Towards Commoditised Near Infrared Spectroscopy

Abstract

Near Infrared Spectroscopy (NIRS) is a sensing technique in which near infrared light is transmitted into a sample, followed by light absorbance measurements at various wavelengths. This technique enables the inference of the inner chemical composition of the scanned sample, and therefore can be used to identify or classify objects. In this paper, we describe how to facilitate the use of NIRS by non-expert users in everyday settings. Our work highlights the key challenges of placing NIRS devices in the hands of non-experts. We develop a system to mitigate these challenges, and evaluate it in a user study. We show how NIRS technology can be successfully utilised by untrained users in an unsupervised manner through a special enclosure and an accompanying smartphone app. Finally, we discuss potential future developments of commoditised NIRS.

Full text

!"#$%&'()"**"&+,+'-&(.-$%(/01%$%-&(23-4,%"'4"35(
Simon Klakegg1, Jorge Goncalves2, Niels van Berkel2, Chu Luo2, Simo Hosio1, Vassilis Kostakos2
1Center for Ubiquitous Computing, University of Oulu
2School of Computing and Information Systems, The University of Melbourne
simon.klakegg@oulu.fi, jorge.goncalves@unimelb.edu.au, n.vanberkel@student.unimelb.edu.au,
chul3@student.unimelb.edu.au, simo.hosio@oulu.fi, vassilis.kostakos@unimelb.edu.au
672!86)!(
Near Infrared Spectroscopy (NIRS) is a sensing technique in
which near infrared light is transmitted into a sample,
followed by light absorbance measurements at various
wavelengths. This technique enables the inference of the
inner chemical composition of the scanned sample, and
therefore can be used to identify or classify objects. In this
paper, we describe how to facilitate the use of NIRS by non-
expert users in everyday settings. Our work highlights the
key challenges of placing NIRS devices in the hands of non-
experts. We develop a system to mitigate these challenges,
and evaluate it in a user study. We show how NIRS
technology can be successfully utilised by untrained users in
an unsupervised manner through a special enclosure and an
accompanying smartphone app. Finally, we discuss potential
future developments of commoditised NIRS.
69,:"%(;-5#"%&'(
Near Infrared Spectroscopy; sample identification; sensor
accessibility; user-induced errors; pharmaceuticals; user
study; gluten detection.
6)<()=$''+1+4$,+"0(;-5#"%&'(
H.5.m. Information interfaces and presentation (e.g., HCI):
Miscellaneous.
/.!8>?@)!/>.(
As can been seen in consumer electronics, sophisticated
hardware is becoming cheaper and more accessible to the
modern customer (e.g., smartphones), this includes Near
Infrared Spectroscopy (NIRS) scanners. NIRS has the ability
to penetrate the surface and traverse the physical structure of
an object. It allows for the retrieval of information about the
inner composition of a sample in the form of a spectrum,
which acts as a proverbial fingerprint [48] of the sample.
Thus, it enables accurate and detailed object identification.
While NIRS scanners have been used in research laboratories
for decades [38], only recently has the technology matured
enough to allow for end-user hardware which is both small
and robust enough to be carried around, while still capable
of producing reliable results. The NIRS device used in our
study (DLP NIRscan Nano [15]) costs under 1000 US dollars
at the time of writing and weighs just 80 grams – a fraction
of the price and weight of high-end NIRS hardware. These
numbers can be expected to continue to decline, which in
turn encourages researchers to start considering everyday
scenarios for this technology. For the first time, we argue, it
can plausibly be placed in the hands of consumers. The main
objective of our work is to explore the improvement of the
accessibility of these devices for non-experts.
Coupling NIRS hardware with commodity devices (e.g.,
tablets, smartphones) opens a range of exciting research
avenues to explore. An in situ scanner placed at supermarkets
could enable classification of products while they are being
weighted, and report to the users the current state of the
product (e.g., the level of ripeness of a fruit [45]). In a home
scenario, augmented shelves can detect the food being
stored, and even determine if it has gone bad [3]. In a
domiciliary health scenario, a user can scan a pill or medicine
to confirm that this pill is the correct one to take at that
moment [9]. There are a vast number of potential use cases
for an everyday device that can identify objects based on
their physical composition and ingredients.
The contribution of our work is three-fold. First, we explore
the challenges in obtaining reliable scanning results, namely
the impact of user-induced errors on scan accuracy. This is
an unexplored territory in the context of miniaturised NIRS
scanning. Second, we map the complexity and required
knowledge to carry out a full sample analysis [27]. Third, we
design and evaluate a set of mechanisms to address these
issues and make NIRS more accessible and usable to non-
experts. Our design consists of a custom enclosure to
physically guide the user in the scanning process, and a
mobile application to assist the user, inform of scanning
errors during usage, and encapsulate the sample analysis
process. We evaluate our design in a user study to collect
feedback and ideas for future improvements. Our findings
show that non-experts can successfully use this technology
when both physical and procedural guides are in place.
