0.99
To classify beef
adulterated with spinal
cord
Beef NA Ground beef Added spinal cord (> 21ppm) R; PLS 1000–1851.85 2
nd
Deriv 87% 119
To quantify
percentages
of crabmeat adulterated
with other species
Canned crab
meat
Atlantic blue
(Callinectes
sapidus);
Blue
swimmer
(Portunus
pelagicus)
Homogenised 100% Atlantic blue; 100% Blue
swimmer; 10 increment of
percentage mixture of the two
(10-90%)
R; PLS
400–2500 None SEP = 5.8 8 %; r
2
= 0.983
120
400–1400 None SEP = 5.02 %; r
2
= 0.988
400–2500 1
st
Deriv SEP = 5.6 4 %; r
2
= 0.984
400–2500 2
nd
Deriv SEP = 7.16 %; r
2
= 0.975
400–1700 2
nd
Deriv SEP = 5.17 %; r
2
= 0.987
400–2500 64-point Combing SEP = 6.94 %; r
2
= 0.979
400–1400 64-point Combing SEP = 4.90 %; r
2
= 0.988
(n)
Number of samples in calibration set;
a
The total number of samples if (n) was not reported in original reference;
b
calculated from correlation coeffi cient (r) reported in original references;
f
Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language;
NA: not available
Table 8(d). Application of near infrared spectroscopy analysis in qualitative analysis of muscle foods—adulteration and low-valued ingredient substitution.

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 35
a food safety implication. Fish, in particular, is a highly perish-
able food and, while fresh fi sh is highly prized and carries a price
premium, it is often necessary to freeze fi sh to prolong shelf-
life and reach retail markets some distance from the sea or
freshwater source. Frozen fi sh therefore generally commands
a lower price than fresh fi sh.
87,117
Two investigations into the
use of NIR spectroscopy for discrimination between fresh and
frozen-then-thawed fish have been published recently
79,117
[Table 8(c)]. In one involving red sea bream (Pagrus major;
n = 108), visible–NIR interactance spectra (400–1100 nm) were
collected from intact fi sh at a location behind the dorsal fi n
using a fi bre-optic accessory and processed using SIMCA and
LDA (linear discriminant analysis) chemometric techniques.
Best results were obtained using raw spectral data with LDA,
producing a 100% correct classifi cation rate for both fresh
and frozen-then-thawed samples. These authors explained
the better performance of raw over scatter-corrected data on
the basis that freeze–thawing induces structural changes in
fi sh fl esh which may be detected in raw spectra but which are
removed during scatter-correction. Fasolato et al.
79
collected
refl ectance spectra (1100–2500 nm) from minced samples of
fresh and frozen-then-thawed sole (Solea vulgaris); correct
classifi cation rates for fresh sole of 98% (PLS) and 93% (SVM)
were reported by these authors while corresponding results
for frozen sole were 80% (PLS) and 83% (SVM).
Inter-species adulteration and ingredient
substitution
Meat is not only purchased as a raw ingredient for meal
preparation but is also a component of ready-to-eat meals
in various forms—ground meat, patties or dices—and may be
present in either raw or cooked forms. Substitution of meat of
a declared species by meat from another or the substitution of
a high-value meat by a cheaper variety have been reported.
122
Recent NIR applications addressing these issues are summa-
rised in Table 8(d). The possibility of detecting Iberian dry-
cured pork sausage mixed with meat from another pork breed
(Standard Landrace) was studied by Ortiz-Somovilla et al.;
118
this study addressed mixture detection in both fresh meat and
fi nished cured products and involved the analysis of samples
comprising 100% Iberian pork, 75% Iberian–25% Standard
Landrace, 50% of each, 25% Iberian–75% Standard Landrace
and 100% Standard Landrace; a total of 375 samples of fresh
and 375 of cured sausages were scanned on a Perten DA7000
instrument between 400 nm and 1700 nm. Best classifi cation
accuracies of 98.3% and 91.7% for fresh and cured mixtures,
respectively, were obtained but prediction accuracies were
much lower at 60% and 80%, respectively. The main errors
arose from mis-classifi cation of pure samples as mixtures in
both cases. These authors also attempted to quantify levels of
admixture using PLS regression. In the case of the fresh meat
samples, a model with r
2
cv
= 0.98 and RMSECV = 4.7% w/w was
reported while for cured sausages, corresponding values of
0.99% and 5.9% w/w were obtained. One note of caution with
regard to this study is that all of the models reported involved
quite large numbers of PLS loadings, for example, 11 to 13.
In an attempt to develop a method to detect meat with added
spinal cord material, Gangidi et al.
119
reported a classifi cation
model which, it is claimed, could correctly detect incorpora-
tion of more than 21 parts per million (ppm) of spinal cord
in ground beef in 87% of samples using spectral data in the
range 1000–1851.85 nm by PLS regression.
In a single fi sh study, Gayo and Hale
120
studied the devel-
opment of models to detect and quantify the adulteration of
Atlantic blue (Callinectes sapidus) crabmeat which is native to
the USA by other imported crab species, i.e. Blue swimmer
(Portunus pelagicus). Atlantic blue crabmeat was adulter-
ated with blue swimmer crabmeat in 10% increments. In a
thorough study which investigated a number of wavelength
ranges and data pre-treatments, the most accurate model
for predicting the percentage of adulterated crabmeat in a
mixture involved three PLS loadings and produced values of
r
2
and RMSECV on a set of prediction samples equal to 0.99%
and 5.2% w/w, respectively. These authors also performed a
PCA on the sample collection and were able to demonstrate
the relationship between PC1 and adulterant content; they did
not, however, report a discriminant or classifi cation model to
detect adulteration.
Food safety, shelf-life and
process monitoring
Food safety is of critical importance in the food chain in order
to prevent foodborne illness and outbreaks as a result of
microbiological growth or contamination. Pathogens found
in meat and meat products which have public health signifi -
cance include Salmonella, Campylobacter and some signifi cant
strains of Escherichia coli such as E. coli O157:H7; in recent
years, the emergence of bovine spongiform encephalopathy
(BSE) in cattle meat
24
has also been a major concern. Fishery
Figure 2. Average refl ectance spectra of fi sh gels prepared
from Walleye Pollack surimi. (Reprinted with permission from
Food Control. 17, 660 (2006). M. Uddin et al.
125
)

36 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
products may be subject to growth or contamination by micro-
organisms such as Vibrio, Salmonella and also by parasitic
23
infestation. In combination with good personal and operational
hygiene, heat treatment is one conventional and effective way
to eliminate harmful organisms but exposure to high temper-
atures may destroy food palatability and wholesomeness; it is
thus necessary to establish effective control of heat treatment
processes (temperature and time).
123
NIR spectroscopy has
been examined for its potential in determining previous heat
treatment and freshness indicators of muscle foods (Table 9).
First, end-point temperature (EPT) is used as a criterion
for deciding whether a food product has received adequate
heat treatment. Uddin et al.