8AB6!A?(C>8;(
<$D+0E(2-0'"%'(644-''+F=-(
Sensors play an increasingly larger role in consumer
electronics. Today, consumers’ personal devices are
Permissio n to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. Copyrights for
components of this work owned by others than the author(s) must be
honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior
specific permission and/or a fee. Request permissions from
Permissio ns@acm.org
.
DIS 2017, June 10-14, 2017, Edinburgh, United Kingdom
Copyright is held by the owner/author(s). Publication rights licensed to
ACM.

ACM 978-1-4503-4922-2/17/06…$15.00

DOI: http://dx.doi.org/10.1145/3064663.3064738

embedded with a plethora of different hardware sensors,
such as accelerometer, gravity, gyroscope, light, and
magnetic [47]. Beyond providing benefits to the users
themselves, these sensors have proven to be valuable tools
when conducting research in fields such as traffic
management [5], HCI [41], localisation [10], and healthcare
[32]. While these electronic components are now
commoditised and embedded as an integral part in a variety
of products, they were originally much larger standalone
devices. For example, the first accelerometer weighted
almost 0.5 kg and measured 1.90 x 4.76 x 21.59 cm in size
[53]. The size, weight, and cost of the accelerometer was
eventually reduced, while the use cases expanded [53].
Initially, the accelerometer was used mostly in industry, but
now benefits end-users with service enhancement in
numerous applications. Further, smart devices do not only
serve as a host for embedded sensors, but frequently act as
control points for smaller and mobile sensors. This,
combined with the widespread nature of smartphones [42],
allows for innovative services. For instance, smartphones
can be connected to portable medical devices (e.g., blood
pressure, glucose, and pulse oximeter) to process, analyse,
and present biohealth data [22,31]. It is increasingly feasible
for end-users to self-monitor their health, without the need
for trained healthcare personnel. In essence, the development
of these sensors and their increased accessibility help bridge
the gap between end-users and advanced medical equipment.
In another example, Samsung’s Galaxy Note 7 includes an
infrared iris sensor, which allows biometric authentication on
consumers’ devices for security and access control [43].
Similarly, while until recently only a few devices came with
fingerprint sensors [19], this technology has become
increasingly more accessible for end-users. Goel et al. [20]
showcase how a hyperspectral sensor, previously used for
research, can assist in a wider set of use cases. Using
functional Near-Infrared Spectroscopy (fNIRS) for Brain–
Computer Interfaces (BCI) has begun to proliferate in the
HCI community [30,33,40,51]. Strait et al. [52] investigate
the reliability of fNIRS in BCI through a user study, and
highlight some of the obstacles. Yuksel et al. [54] develop a
system that automatically adjusts musical learning tasks
based on the user’s cognitive workload measured by fNIRS.
In our work, we take initial usability steps towards
facilitating the use of NIRS scanners by non-experts and by
using smartphones as a control point.
.-$%(/01%$%-&(23-4,%"'4"35(
NIRS sends near infrared light into a sample and measures
the absorbance at various wavelengths, thus allowing for
object identification [7]. Because of the characteristics of the
NIR band (780 nm to 2500 nm), it can penetrate objects up
to several millimetres. This enables quick and accurate
analysis of the inner composition of samples (e.g., analysing
if a food product contains gluten), something that cannot be
attained using computer vision. NIRS has been shown to be
a useful technique in research across many different fields
[16,17,49]. One of the more popular use cases of NIRS is to
verify the quality of food. Sinelli et al. [49] used NIRS to
analyse the freshness of minced beef. By comparing scans
taken at different times, they were able to accurately
determine the expiry date of the product. Others have
explored NIRS as a way of developing a non-destructive
solution for analysing fruit quality [28]. NIRS methods can
reveal information about the inside of the fruit, something
impossible to achieve solely through visual inspection. The
reflected spectra contain the information used to infer the
fruit’s maturity, pH factor, solid content, and flesh elasticity.
Another distinct feature of NIRS is the possibility to scan an
item with no manipulation or pre-treatment applied to the
object itself [6]. For this reason, NIRS has also gained
popularity in the pharmaceutical industry [4]. This industry
is heavily regulated, with a need for fast and safe quality
control. NIRS can also help to address the challenge of
counterfeit drugs. It is estimated that around 7% of the
pharmaceuticals sold in the world are fake [13]. These
counterfeit drugs are typically defined as drugs that have
active substances which have been modified [13]. Similarly,
in textile industries there is a need to verify the fibres of
clothes at different stages in the production and recycling.