123
described a study designed to
estimate the end-point temperature in intact kamabogo gel,
a fi sh-based food produced from Walleye Pollack (Theragra
chalcogramma). Using a range of experimental end-point
temperatures between 30°C and 90°C in 10°C increments and
various scanning ranges, good prediction accuracy (SEP = 2.09,
r
2
= 0.96) was achieved using spectral data in the wavelength
range 1300–1600 nm. The authors attributed the success of
the model to the ability of NIR spectroscopy to detect changes
in the secondary structure of protein in the gels; a number of
wavelength ranges and mathematical techniques (PLS and
MLR) produced very similar prediction accuracies. In another
similar study,
125
fi sh-meat gel samples from Walleye Pollack
(Theragra chalcogramma) and Horse mackerel (Trachurus
japonicus) were studied using transmission spectroscopy; an
example of the spectral variation found is shown in Figure 2
and the absorbance changes have been attributed to changes
in protein secondary structure and states of water.
123,125,128
The degree of freshness of a food product can be estimated
by various indicators such as total viable count of micro-
organisms and the presence of certain volatile compounds.
In an attempt to predict total viable count on Rainbow trout,
Lin et al.
124
devised a study in which three sample types
were examined (intact fl esh stored at 4°C, intact skin stored
at 4°C and minced, whole fish held at 21°C) using visible–
shortwave NIR refl ectance spectra (600–1100 nm). Prediction
models were developed by PLS regression and the highest
prediction accuracy [r
2
= 0.94, SEP = 0.38 log colony-forming
units (CFU) g
−1
] was observed for the model developed for
intact fl esh. The poorest performing model was obtained for
minced fi sh stored at 21°C (r
2
= 0.67, SEP = 0.82, RPD = 1.75). In
another study involving pork meat (tenderloin samples from
Taiwan black-hair hogs), Chou et al.
126
aimed to quantitatively
predict pH, volatile basic nitrogen (VBN), aerobic plate count
(APC) using visible–NIR spectra (600–1368 nm) from aqueous
extracts of pig meat; they reported that four wavelengths
(608 nm, 624 nm, 656 nm and 732 nm) were selected for fresh-
ness evaluation based on a partial least squares regression
(PLSR) model using different freshness criteria for dummy
regression. Accuracies of freshness evaluation reached 90%
with the PLSR models by excluding the data in day 3; given
that the original manuscript is in Chinese, it is difficult to
fully understand the study. Finally, a study by Stawczyk et
al.
127
aimed to develop a model to predict water activity (A
w
)
in dry pork sausages using a remote Fourier transform–NIR
sensor to monitor and control associated relative humidity
(RH) during the drying of pork sausages (made from 70%
shoulder lean and 30% bellies) to prevent the development of
a hard and crusty surface. Using a fi bre-optic accessory for
spectral collection, partial least squares (PLS) regression was
used to predict chemical and technological properties. The
reported a
w
prediction error of calibration (RMSEP) was 0.007
and the corresponding residual prediction deviation (RPD) was
7.96. These authors concluded that a control system based on
NIR input data was able to properly modify the drying condi-
tions of the pork sausage drying process studied.
Conclusions and future
challenges
Near infrared spectroscopic applications in the assessment of
muscle food quality reported in the last fi ve years have devel-
oped in directions that serve modern industrial and consumer
needs and trends.
NIR spectroscopic quantifi cation of quality indicators and
attributes in meat and fi sh can be categorised into chemical,
physical and sensory traits and applications exist in process
and post-process monitoring. With regard to qualitative appli-
cations, most reported studies have focused on confi rmation
of feeding regimen, housing systems and animal breed iden-
tifi cation. For instance, dairy cows can produce more body fat
in a shorter time period than beef breeds; to avoid substitu-
tion of prime beef by that from dairy cows, qualitative NIR
spectroscopy can be deployed. Besides, NIR devices allow
non-invasive (no contact between the device and food sample)
implementations which are advantageous from the point-of-
view of hygiene control in the food industry.
Despite the many advantages of NIR technology, the food
industry has demonstrated a degree of reluctance in the
adoption and application of this technology. Reports of
unstable calibration models and the lack of proper industrial
validation are possible reasons for this poor uptake; however,
external validation sets have been included in many of the
studies cited in this review paper to a greater extent than
may have been the case previously. Being dependent on
reference analytical methods is still a diffi culty for NIR spec-
troscopy, especially with regard to the empirical nature of
many of the measurement methods used by the food industry
in particular.
One recent significant development in NIR-based tech-
nology has been the emergence of hyperspectral image
analysis. In hyperspectral imaging, the combination of a
camera, spectral detector and moving sample bench results
in the collection of both pixel-based (spatial) and spectral-
based information; this technology is eminently suitable for
monitoring food products on moving conveyor belts or in
continuous processes for determining the physical distri-
bution of components or for the detection of defects. The

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 37
Study purposes
Sample description
Classifi cation
(n)
Measurement mode;
regression method
Wavelength
ranges (nm)
Spectral
pre-treatment
% Correct classifi cation/
performance
Ref.
Animal breeds/
variety
Selected
muscles
Sample
presentation
To estimate processing
end-point temperature of
kamabogo gel—fi sh-based
product
Kamabogo gel from:
Walleye Pollack
fi sh (Theragra
chalcogramma)
NA Intact Endpoint of previous cooking tempera-
ture
(70)
from 30 to 90°C
R; MLR 1350–1580 SNV-D
+
+ 2
nd
Deriv
SEP = 2.16°C; R
2
= 0.96
b
123
R; PL S 1100–2500 SN V-D + 2
nd
Deriv
SEP = 2.21°C; r
2
= 0.96
b
;
R; PLS 1300 –1600 SN V-D + 2
nd
Deriv
SEP = 2.09°C; R
2
= 0.96
b
To detect the onset of
spoilage caused by
psychrotrophic Gram-
negative organisms and to
quantify microbial loads in
rainbow trout
Rainbow trout;
(Oncorhynchus mykiss)
Flesh; Skin
Intact fl esh stored
at 4°C
Total Viable Count 3.90–7.85 log CFU
g
-1
R; PLS 600–1100 Binning, smoothing,
2
nd
Deriv
SEP = 0.38 log CFU g
-1
R
2
= 0.94
b
; RPD = 4.14
124
Intact skin at 4° C
Total Viable Count 3.90–7.85 log CFU
g
-1
R; PLS 600–1100 Binning, smoothing,
2
nd
Deriv
SEP = 0.53 log CFU g
-1
R
2
= 0.88
b
; RPD = 2.99
Minced stored at
21°C
Total Viable Count 3.00–7.46 log CFU
g
-1
R; PLS 600–1100 Binning, smoothing,
2
nd
Deriv
SEP = 0.82 log CFU g
-1
R
2
= 0.67
b
; RPD = 1.75
To estimate previous
heating temperature of
fi sh-meat gel—fi sh-based
product
Fish-meat gel: Walleye
Pollack fi sh (Theragra
chalcogramma)
NA Intact Previous cooking temperature
(112a)
from 30 to 90°C
T; MLR 900 –95 0 SNV-D + 2
nd
Deriv
SEP = 1.71° C
R
2
= 0.96
b
; RPD = 5.32
125
T; PLS 650–1100 SNV-D + 2
nd
Deriv
SEP = 1.87 °C
R
2
= 0.96
b
; RPD = 5.63
T; PL S 890–950 SN V-D + 2
nd
Deriv
SEP = 1.76°C
R
2
= 0.94
b
; RPD = 6.07
Horse mackerel
(Trachurus
japonicus)
NA Intact Previous cooking temperature
(112a)
from 30 to 90°C
T; MLR 900 –95 0 SNV-D + 2
nd
Deriv
SEP = 1.84° C
R
2
= 0.96
b
; RPD = 5.09
T; PLS 650–1100 SNV-D + 2
nd
Deriv
SEP = 1.94°C
R
2
= 0.96
b
; RPD = 4.25
T; PL S 890–950 SN V-D + 2
nd
Deriv
SEP = 1.81°C
R
2
= 0.96
b
; RPD = 5.49
To estimate freshness indi-
cators of ground pork and
classify fresh samples from
spoiled group
Pork: Taiwan black-
hair hogs
Tenderloin Liquid extracts
varied from 36hrs
storage to 5 days
under 7°C
VBN
(120a)
(9.80–53.20 mg 100g
-1
) T; PLS 600–1368 and
1100–1368
1
st
and 2
nd
difference
spectra
SEP = 3.60 mg 100g
-1
r
2
= 0.76
b
126
APC
(120a)
(3.00–10.20 log CFU g
-1
) T; PLS
600–1368 and
1100–1368
1
st
and 2
nd
difference
spectra
SEP = 5.19 log CFU g
-1
r
2
= 0.74
b
Fresh meat; Spoiled meat T; PLS-DA 600–840 NA 70.2%–96%
To monitor A
w
in relation to
relative humidity control of
drying sausage
Dry pork sausages NA NA Water activity (A
w
) R; PLS NA NA RMSEP = 0.007; RPD = 7.96
R
2
= not repor ted
127
(
n
) Number of samples in calibration set; a The total number of samples if (n) was not reported in original reference; b calculated from correlation coeffi cient (r) reported in original references; f Mean ± SD; Deriv = Derivative(s); ** The original article was written in another language;
NA: not available
Table 9. Applications of near infrared spectroscopy analysis in muscle foods—quality identifi cation for food safety, shelf-life and process monitoring.

38 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
application of hyperspectral imaging to food quality analysis
and control has recently been reviewed.
129
A study on turkey
hams of varying quality showed that they can be sorted into
quality classes using only eight out of 241 wavelengths.
130
The recent emergence of hand-held or portable NIR spec-
trophotometers opens the way for the effi cient collection of
high-quality spectral data in the fi eld, for example, from live
animals or carcasses in abattoirs. This might be useful for
making modifi cations to feed formulations, for example, to
adjust certain nutrient contents in order to produce better
quality farmed-fi sh more effi ciently.
References
1. A.W. Speedy, “Global production and consumption of
animal source foods”, J. Nutr. 133, 4048S (2003).
2. Food and Agriculture Organization, FAO Statistical
Yearbook 2009. Available at:
http://www.fao.org/eco-
nomic/ess/publications-studies/statistical-yearbook/
fao-statistical-yearbook-2009/d-consumption/en/
3. Food and Agriculture Organization, FAO Statistical
Yearbook 2007–2008. Available at:
http://www.fao.org/
economic/ess/publications-studies/statistical-year-
book/fao-statistical-yearbook-2007-2008/b-agricul-
tural-production/en/
4. G. Downey and K.I. Hildrum, “Analysis of meats”, in
Near-infrared Spectroscopy in Agriculture, Agronomy
Monograph no. 44, Ed by C.A. Roberts, J. Workman, Jr
and J.B. Reeves, III. ASA, CSSA, SSSA, Madison, WI,
USA, pp. 599–634 (2004).
5 United States Department of Agriculture, Livestock and
Poultry: World Markets and Trade. Foreign Agricultural
Service/USDA (2009).
6. United States Department of Agriculture, Livestock and
Poultry: World Markets and Trade. Foreign Agricultural
Service/USDA (2009).
7. R.J. Polkinghorne and J.J. Thompson, “Meat standards
and grading. A world view”, Meat Sci. 86, 15 (2010).
8. T. Becker, “Defi ning meat quality”, in Meat Processing:
Improving Quality, Ed by J. Kerry, J. Kerry and D.
Ledward. Woodhead Publishing, Cambridge, UK, p. 3
(2002).
9. N. Aldai, A.I. Nájera, A. Martínez, R. Celaya and K.
Osoro, “Correlation between carcass conformation
and fat cover degree, and muscle fatty acid profi le
of yearling bulls depending on breed and mh-gen-
otype”, Livestock Sci. 107, 199 (2007). doi:
10.1016/j.
livsci.2006.09.014
10. T. Woodcock, G. Downey and C.P. O’Donnell, “Review:
Better quality food and beverages: the role of near
infrared spectroscopy”, J. Near Infrared Spectrosc. 16, 1
(2008). doi:
10.1255/jnirs.758
11. R.S. Singhal, P.R. Kulkarni and D.V. Rege, Handbook
of Indices of Food Quality and Authenticity. Woodhouse
Publishing, Cambridge, UK (1997).
12. R. Bro, F. van den Berg, A. Thybo, C.M. Andersen, B.M.
Jørgensen and H. Andersen, “Multivariate data analysis
as a tool in advanced quality monitoring in the food pro-
duction chain”, Trends Food Sci. Tech. 13, 235 (2002). doi:
10.1016/s0924-2244(02)00138-3
13. T. Isaksson and V.H. Segtnan, “Meat and fi sh products”,
in Near-Infrared Spectroscopy in Food Science and
Technology, Ed by Y. Ozaki, W.F. McClure and A.A.
Christy. John Wiley & Sons, Hoboken, New Jersey, USA,
p. 247 (2007).
14. B. Leroy, S. Lambotte, O. Dotreppe, H. Lecocq, L.
Istasse and A. Clinquart, “Prediction of technological
and organoleptic properties of beef Longissimus thora-
cis from near-infrared refl ectance and transmission
spectra”, Meat Sci. 66, 45 (2004). doi:
10.1016/s0309-
1740(03)00002-0
15. R.S. Singhal, P.R. Kulkarni and D.V. Rege, “Meat, fi sh
and poultry”, in Handbook of Indices of Food Quality and
Authenticity. Woodhead Publishing, Cambridge, UK, pp.
209–299 (1997).
16. M. Prevolnik, M. Candek-Potokar and D. Skorjanc,
“Ability of NIR spectroscopy to predict meat chemical
composition and quality—a review”, Czech J. Anim. Sci.
49, 500 (2004).