However, many of the current methods for textile
classification are both time consuming and involve
dangerous chemicals. Cleve et al. [12] report how they
accurately identified fabrics using NIRS. Durand et al. [16]
argue that NIRS can replace existing methods and report an
identification accuracy of over 96% for textiles. The
aforementioned use cases (i.e., food, pharmaceutical, textile)
provide realistic real-world usage scenarios and are suitable
for testing in a miniaturised NIRS context.
Miniaturised+Near+Infrared+Spectroscopy!
Most of the previous examples rely on heavy desktop NIRS
equipment or larger portable devices. Given their success,
miniaturised versions for field use have recently been
developed. For instance, a mobile NIRS device has been
used to detect tomato’s pathogen [1]. By scanning the fruit
just before it is harvested, it is possible to avoid large losses
of fruit being rejected in quality control. Instead of sending
the sample to a lab and waiting for a report, results can be
obtained in situ in a matter of seconds. Unlike many of the
current in situ chemical approaches, NIRS is also a non-
destructive method. However, farmers reported that they
considered the equipment to be both too expensive and too
difficult to use [13].
Recently, cheaper and smaller devices have entered the
market. However, there is a lack of research that investigates
the effects of user-noise and usability when the device is
placed in the hands of non-expert. Specifically, the effect of
device motion [26], sample distance [46], sample angle [35],
sample surface [36], sample interference [56], and ambience
[50] have not been tested in the context of end-user
miniaturised NIRS usage. Therefore, in this paper we
investigate the effect of these parameters and consider ways
to overcome the challenges they may impose. Furthermore,
we propose a non-expert assistance for end users.

)G6BBA.HA2(>I(./82(2)6../.H(
There are two main challenges that users face when utilising
NIRS technologies: 1) the impact of user-induced errors on
the reliability of the results, and 2) the complexity of the
sample analysis process. The effect of the former is
exacerbated when NIRS devices are placed in the hands of
non-experts. The latter is a challenge pertaining to
commoditising this technology, since we expect that
everyday users lack the skills to extract, analyse, and
interpret the output generated by the device. Next, we
describe in detail the impact of both these challenges on the
reliability of the results.
):$==-0E-(J(K(@'-%L/0&94-&(A%%"%'(
We experimentally investigate the limitations of NIRS
technology in terms of user-induced errors in a scanning
scenario with everyday materials. When a user interacts with
the device (e.g., place an object, hold the device), the user
may introduce some noise to the system that can degrade
scan accuracy. This could for example be caused by
improper sample placement or an (unintended) shaking
motion. To support non-expert NIRS usage, we have to
identify the magnitude of the various types of user-induced
errors. Moreover, based on the data from these test, we can
inform the design of our non-expert assistance.
To showcase the impact of each type of user-induced error,
we chose three different sample types (fruit, pharmaceutical,
textile) with varying characteristics, as shown in Table 1. In
some cases, we used different items from the same sample
type for a more appropriate evaluation of each parameter. For
example, when testing sample surface, it would not make
sense to test a round object with no edges or varying texture.
A NIRS scan was deemed inadequate when it produced
values that are significantly different from the values
obtained under ideal conditions (significance tested through
Wilcoxon signed-rank tests).
Sample
Properties
Test(s)
100% Cotton
Area (Square): 196 cm
Height: 2,5 cm
1 - 6
Omega 3
Brand: Möller
Weight: 1200 mg
2, 5, 6
Multivitamin Plus M
Brand: Orion Pharma
Weight: 250 mg
1, 3, 4
Banana
Uncut
Diameter: 20 cm
Weight: 120 g
1 - 5
Apple
Uncut
Diameter: 6 cm
Weight: 80 g
5, 6
Grape
Uncut
Diameter: 2 cm
Weight: 5 g
5
Table 1. Selected samples, properties, and respective tests.
Scan+setup+and+configuration+
Our scan setup consists of a NIRS scanner (DLP NIRscan
Nano [15]) and a camera tripod. We use a custom-built
aluminium plate and the tripod’s spirit level to align the
scanner with the scanned object. We use a separate spirit
level to ensure both an overall alignment and the incident
beam hitting the sample orthogonally. Figure 1 depicts the
scan setup. Our scan setup was installed in a room with
controlled light conditions, and a lux meter was used to
detect the amount of light visible in the room. The lux meter
detects wavelengths within the visible light range (400-
700nm). Fluorescent lamps, such as the ones in our testing
room, mainly radiate within the same spectra [18].
Therefore, the level of stray light interference can be
measured by the lux meter.
The DLP NIRscan Nano software allows the user to
configure and calibrate the device before starting a scan. A
reference scan was set using a labsphere Spectralon Diffuse
Reflectance Standard. This calibration tool reflects up to
99% of incoming light, allowing the machine to adjust to
changing hardware performance. The configuration during
all experiments used the whole available NIRS wavelength
range (900-1700nm). The resolution of the hardware
describes the smallest spectral features that it is able to
detect. In our case the hardware has an optical resolution of
10nm. The digital resolution can be increased at the cost of
decreasing the Signal-to-Noise Ratio (SNR). The lowest
setting with the manufacturer’s software is 7.02nm without
receiving warnings about degrading performance. Therefore
this setting was chosen as literature suggests that a higher
resolution helps to detect all spectral features [12,21,37]. The
manufacturer ships the device with four typical scan
configurations [15]. For a digital resolution of 7.02nm, it is
suggested that the spectra should be sampled approximately
228 times, meaning that it will be oversampled by 2 to satisfy
the Nyquist-Shannon Sampling Theorem [8].
Figure 1. Left: Scanning platform with the DLP NIRscan
Nano mounted and the lux meter. Right: DLP NIRscan Nano.
Specifically, the spectral bandwidth being sampled is 800nm
(i.e., 1700nm-900nm). With a digital resolution of 7.02nm,
we need 114 (i.e., 800nm / 7.02nm) patterns to sample the
whole spectra. A common way to increase the SNR while
sampling is by signal averaging [23]. We scan each object
six times, as advised by the manufacturer. Each sample is
recorded in the same manner, and all are averaged to increase
the SNR. Each data point in Figure 2 – 7 represents the
average mean absorbance of six individual scans.

There are two available scan methods for the device, which
determine how the wavelengths are scanned: Hadamard scan
and Column scan. Hadamard multiplexes several
wavelengths together and decodes individual wavelengths.
Noise in the incident signal is distributed evenly over the
spectrum to minimize the effect. This method also collects
more light and provides a greater SNR than the Column scan
[15], and was therefore chosen for our study.
Test+1:+Device+Motion+
The device could be used either as a handheld point and scan
device (e.g., baggage scanner) or as a stationary device (i.e.,
mobile, but placed on a table when scanning). We
investigated the effect of device motion on scan quality. If
even small movements from events such as hand tremor
cause inadequate scans, then the latter positioning technique
would be the best solution when using these devices.
Movement of the NIRS while scanning can cause the lens
window to lose contact with the sample. This may lead to
unwanted reflections, distortion of the signal, and exposure
to ambient light [26]. To explore the effect of device motion
on scan accuracy, we scan each sample while the device is
under three different levels of motion. First, we scan the
sample with the device placed on the table. Second, we scan
the sample while holding the device (i.e., light movement).
Lastly, we conduct scans with moderate motion to reflect
careless usage.
Figure 2. Mean absorbance of objects with different levels of
device motion.
Figure 2 shows how the quality of the scans degrades when
motion is applied to the NIRS when scanning. The
absorbance also increases for both the 100% cotton and the
banana sample, meaning less light is reflected. The
multivitamin experiences a decrease in absorbance, likely
caused by an increase in the lens exposure to ambient light.
Test+2:+Sample+Distance+
Device positioning depends on the sample distance range for
which it can deliver adequate results. Typically, lower ranges
yield more accurate results, and in such cases the scanner
should ideally be positioned upwards to enforce that the
samples are directly on top of the lens window. However, a
larger range would facilitate a more flexible design. The two
lens-end broadband tungsten filament lamps in the NIRS
hardware emit light through the device’s sapphire window.
Both the light source paths and the vision cone of the
receiving lens intersect directly in front of the window. The
further away the sample is positioned, the less light the
system is able to collect because of path loss [44]. In
addition, the reflected signal contains less information and is
more prone to noise. NIRS scanners assume that all light
which is not reflected has been absorbed, even though it may
have radiated through or around the object. While different
objects absorb at varying rate, lower absorbance (i.e., a
higher reflectance) often indicates that the object is closer
and more ideally positioned. In addition, a sample positioned
at close range will ensure less stray light entering the
spectrometer, thereby increasing accuracy [46].
To test the accuracy of our scans over different distances, we
attached a ruler to the table. Each sample was then measured
from 0cm to 3cm with increments of 1mm. The effect of the
distance on the absorbance is shown in Figure 3. All of the
curves display a similar characteristic, with the main
difference seen in the level of object absorbance (dependent
on the chemical composition of the sample). We can see that
pure cotton reflects back more light back than the omega-3
sample. This is because the omega-3 is transparent, with the
majority of incoming light being scattered in multiple
directions.
Figure 3. Mean absorbance of objects over different distances
to the scanner
Test+3:+Sample+Angle+
To design an appropriate sample holder for the hardware, we
identify the sample angle range in which the device delivers
adequate result. The holder needs to hold the sample flat
down if the range is limited. With adequate results in a larger
range, the design could be more universal with less focus on
holding the sample in a certain angle.