17. Y.J. Ru and P.C. Glatz, “Application of near infrared
spectroscopy (NIR) for monitoring the quality of milk,
cheese, meat and fi sh”, Asian–Aust. J. Anim. Sci. 13, 1017
(2000).
18. T. Isaksson and V.H. Segtnan, “Meat and fi sh prod-
ucts”, in Near-Infrared Spectroscopy in Food Science and
Technology, Ed by Y. Ozaki, W.F. McClure and A. Christy.
John Wiley & Sons, Hoboken, NJ, USA, pp. 247–277
(2006).
19. A.J. McAfee, E.M. McSorley, G.J. Cuskelly, B.W. Moss,
J.M.W. Wallace, M.P. Bonham and A.M. Fearon, “Red
meat consumption: An overview of the risks and
benefi ts”, Meat Sci. 84, 1 (2010). doi:
10.1016/j.meat-
sci.2009.08.029
20. Food and Agriculture Organization, FAO Yearbook
2
007. Food and Agriculture Organization of the United
Nations, Rome, Italy (2009).
21. D. Cozzolino, I. Murray and J.R. Scaife, “Near infrared
refl ectance spectroscopy in the prediction of chemical
characteristics of minced raw fi sh”, Aquaculture Nutri. 8,
1 (2002). doi:
10.1046/j.1365-2095.2002.00176.x
22. P. Fellows and A. Hampton, Small-Scale Food Processing
– A Guide for Appropriate Equipment. Intermediate
Technology Publications, London, UK (1992).
23. V. Venugopal, “Biosensors in fi sh production and qual-
ity control”, Biosens. Bioelectron. 17, 147 (2002). doi:
10.1016/S0956-5663(01)00180-4
24. P.D. Warriss, Meat Science: An Introduction Text. CABI,
Wallingford, UK (2010).
25. G. Shim-Prydon and H. Camacho-Barreto, New Animal
Products: New Uses and Markets for By-products and
Coproducts of Crocodile, Emu, Goat, Kangaroo and

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 39
Rabbit. Rural Industries Research and Development
Corporation, Australian Government (2007).
26. Food and Agriculture Organization, Utilization of
Tropical Foods: Animal Products. Food and Agriculture
Organization of the United Nations, Rome, Italy (1990).
27. DAFF, Intensive Livestock and Game (2009). Available at:
http://www.daff.gov.au/agriculture-food/meat-wool-
dairy/ilg
28. C. Maltin, D. Balcerzak, R. Tilley and M. Delday,
“Determinants of meat quality: tenderness”, Proc. Nutri.
Soc. 62, 337 (2003). doi:
10.1079/PNS2003248
29. R.A. Mancini and M.C. Hunt, “Current research in meat
color”, Meat Sci. 71, 100 (2005). doi:
10.1016/j.meat-
sci.2005.03.003
30. M.D. Aaslyng, “Trends in meat consumption and the
need for fresh meat and meat products of improved
quality”, in Improving the Sensory and Nutritional
Quality of Fresh Meat, Ed by J.P. Kerry and D. Ledward.
Woodhead Publishing, Cambridge, UK, p. 3 (2009).
31. J.R. Botta, Evaluation of Seafood Freshness Quality. VCH
Publishers, New York (1995).
32. A. Ouali, C.H. Herrera-Mendez, G. Coulis, S. Becila, A.
Boudjellal, L. Aubry and M.A. Sentandreu, “Revisiting
the conversion of muscle into meat and the underly-
ing mechanisms”, Meat Sci. 74, 44 (2006). doi:
10.1016/j.
meatsci.2006.05.010
33. G. Ripoll, P. Albertí, B. Panea, J.L. Olleta and C. Sañudo,
“Near-infrared refl ectance spectroscopy for predict-
ing chemical, instrumental and sensory quality of
beef”, Meat Sci. 80, 697 (2008). doi:
10.1016/j.meat-
sci.2008.03.009
34 O. Raatikainen, V. Reinikainen, P. Minkkinen, T. Ritvanen,
P. Muje, J. Pursiainen, T. Hiltunen, P. Hyvönen, A. v.
Wright and S.-P. Reinikainen, “Multivariate modelling of
fi sh freshness index based on ion mobility spectrometry
measurements”, Anal. Chim. Acta 544, 128 (2005). doi:
10.1016/j.aca.2005.02.029
35. L.M. Reid, C.P. O’Donnell and G. Downey, “Recent
technological advances for the determination of food
authenticity”, Trends Food Sci. Technol. 17, 344 (2006).
doi:
10.1016/j.tifs.2006.01.006
36. N. Prieto, S. Andrés, F.J. Giráldez, A.R. Mantecón and P.
Lavín, “Potential use of near infrared refl ectance spec-
troscopy (NIRS) for the estimation of chemical composi-
tion of oxen meat samples”, Meat Sci. 74, 487 (2006). doi:
10.1016/j.meatsci.2006.04.030
37. R.J. Polkinghorne and J.M. Thompson, “Meat standards
and grading: A world view”, Meat Sci. 86, 227 (2010). doi:
10.1016/j.meatsci.2010.05.010
38. B.M. Franke, G. Gremaud, R. Hadorn and M. Kreuzer,
“Geographic origin of meat—elements of an analytical
approach to its authentication”, Eur. Food Res. Tech. 221,
493 (2005). doi:
10.1007/s00217-005-1158-8
39. N.F. Haard, “Control of chemical composition and food
quality attributes of cultured fi sh”, Food Res. Int. 25, 289
(1992). doi:
10.1016/0963-9969(92)90126-P
40. I. Martinez, M. Aursand, U. Erikson, T.E. Singstad, E.
Veliyulin and C. van der Zwaag, “Destructive and non-
destructive analytical techniques for authentication and
composition analyses of foodstuffs”, Trends Food Sci.
Technol. 14, 489 (2003). doi:
10.1016/j.tifs.2003.07.005
41. R. Clarke, “On-line measurement of meat composi-
tion”, Encyclopedia of Meat Science, Ed by M. Dikeman,
C. Devine, and W.K. Jensen. Elsevier, Amsterdam,
The Netherlands, p. 935 (2004). doi:
10.1016/B0-12-
464970-X/00100-8
42. D. Tejerina, M.M. López-Parra and S. García-Torres,
“Potential use of near infrared refl ectance spectros-
copy to predict meat physico–chemical composition of
guinea fowl (Numida meleagris) reared under different
production systems”, Food Chem. 113, 1290 (2009). doi:
10.1016/j.foodchem.2008.08.044
43. V.H. Segtnan, K.I. Hildrum and J.P. Wold, “New meth-
ods for analysis of factors affecting meat eating qual-
ity”, in Improving the Sensory and Nutritional Quality of
Fresh Meat, Ed by J. Kerry and D. Ledward. Woodhead
Publishing, Cambridge, UK, p. 519 (2009).