For scans to be of high quality, the scanner needs to collect
as much of the reflected light as possible. Ideally, the
incident wave should hit a flat surface with an angle of 0°,
following the law of reflection [35]. To measure the effects
of scanning an item at various angles, we scan samples from
0° to 90° with increments of 15°. Figure 4 shows the effect
of object angle on absorbance levels. For every increment of
15°, the quality deteriorates rapidly. While 0° is the optimal
angle to retain all the spectral properties, small items (e.g.,
multivitamins) reflect an adequate amount of light up to 15°
(as shown in Figure 4) because of the illumination angle of
the device.

Figure 4. Mean absorbance of objects over different scanning
angles.
Test+4:+Sample+Surface+
The sample may have a surface with varying texture. It is
crucial to know if the different textures will affect the scan
accuracy. If the texture degrades scan accuracy significantly,
the non-expert user should be instructed on how to correctly
place the item to scan the most appropriate surface. A flat
surface can reflect the signal in one direction, whereas a
rough surface may cause signal scattering [36]. To
investigate the consequences of scanning at various areas, we
scanned the same samples from 3 reference points and 3
uneven surfaces. The reference scans are taken at 0mm
distance and at a 0° angle at a flat surface covering most of
the lens. The uneven surfaces consist of parts of the object
that have edges or are rugged.
In Figure 5, we can observe that all the reference scans are
of adequate quality and have little variance between them. In
contrast, just one out of the nine uneven samples is of
adequate quality. The uneven cotton samples also have a
large variance compared to the rest. Since the cotton samples
were placed unfolded in front of the lens, they had a varying
thickness when we were scanning the three uneven samples.
Figure 5. Mean absorbance of objects with different surface
evenness
Test+5:+Sample+Interference+
When scanning a sample, nearby objects may be a source of
interference. If the magnitude of this noise causes inadequate
scans, then the user should be instructed on correct scan
behaviour (i.e., only scan one sample at a time). Furthermore,
this may also further encourage a stationary design, where
the object is placed on a sample holder. The ratio of the NIRS
waves that is reflected, absorbed, or passed through the
scanned object depends on the object characteristics. Fruits
[29], textiles [56], and pharmaceuticals [11] all have
different penetration properties. Furthermore, a secondary
object placed directly behind the sample might also reflect
some light and affect the resulting spectra. To inspect the
effects of this potential source of interference, we placed one
object to be scanned in an ideal position while a secondary
object was positioned directly behind the scanned object,
moving up to 5cm away from the primary object at
increments of 2mm.
Figure 6 shows how the distance of a secondary object to the
scanned object can affect the quality of the signal. For the
first and second graph, there was only one instance of
interference when the secondary object (i.e., apple, banana)
was placed directly behind the sample (i.e., cotton, grape).
Finally, the level of interference from cotton on the omega-3
decreased in magnitude from 0mm to 37mm. This is because
the omega-3 is transparent and therefore light is transmitted
through the object and hits the cotton. It is then reflected back
to the lens, causing the resulting spectra to be distorted.
Figure 6. Mean absorbance of objects over different distances
of interfering objects to the main object
Test+6:+Ambience!
The user may add noise to the spectrum by conducting scans
with the NIRS in a context with unsuitable levels of ambient
light, humidity, or temperature. We focus on light, as
humidity and temperature’s effect on the NIR spectra is
limited in non-extreme climates [55]. The user should be
instructed through the interface if the ambient light can cause
the scan quality to drop significantly.
Radiation from nearby light sources can affect the accuracy
of the NIRS [50]. The level of interference depends on the
magnitude of the illuminance and the distance between
scanner and sample. We use a 120W halogen lamp as an
interference source and a lux meter to measure its
illuminance. As the halogen lamp transmits on wavelengths
beyond the lux meter’s range, not all the light was detected.
However, we still obtained a good indication of the
illuminance level. We measured the effect of five different
lux levels: 500 (bright office), 800, 1200 (daylight), 1600,
and 2000 lux (shop window) [14]. The measurements were
taken at four distances: 0, 10, 20, and 30mm. The effect of
increased illuminance on SNR can be observed in Figure 7.
The graphs are divided into bins. Samples in the same bin
were taken at equal distance. Cotton and apple were
unaffected by higher amounts of lux at 0mm range, while
cotton showed varying results at 10, 20, and 30mm. For the

apple, the scan contained substantial noise when increasing
the lux beyond 0mm range. Omega-3 was affected more by
stray light than the other two samples. Readings became
noisy when the measured lux was above 500. This was
expected as omega-3 is transparent and the interfering light
can easily penetrate and cause noise.
Figure 7. Mean absorbance of objects over different lux levels
):$==-0E-(M(L()"*3=-N+,5("1(,:-(2$*3=+0E(O%"4-''(
In addition to the challenges associated with user-induced
errors when using NIRS, interaction with the scanning
device is another important aspect. To facilitate novice end-
user interaction with NIRS, a graphical user interface is
required [27]. Figure 8 shows a procedural overview of the
steps a user has to complete when using a NIRS device. The
first two columns contain the steps and their respective
descriptions. The third column indicates whether a user has
to conduct that step when using a default NIRS device.
Figure 8. Necessary steps a user needs to complete with a
typical NIRS device. Adapted from [26].
First, users have to operate the instrument through typically
complex software (and sometimes hardware) operations.
Current NIRS systems require users to configure the device,
set scan parameters, and navigate complex menus. Following
these steps, the user also has to handle information produced
by the device. A major challenge for non-experts is that data
is usually returned in a raw format, not revealing much
information to the end user. To extract the data and turn it
into understandable knowledge, analysis methods (i.e.,
chemometrics) have to be applied to the dataset. This
involves pre-processing the data when needed, and using
multivariate analysis to classify the information [34].
Furthermore, a reference library is required to serve as
training data for the classification model. The whole process
hinders non-experts without strong analytical capabilities to
interpret the information produced by the instrument.
Consequently, training and education is a prerequisite for
personnel wanting to utilise this technology in its present
state. Our work is a step towards a future where users are not
required to perform complex steps that require specific skills.
.>.LAPOA8!(622/2!6.)A(
To overcome the challenges related to scanning and
identifying samples, we adopt a combined hardware-
software approach. We address potential user-induced errors
and the procedural complexities through a 3D printed
enclosure and by guiding the user through a smartphone
application capable of automating the required analysis.
O:5'+4$=(6''+',$04-Q(A04="'9%-(
We designed a 3D printed enclosure to both protect the
scanner, reduce the effect of user-induced errors, and
facilitate the scanning process. The design of our enclosure
is informed by the findings of our tests. The design consists
of a modular approach (protective casing and two different
sample holders), which can be replaced by the user. The
sample holders are distinctive in both shape and size to allow
for different types of objects to be scanned. Figure 9 shows
the final iteration of the enclosure and sample holders.
Figure 9. Enclosure and sample holders. The smaller sample
holder contains two walls for small samples to be held in place.
The enclosure is formed to be positioned on the table, which
removes the chance for device motion noise during
operation. It is also pointing upwards, guaranteeing that
object placement is within close proximity of the lens. This
increases the accuracy as discovered in the sample distance
test. Furthermore, it also the reduces the interference from
ambient light, as the object is now covering the lens. The
holder ensures that items are by default positioned at an even
sample angle. In addition, our design prevents users from
accidentally placing another object behind the scanned
object, affecting the scan results (sample interference). Two
arrows were placed on each platform to guide the user as to
where the sample should be positioned.
The larger sample holder has a dimension of 20 x 20 cm. The
considerable surface area enables comfortable placement of
samples, without concern for the sample falling off. The
smaller sample holder consists of two small walls placed at
an incline, allowing for small samples to be held securely in
place. This addition can be seen more in detail in Figure 12.
The walls are covered using insulating tape with IR
absorbing characteristics, to reduce light scattering and to

shield from interference caused by any nearby objects. The
base of the walls are positioned at the edge of the lens, so that
any item positioned between the walls is covering the lens.
The final setup as described above is the result of three
iterations of design, 3D printing, and testing.
O%"4-&9%$=(6''+',$04-Q(2"1,#$%-(
We developed a smartphone application to allow end-users
to interact with the NIRS device. The application provides a
single interaction point. Furthermore, it provides the user
with instructions and warnings regarding scan quality issues
in order to reduce the effect of user-induced errors. The
software also automates and encapsulates all the complex
analysis stages, thus reducing the burden for the user. Figure
10 displays the various screens of the application.
Figure 10. Top row from left: Initial screen, settings, example
of scanning instructions. Bottom row from left: Example of
sample identification result, example of scan result warning
about the presence of a component, example of scan result
regarding the absence of a component.
The home screen allows users to adjust the settings or start a
new scan. In the settings screen, the user selects the type of
sample that will be scanned. Instead of having to adjust a
long list of parameters for the scan, the user can choose a pre-
determined category. Based on this configuration, the scan
screens will adjust to show the relevant instructions, guiding
the user in correctly scanning the sample. For example, the
text instructs the user to position an even surface down and
only scan one sample at a time. These guidelines were based
on our tests investigating user-induced errors. After a scan is
completed, the application displays the results. We utilise the
built-in lux sensor of the phone to infer the level of
illuminance in the area. Lux is measured in luminous flux per
square metre, so while the phone and sensor are not in the
exact same position, it can still be used as a baseline for how
much light is reaching the NIRS. The lux measurement (top
right corner of scan screen) changes from green, to yellow,
to red, depending on the level of illuminance based on our
ambient light tests.