44. L. Ababouch, “Assuring fi sh safety and quality in inter-
national fi sh trade”, Mar. Pollut. Bull. 53, 561 (2006). doi:
10.1016/j.marpolbul.2006.08.011
45. World Trade Organisation, Sanitary and Phytosanitary
Measures (2005). Available at:
http://www.wto.org/eng-
lish/tratop_e/sps_e/sps_e.htm
46. World Trade Organisation, Agreement on Technical
Barriers to Trade (2005). Available at:
http://www.wto.
org/english/docs_e/legal_e/17-tbt_e.htm
47. D.J. Dahm and K.D. Dahm, “Representative layer theory
for diffuse refl ectance”, Appl. Spectrosc. 53, 647 (1999).
48. D.J. Dahm and K.D. Dahm, Interpreting Diffuse
Refl ectance and Transmittance. IM Publications,
Chichester, UK (2007).
49. S.H. Lee, S.T. Joo and Y.C. Ryu, “Skeletal muscle fi ber
type and myofi brillar proteins in relation to meat
quality”, Meat Sci. 86, 166 (2010). doi:
10.1016/j.meat-
sci.2010.04.040
50. G.L. Bosco, “James L. Waters Symposium 2009 on
near-infrared spectroscopy”, Trends Anal. Chem. 29, 197
(2010). doi:
10.1016/j.trac.2010.01.003
51. M. Viljoen, L.C. Hoffman and T.S. Brand, “Prediction of
the chemical composition of freeze dried ostrich meat
with near infrared refl ectance spectroscopy”, Meat Sci.
69, 255 (2005). doi:
10.1016/j.meatsci.2004.07.008
52. Å. Rinnan, L. Nørgaard, F. v.d. Berg, J. Thygesen, R. Bro
and S.B. Engelsen, “Data pre-processing”, in Infrared
Spectroscopy for Food Quality Analysis and Control.
Academic Press, San Diego, California, p. 29 (2009).
53. B. Igne, J.B. Reeves, G. McCarty, W.D. Hively, E. Lund
and C.R.J. Hurburgh, “Evaluation of spectral pre-
treatments, partial least squares, least squares sup-
port vector machines and locally weighted regression
for quantitative spectroscopic analysis of soils”, J. Near
Infrared Spectrosc. 18, 167 (2010). doi:
10.1255/jnirs.883

40 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
54. D. Alomar, C. Gallo, M. Castañeda and R. Fuchslocher,
“Chemical and discriminant analysis of bovine meat by
near infrared refl ectance spectroscopy (NIRS)”, Meat
Sci. 63, 441 (2003). doi:
10.1016/s0309-1740(02)00101-8
55. M. Prevolnik, M
ˇ
Candek-Potokar, D. Škorjanc, Š.
Velikonja-Bolta, M. Škrlep and T. Žnidaršiè, “Predicting
intramuscular fat content in pork and beef by near
infrared spectroscopy”, J. Near Infrared Spectrosc. 13,
77 (2005). doi:
10.1255/jnirs.460
56. V. Sierra, N. Aldai, P. Castro, K. Osoro, A. Coto-Montes
and M. Olivan, “Prediction of the fatty acid composi-
tion of beef by near infrared transmittance spectros-
copy”, Meat Sci. 78, 248 (2008). doi:
10.1016/j.meat-
sci.2007.06.006
57. S.G. Bajwa, J. Kandaswamy and J.K. Apple,
“Spectroscopic evaluation of the nutrient value of ground
beef patties”, J. Food Eng. 92, 454 (2009). doi:
10.1016/j.
jfoodeng.2008.12.025
58. K. Rosenvold, E. Micklander, P.W. Hansen, R. Burling-
Claridge, M. Challies, C. Devine and M. North,
“Temporal, biochemical and structural factors that
infl uence beef quality measurement using near infrared
spectroscopy”, Meat Sci. 82, 379 (2009). doi:
10.1016/j.
meatsci.2009.02.010
59. J.S. Yang, Q.X. Meng, L.P. Ren, Z.M. Zhou and X.X. Xie,
“Rapid evaluation of beef quality by NIRS technology”,
Spectrosc. Spectr. Anal. 30, 685 (2010). doi:
10.3964/j.
issn.1000-0593(2010)03-0685-03
60. A.H. Hoving-Bolink, H.W. Vedder, J.W.M. Merks, W.J.H.
de Klein, H.G.M. Reimert, R. Frankhuizen, W.H.A.M.
van den Broek and E. Lambooij, “Perspective of NIRS
measurements early post mortem for prediction of pork
quality”, Meat Sci. 69, 417 (2005). doi:
10.1016/j.meat-
sci.2004.08.012
61. B. Savenije, G.H. Geesink, J.G.P. van der Palen
and G. Hemke, “Prediction of pork quality
using visible/near-infrared refl ectance spectroscopy”,
Meat Sci. 73, 181 (2006). doi:
10.1016/j.meat-
sci.2005.11.006
62. V. Ortiz-Somovilla, F. España-España, A.J. Gaitán-
Jurado, J. Pérez-Aparicio and E.J. De Pedro-Sanz,
“Proximate analysis of homogenized and minced mass
of pork sausages by NIRS”, Food Chem. 101, 1031 (2007).
doi:
10.1016/j.foodchem.2006.02.058
63. A.J. Gaitán-Jurado, V. Ortiz-Somovilla, F. España-
España, J. Pérez-Aparicio and E.J. De Pedro-Sanz,
“Quantitative analysis of pork dry-cured sausages to
quality control by NIR spectroscopy”, Meat Sci. 78, 391
(2008). doi:
10.1016/j.meatsci.2007.07.005
64. M. Müller and M.R.L. Scheeder, “Determination of fatty
acid composition and consistency of raw pig fat with
near infrared spectroscopy”, J. Near Infrared Spectrosc.
16, 305 (2008). doi:
10.1255/jnirs.792
65. D. Pérez-Marín, E. De Pedro Sanz, J.E. Guerrero-Ginel
and A. Garrido-Varo, “A feasibility study on the use of
near-infrared spectroscopy for prediction of the fatty
acid profi le in live Iberian pigs and carcasses”, Meat Sci.
83, 627 (2009). doi:
10.1016/j.meatsci.2009.07.012
66. M.I. González-Martín, C.F. Bermejo, J.M.H. Hierro
and C.I.S. González, “Determination of hydroxy-
proline in cured pork sausages and dry cured beef
products by NIRS technology employing a fi bre-optic
probe”, Food Control 20, 752 (2009). doi:
10.1016/j.food-
cont.2008.09.015
67. C. Collell, P. Gou, P. Picouet, J. Arnau and J.
Comaposada, “Feasibility of near-infrared spectros-
copy to predict a
w
and moisture and NaCl contents of
fermented pork sausages”, Meat Sci. 85, 325 (2010). doi:
10.1016/j.meatsci.2010.01.022
68. V.M. Fernández-Cabanás, O. Polvillo, R. Rodríguez-
Acuña, B. Botella and A. Horcada, “Rapid determination
of the fatty acid profi le in pork dry-cured sausages by
NIR spectroscopy”, Food Chem. 124, 373 (2011). doi:
10.1016/j.foodchem.2010.06.031
69. D. Pérez-Marín, T. Fearn, J.E. Guerrero and A. Garrido-
Varo, “Robustness in pig fat NIRS calibrations by
orthogonal projection”, Chemometr. Intell. Lab. Syst. 100,
36 (2010). doi:
10.1016/j.chemolab.2009.10.003
70. M. Pérez-Juan, N.K. Afseth, J. González, I. Díaz, M.
Gispert, M.F.i. Furnols, M.A. Oliver and C.E. Realini,
“Prediction of fatty acid composition using a NIRS
fi bre optics probe at two different locations of ham
subcutaneous fat”, Food Res. Int. 43, 1416 (2010). doi:
10.1016/j.foodres.2010.04.006
71. S. Andrés, I. Murray, E.A. Navajas, A.V. Fisher, N.R.
Lambe and L. Bünger, “Prediction of sensory char-
acteristics of lamb meat samples by near infrared
refl ectance spectroscopy”, Meat Sci. 76, 509 (2007). doi:
10.1016/j.meatsci.2007.01.011
72. M. Viljoen, L.C. Hoffman and T.S. Brand, “Prediction of
the chemical composition of mutton with near infrared
refl ectance spectroscopy”, Small Ruminant Res. 69, 88
(2007). doi:
10.1016/j.smallrumres.2005.12.019
73. I.T. Kadim, O. Mahgoub, W. Al-Marzooqi and K.
Annamalai, “Prediction of crude protein, extractable
fat, calcium and phosphorus contents of broiler chicken
carcasses using near-infrared refl ectance spectros-
copy”, Asian Austral. J. Anim. Sci. 18, 1036 (2005).
74. R.M. McDevitt, A.J. Gavin, S. Andrés and I. Murray, “The
ability of visible and near infrared refl ectance spec-
troscopy to predict the chemical composition of ground
chicken carcass and to discriminate between carcasses
from different genotypes”, J. Near Infrared Spectrosc. 13,
109 (2005). doi:
10.1255/jnirs.463
75. P. Berzaghi, A. Dalla Zotte, L.M. Jansson and I.
Andrighetto, “Near-infrared refl ectance spectroscopy
as a method to predict chemical composition of breast
meat and discriminate between different n-3 feeding
sources”, Poult. Sci. 84, 128 (2005).
76. M. Pla, P. Hernández, B. Ariño, J.A. Ramírez and I. Díaz,
“Prediction of fatty acid content in rabbit meat and dis-
crimination between conventional and organic produc-

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 41
tion systems by NIRS methodology”, Food Chem. 100,
165 (2007). doi:
10.1016/j.foodchem.2005.09.029
77. M. Uddin, E. Okazaki, H. Fukushima, S. Turza, Y. Yumiko
and Y. Fukuda, “Nondestructive determination of
water and protein in surimi by near-infrared spectros-
copy”, Food Chem. 96, 491 (2006). doi:
10.1016/j.food-
chem.2005.04.017
78. A. Folkestad, J.P. Wold, K.-A. Rørvik, J. Tschudi, K.H.
Haugholt, K. Kolstad and T. Mørkøre, “Rapid and non-
invasive measurements of fat and pigment concentra-
tions in live and slaughtered Atlantic salmon (Salmo
salar L.)”, Aquaculture 280, 129 (2008). doi:
10.1016/j.
aquaculture.2008.04.037
79. L. Fasolato, A. Manfrin, C. Corrain, A. Perezzani, G.
Arcangeli, M. Rosteghin, E. Novelli, R.M. Lopparelli, S.
Balzan, M. Mirisola, L. Serva, S. Segato and E. Bianchi,
“Assessment of quality-parameters and authentication
in sole (Solea vulgaris) by NIRS (near infrared refl ect-
ance spectroscopy)”, Ind. Aliment-Italy 47, 355 (2008).
80. A. Ervik, P.K. Hansen, J. Aure, A. Stigebrandt, P.
Johannessen and T. Jahnsen, “Regulating the local
environmental impact of intensive marine fi sh farming I.
The concept of the MOM system (Modelling–Ongrowing
fi sh farms–Monitoring)”, Aquaculture 158, 85 (1997). doi:
10.1016/S0044-8486(97)00186-5
81. R.K. Miller, “Factors affecting the quality of raw meat”,
in Meat Processing, Ed by J. Kerry, J. Kerry and D.
Ledward. Woodhead Publishing, Cambridge, UK, p. 27
(2002).
82. D.N. D’Souza, D.W. Pethick, F.R. Dunshea, D. Suster,
J.R. Pluske and B.P. Mullan, “The pattern of fat and lean
muscle tissue deposition differs in the different pork
primal cuts of female pigs during the fi nisher growth
phase”, Livestock Prod. Sci. 91, 1 (2004). doi:
10.1016/j.
livprodsci.2004.04.005
83. J.D. Wood, R.I. Richardson, G.R. Nute, A.V. Fisher,
M.M. Campo, E. Kasapidou, P.R. Sheard and M. Enser,
“Effects of fatty acids on meat quality: a review”, Meat
Sci. 66, 21 (2004). doi:
10.1016/S0309-1740(03)00022-6
84. C.E. Realini, S.K. Duckett and W.R. Windham, “Effect
of vitamin C addition to ground beef from grass-fed
or grain-fed sources on color and lipid stability and
prediction of fatty acid composition by near-infrared
refl ectance analysis”, Meat Sci. 68, 35 (2004). doi:
10.1016/j.meatsci.2004.02.002
85. I. González-Martín, C. González-Pérez, N. Alvarez-
GarcIa and J.M. González-Cabrera, “On-line determi-
nation of fatty acid composition in intramuscular fat
of Iberian pork loin by NIRS with a remote refl ectance
fi bre optic probe”, Meat Sci. 69, 243 (2005). doi:
10.1016/j.
meatsci.2004.07.003
86. Y. Huang, A.G. Cavinato, D.M. Mayes, G.E. Bledsoe and
B.A. Rasco, “Nondestructive prediction of moisture and
sodium chloride in cold smoked Atlantic salmon (Salmo
salar)”, J. Food Sci. 67, 2543 (2002). doi:
10.1111/j.1365-
2621.2002.tb08773.x
87. P. Berzaghi and R. Riovanto, “Near infrared spectros-
copy in animal science production: principles and appli-
cations”, Ital. J. Anim. Sci. 8, 39 (2009).
88. S. Andrés, A. Silva, A.L. Soares-Pereira, C. Martins,
A.M. Bruno-Soares and I. Murray, “The use of visible
and near infrared refl ectance spectroscopy to pre-
dict beef M. longissimus thoracis et lumborum quality
attributes”, Meat Sci. 78, 217 (2008). doi:
10.1016/j.meat-
sci.2007.06.019
89. N. Prieto, S. Andrés, F.J. Giráldez, A.R. Mantecón and P.
Lavín, “Ability of near infrared refl ectance spectroscopy
(NIRS) to estimate physical parameters of adult steers
(oxen) and young cattle meat samples”, Meat Sci. 79, 692
(2008). doi:
10.1016/j.meatsci.2007.10.035
90. M.B. Bowling, D.J. Vote, K.E. Belk, J.A. Scanga, J.D.
Tatum and G.C. Smith, “Using refl ectance spectroscopy
to predict beef tenderness”, Meat Sci. 82, 1 (2009). doi:
10.1016/j.meatsci.2008.09.012
91. J.W.S. Yancey, J.K. Apple, J.F. Meullenet and J.T.
Sawyer, “Consumer responses for tenderness and
overall impression can be predicted by visible and near-
infrared spectroscopy, Meullenet–Owens razor shear
and Warner–Bratzler shear force”, Meat Sci. 85, 487
(2010). doi:
10.1016/j.meatsci.2010.02.020
92. M.
ˇ
Candek-Potokar, M. Prevolnik and M. Škrlep, “Ability
of near infrared spectroscopy to predict pork techno-
logical traits”, J. Near Infrared Spectrosc. 14, 269 (2006).
doi:
10.1255/jnirs.644
93. M. Prevolnik, M.
ˇ
Candek-Potokar, M. Novic and D.
Škorjanc, “An attempt to predict pork drip loss from
pH and colour measurements or near infrared spec-
tra using artifi cial neural networks”, Meat Sci. 83, 405
(2009). doi:
10.1016/j.meatsci.2009.06.015
94. Y.H. Hu, K.Q. Guo, N. Gou, K. Sumio and S. Takaaki,
“Prediction of water holding capacity of fresh pork using
near infrared spectroscopy”, Spectrosc. Spectr. Anal.
29, 3259 (2009). doi:
10.3964/j.issn.1000-0593(2009)12-
3259-04
95. Y.T. Liao, Y.X. Fan, X.Q. Wu and F. Cheng, “Online deter-
mination of pH in fresh pork by visible/near-infrared
spectroscopy”, Spectrosc. Spectr. Anal. 30, 681 (2010).
doi:
10.3964/j.issn.1000-0593(2010)03-0681-04
96. M. Prevolnik, M. Candek-Potokar and D. Skorjanc,
“Predicting pork water-holding capacity with NIR
spectroscopy in relation to different reference meth-
ods”, J. Food Eng. 98, 347 (2010). doi:
10.1016/j.jfood-
eng.2009.11.022
97. V.A. McGlone, C.E. Devine and R.W. Wells, “Detection
of tenderness, post-rigor age and water status
changes in sheep meat using near infrared spectros-
copy”, J. Near Infrared Spectrosc. 13, 277 (2005). doi:
10.1255/jnirs.558
98. Y. Liu, B.G. Lyon, W.R. Windham, C.E. Lyon and E.M.
Savage, “Prediction of physical, color and sensory char-
acteristics of broiler breasts by visible/near infrared
refl ectance spectroscopy”, Poult. Sci. 83, 1467 (2004).

42 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
99. D.H. Clark, H.F. Mayland and R.C. Lamb, “Mineral
analysis of forages with near infrared refl ectance spec-
troscopy”, Agron. J. 79, 485 (1987). doi:
10.2134/agronj19
87.00021962007900030016x
100. S.D. Shackelford, T.L. Wheeler and M. Koohmaraie,
“On-line classifi cation of US Select beef carcasses for
longissimus tenderness using visible and near-infrared
refl ectance spectroscopy”, Meat Sci. 69, 409 (2005). doi:
10.1016/j.meatsci.2004.08.011
101. D.A. King, T.L. Wheeler, S.D. Shackelford and M.
Koohmaraie, “Fresh meat texture and tenderness”, in
Improving the Sensory and Nutritional Quality of Fresh
Meat, Ed by J.P. Kerry and D. Ledward. Woodhead
Publishing, Cambridge, UK, p. 61 (2009).
102. J.J. Xia, E.P. Berg, J.W. Lee and G. Yao, “Characterizing
beef muscles with optical scattering and absorption
coeffi cients in VIS–NIR region”, Meat Sci. 75, 78 (2007).
doi:
10.1016/j.meatsci.2006.07.002
103. N. Prieto, R. Roehe, P. Lavín, G. Batten and S. Andrés,
“Application of near infrared refl ectance spectroscopy to
predict meat and meat products quality: A review”, Meat
Sci. 83, 175 (2009). doi:
10.1016/j.meatsci.2009.04.016
104. C. Lopez-Bote, P.D. Warriss and S.N. Brown, “The use
of muscle protein solubility measurements to assess
pig lean meat quality”, Meat Sci. 26, 167 (1989). doi:
10.1016/0309-1740(89)90018-1
105. D. Cozzolino, D.D. Mattos and D. Vaz Martins, “Visible/
near infrared refl ectance spectroscopy for predicting
composition and tracing system of producton of beef
muscle”, Anim. Sci. 74, 477 (2002).
106. H.J. Swatland, “On-line monitoring of meat qual-
ity”, in Meat Processing, Ed by J. Kerry, J. Kerry and D.
Ledward. Woodhead Publishing, Cambridge, UK, p. 193
(2002).
107. N. Prieto, D.W. Ross, E.A. Navajas, G.R. Nute, R.I.
Richardson, J.J. Hyslop, G. Simm and R. Roehe,
“On-line application of visible and near infrared refl ect-
ance spectroscopy to predict chemical–physical and
sensory characteristics of beef quality”, Meat Sci. 83, 96
(2009). doi:
10.1016/j.meatsci.2009.04.005
108. R.M. García-Rey, J. García-Olmo, E. De Pedro, R.
Quiles-Zafra and M.D. Luque de Castro, “Prediction of
texture and colour of dry-cured ham by visible and near
infrared spectroscopy using a fi ber optic probe”, Meat
Sci. 70, 357 (2005). doi:
10.1016/j.meatsci.2005.02.001
109. N.Z. Ballin, “Authentication of meat and meat prod-
ucts”, Meat Sci. 86, 577 (2010). doi:
10.1016/j.meat-
sci.2010.06.001
110. G. Xiccato, A. Trocino, F. Tulli and E. Tibaldi, “Prediction
of chemical composition and origin identifi cation of
european sea bass (Dicentrarchus labrax L.) by near
infrared refl ectance spectroscopy (NIRS)”, Food Chem.
86, 275 (2004). doi:
10.1016/j.foodchem.2003.09.026
111. N. Prieto, S. Andrés, F.J. Giráldez, A.R. Mantecón and
P. Lavín, “Discrimination of adult steers (oxen) and
young cattle ground meat samples by near infrared
refl ectance spectroscopy (NIRS)”, Meat Sci. 79, 198
(2008). doi:
10.1016/j.meatsci.2007.08.001
112. M.T. Osorio, J.M. Zumalacárregui, N. Prieto, F.J.
Giráldez, S. Andrés and J. Mateo, “Differentiation
between carcasses from suckling lambs reared with
ewe milk or milk replacers by near infrared refl ectance
spectroscopy of perirenal fat”, Small Ruminant Res. 72,
221 (2007). doi:
10.1016/j.smallrumres.2006.10.011
113. P.H.M. Dian, D. Andueza, M. Jestin, I.N. Prado and
S. Prache, “Comparison of visible and near infrared
refl ectance spectroscopy to discriminate between pas-
ture-fed and concentrate-fed lamb carcasses”, Meat Sci.