When the user presses the scan button, the button becomes
disabled and a progress bar appears together with a text field
describing the stage (i.e., scanning, analysis, or processing).
The whole process takes around 20 seconds (approximately
10 seconds for scan, 5 seconds for analysis, and 5 seconds
for processing) to complete. Upon completion, the
application will display the results and also vocalise the
results using text-to-speech. The current version of the
application has the ability to identify pharmaceuticals and
detect whether a bread contains gluten. While the application
could be implemented for a vast number of scenarios, we
focus on the two aforementioned scenarios in our user study.
Communication (e.g., starting a scan, receiving scan results)
between the smartphone and the NIRS device occurs over
Bluetooth Low Energy. When the application receives the
scan results, they are sent to a server for processing (e.g.,
filtering, classification, regression). After the sample is
analysed, the results are sent back to the smartphone and
presented to the user. In addition, there are four scenarios in
which the application will display a warning message to the
user. While the warnings are active (10s), the scan button is
greyed out and disabled. This is to ensure that the user reads
it and does not accidently press the scan button again. After
ten seconds, the button becomes active again. Warnings are
intended to correct scan behaviour and limit the effect of
user-induced errors, and are displayed in Figure 11. The
thresholds for triggering these warnings were informed by
the user-induced errors tests.
Figure 11. Warnings 1 - 4 from top left to bottom right.
Warning 1. If insufficient light is reflected back to the
receiver, this is interpreted by the software as the object
being misplaced.

Warning 2. If substantial noise is detected in the resulting
spectra, this can be caused either by object transparency or
uneven surfaces.
Warning 3. When the illuminance is larger than 800 lux, the
text turns yellow and the user will be informed that the
accuracy of the scans might be inaccurate.
Warning 4. When the illuminance is larger than 1200 lux,
the text turns red and scanning is temporarily disabled until
the user can find a darker location.
AR$=9$,+"0(
To evaluate our proposed software and physical guides, we
conducted a user study with participants who had no prior
experience with NIRS technology. The main goal was to
investigate whether non-experts are able to successfully
conduct the scanning procedure without training. We
investigated how people interact with the technology through
a scanning experiment, and conducted a semi-structured
interview to capture their opinions and feedback. Thus, our
objective here is to explore the practical usability and
perceived usefulness of the developed solution.
Participants+and+Procedure+
We recruited 15 people using mailing lists of our university
(9 males, 6 females; ages: 20-34 years old, M=26.6).
Participants had a diverse range of educational backgrounds
(e.g., Finance, Biology, Anthropology, and Computer
Science). We carried out the experiments with each
participant individually. The experiment duration was
approximately 30 minutes per participant. Participants were
rewarded for their participation with a movie voucher. After
a short introduction to the experiment, participants were
asked to scan a set of objects placed on the table before them
using the NIRS scanner within the enclosure and a provided
mobile phone running our application. The study was
divided in two parts: 1) sample identification using 10
different pharmaceuticals, and 2) gluten detection using 10
pieces of bread. Both parts had items with varying
characteristics (e.g., shape, size, texture, and transparency)
to thoroughly test our non-expert assistance. We
counterbalanced the scanning order.
Participants did not receive any instructions on how to use
the devices, as we wanted to assess the usability of our
proposed solution in a realistic setting (e.g., the user would
encounter this device in a supermarket without additional
guidance from the staff). The participants were asked to
report the scan results to the observing researcher. During the
experiment, we recorded the participants’ answers and how
they interacted with the non-expert assistance, as well as any
comments from the participants while they scanned the
objects. In addition, we also recorded any warning messages
that were displayed by the application, for which object this
warning message was displayed, and the user’s reaction.
After all scans were completed, we conducted a semi-
structured interview in which we asked participants to
comment on the application (e.g., usage, instructions,
warnings, results), the enclosure, and the technology in
general. The interview was structured as follows.
Age, gender, occupation, background.
Statements ranked on a 5-point Likert scale.
–!The scan time was acceptable.
–!The instructions/warnings/results (asked individually) were
easy to understand.
–!The instructions/warnings (asked individually) had enough
details.
–!It was easy to understand where to place the object.
–!It was easy to place the object in the correct position.
–!I felt comfortable when I used the device.
–!I would trust this type of technology.
–!I can see myself using this technology daily.
Open-ended questions.
–!What is your overall experience using our solution?
–!Did you have any issues understanding how to use the
application?
–!Did you make any adjustments when you received the
warnings?
–!What steps would you take upon receiving a warning?
–!What is your impression of the accuracy of the device?
–!What type of information would you like to receive with the
result?
–!What features did you like on the enclosure?
–!What features did you not like on the enclosure?