80, 1157 (2008). doi:
10.1016/j.meatsci.2008.05.009
114. L. Arce, A. Domínguez-Vidal, V. Rodríguez-Estévez,
S. López-Vidal, M.J. Ayora-Cañada and M. Valcárcel,
“Feasibility study on the use of infrared spectroscopy
for the direct authentication of Iberian pig fattening
diet”, Anal. Chim. Acta 636, 183 (2009). doi:
10.1016/j.
aca.2009.01.058
115. F.G. del Moral, A. Guillén, L.G. del Moral, F. O’Valle, L.
Martínez and R.G. del Moral, “Duroc and Iberian pork
neural network classifi cation by visible and near infra-
red refl ectance spectroscopy”, J. Food Eng. 90, 540
(2009). doi:
10.1016/j.jfoodeng.2008.07.027
116. A. Guillen, F.G. del Moral, L.J. Herrera, G. Rubio, I.
Rojas, O. Valenzuela and H. Pomares, “Using near-
infrared spectroscopy in the classifi cation of white and
Iberian pork with neural networks”, Neural Comput.
Appl. 19, 465 (2010). doi:
10.1007/s00521-009-0327-2
117. M. Uddin, E. Okazaki, S. Turza, Y. Yumiko, M. Tanaka and
Y. Fukuda, “Non-destructive visible/NIR spectroscopy
for differentiation of fresh and frozen-thawed fi sh”, J.
Food Sci. 70, c506 (2005). doi:
10.1111/j.1365-2621.2005.
tb11509.x
118. V. Ortiz-Somovilla, F. España-España, E.J. De Pedro-
Sanz and A.J. Gaitán-Jurado, “Meat mixture detection
in Iberian pork sausages”, Meat Sci. 71, 490 (2005). doi:
10.1016/j.meatsci.2005.04.028
119. R.R. Gangidi, A. Proctor, F.W. Pohlman and J.-F.
Meullenet, “Rapid determination of spinal cord content in
ground beef by near-infrared spectroscopy”, J. Food Sci.
70, c397 (2005). doi:
10.1111/j.1365-2621.2005.tb11436.x
120. J. Gayo and S.A. Hale, “Detection and quantifi cation of
species authenticity and adulteration in crabmeat using
visible and near-infrared spectroscopy”, J. Agric. Food
Chem. 55, 585 (2007). doi:
10.1021/jf061801+
121. M.R. Ramírez and R. Cava, “Effect of Iberian × Duroc
genotype on dry-cured loin quality”, Meat Sci. 76, 333
(2007). doi:
10.1016/j.meatsci.2006.11.017
122. G. Downey and D. Beauchêne, “Discrimination between
fresh and frozen-then-thawed beef m. longissimus dorsi
by combined visible-near infrared refl ectance spectros-
copy: A feasibility study”, Meat Sci. 45, 353 (1997). doi:
10.1016/s0309-1740(96)00127-1
123. M. Uddin, E. Okazaki, M.U. Ahmad, Y. Fukuda and M.
Tanaka, “Noninvasive NIR spectroscopy to verify end-

J. Weeranantanaphan et al., J. Near Infrared Spectrosc. 19, xxx–xxx (2011) 43
point temperature of kamaboko gel”, LWT – Food Sci.
Technol. 38, 809 (2005). doi:
10.1016/j.lwt.2004.10.008
124. M. Lin, M. Mousavi, M. al-Holy, A.G. Cavinato and B.A.
Rasco, “Rapid near infrared spectroscopic method for
the detection of spoilage in rainbow trout (Oncorhynchus
mykiss) fi llet”, J. Food Sci. 71, S18 (2006). doi:
10.1111/
j.1365-2621.2006.tb12400.x
125. M. Uddin, E. Okazaki, M.U. Ahmad, Y. Fukuda and
M. Tanaka, “NIR spectroscopy: A non-destructive
fast technique to verify heat treatment of fi sh-meat
gel”, Food Control 17, 660 (2006). doi:
10.1016/j.food-
cont.2005.04.009
126. C.H. Chou, C.Y.L. Chyr, R.L.C. Chen and J.T. Shaw,
“Evaluation of freshness of ground pork tenderloin by
visible/near infrared spectroscopy”, J. Food Drug Anal.
18, 107 (2010).
127. J. Stawczyk, I. Munoz, C. Collell and J. Comaposada,
“Control system for sausage drying based on on-line
NIR a(w) determination”, Dry. Technol. 27, 1338 (2009).
doi:
10.1080/07373930903383620
128. K. Thyholt, G. Enersen and T. Isaksson, “Determination
of endpoint temperatures in previously heat treated
beef using refl ectance spectroscopy”, Meat Sci. 48, 49
(1998). doi:
10.1016/s0309-1740(97)00075-2
129. D.-W. Sun (Ed.), Hyperspectral Imaging for Food Quality
and Control. Academic Press/Elsevier, Missouri, USA
(2010).
130. G. El Masry, A. Iqbal, D.-W. Sun, P. Allen and P. Ward,
“Quality classifi cation of cooked, sliced turkey hams
using NIR hyperspectral imaging system”, J. Food Eng.
103, 333 (2011).
APC aerobic plate count
A
w
water activity
BFA branched fatty acids
BFNN basis function neural network
BPANN back propagation artifi cial neural network
CLA conjugated linoleic acids
CP-ANN counter-propagation artifi cial neural network
DT de-trending
DM dry matter
EMSC extended multiplicative scatter correction
FM fresh matter
GLS generalised least squares
I interactance
IMF intramuscular fat
LDA linear discriminant analysis
MI mutual information ranking
MLR multi-linear regression
MSC multiplicative scatter correction
MUFA monounsaturated fatty acids
PCR principal component regression
PLS partial least squares regression
PM post-mortem
PUFA polyunsaturated fatty acids
Rrefl ectance
R
2
coeffi cient of determination in calibration
r
2
coeffi cient of determination in validation
r
2
cv
coeffi cient of determination in cross-validation
RBFNN radial basis function neural network
RPD ratio of prediction error (SEP) to range in reference values (SD)
RMSECV root mean square error of cross-validation
SEP standard error of prediction
SD standard deviation
SEL standard error of laboratory
SFA saturated fatty acids
SFC solid fat content
Appendix

44 Near Infrared Spectroscopy in Muscle Food Analysis: 2005–2010
SIMCA soft independent modelling of class anology
SNV-DT standard normal variate and de-trending
SSF slice shear force
SVM support vector machines
T transmittance
TFC solid fat content
TOP transfer by orthogonal projection algorithm
VBN volatile basic nitrogen
VN vector normalisation
WBSF Warner–Bratzler shear force
WHC water-holding capacity