–!What features would you change on the enclosure?
–!What would you scan with a device like this?
Since not all the warnings were shown during the scans, we
printed them on paper and showed them individually during
the interview in order to collect feedback. We asked
participants to envision receiving the warnings and
describing their subsequent actions. In addition, we asked the
participants what additional information they would like to
receive in scenarios besides the two we presented. The main
focus of the study was to investigate whether novice end-
users would be able to successfully utilise a NIRS device
using our design. We were also interested in collecting
opinions on our design and the technology in general.
Figure 12. Left: Object placement on small sample holder.
Right: Participant during the user study.
8-'9=,'(
Perceptions+
The majority of participants (N=14) immediately understood
the instructions and started using the application with no
notable problems. One participant (P07) tried scanning the
object by holding the phone towards it instead of placing the
object on the platform. The participant quickly corrected his
behaviour and explained that the confusion was due to
having used several applications that rely on the camera.
Participants strongly agreed that the instructions were easy

to understand (M = 4.53, SD = 1.06), and also agreed that the
instructions they were provided with by the smartphone were
of sufficient detail (M = 4.40, SD = 0.91). Some participants
commented that the instructions could be presented in a
stepwise fashion, allowing for a more legible presentation: “I
would like stepwise instructions, where you swipe to show
the next instructions. It would allow for bigger pictures and
text.” (P09). None of the participants reported having any
issues with using the application and generally agreed that
the scan time was acceptable (M = 3.80, SD = 0.68). Context
was also deemed an important factor: “A user in the
supermarket would not be able to wait that long” (P01).
Warnings+
A few warnings were caused by participants pressing the
scan button before positioning the sample, triggering
Warning 1. In these cases, the users quickly recovered. There
were also events in which transparent samples (omega-3)
generated Warning 2. We showed each individual warning
to the participants, and they generally agreed that they were
understandable (M = 4.34, SD = 0.80) and contained enough
details (M = 4.41, SD = 0.85). P04, P09, P11, and P12
commented that the icon for Warning 2 did not correlate with
the text in the warning. “The icon in Warning 2 does not
indicate transparency” (P04). P05 recommended more
visual instructions, instead of text. P06 requested more
information about the lux range and more distinct difference
for the icons in warning 3 and 4. Regarding the actions they
would take upon receiving a warning, the answers were in
line with the steps we envisioned a user would take (e.g.,
reposition object, place flat surface down, shade the machine
from interfering light, or move to a darker area).
Scan+Presentation+
Participants strongly agreed that the results were easy to
understand (M = 4.87, SD = 0.35). When questioned about
what additional information participants would like to have
included with the results, most answers were related to more
nutritional information (e.g., protein, carbohydrates, fats,
sugar) but also other allergens (lactose). We received
positive comments for the pharmaceutical results, such as “I
really like the vitamin feedback. It is cool that it tells you
what it contains. Reminds me of a video game, where you can
inspect items and see their stats” (P05).
Physical+Enclosure+
Multiple participants positively commented on the
enclosure, with comments such as, “It’s small, light, and
portable” (P04), and “The large surface area makes it easy
to position the bread” (P01). Participants strongly agreed
that it was easy to understand where to place the sample (M
= 4.67, SD = 0.72). “I liked the arrows and colour
indications on the box” (P05). They also agreed that it was
easy to place the item in the correct position (M = 4.20, SD
= 0.86). When asked about potential improvements to the
enclosure, a few remarks about aesthetics were noted.
“Would change the package a bit, make it less bulky and
more flat.” (P08). Another participant believed the scanning
process of smaller items could be improved with a redesign
of the enclosure, “Make the box so that you put the item
inside and scan it, and change the platform to be a bowl, so
the item falls to the middle.” (P10).
Future+Use+
The participants reported being comfortable when using the
device (M = 4.47, SD = 0.64), and the general consensus was
that they would use the technology on a daily basis (M =
4.47, SD = 0.74). In addition, participants stated they would
trust this type of technology (M = 4.53, SD = 0.64), "I think
it was accurate and made correct identifications. I trust this
machine more than my own judgement or some labels"
(P12). However, a couple of participants expressed the
importance of getting the scans right in certain situations,
“False classifications could potentially be dangerous for the
user. In those cases, accuracy is really important.” (P04).
When probed about potential scenarios where they would use
this technology, several participants mentioned food,
pharmaceuticals, textiles, and various chemicals. Three
participants proposed creative ways of using the technology:
“It would be interesting to scan makeup to look for micro
plastic. I don’t use makeup that can potentially damage the
nature.” (P05). “I would like to scan food products for
gelatine. We Muslims don’t eat products that contain it.”
(P12). “It would be nice to have a survival scanner that could
detect if food found in the nature (e.g., berries and
mushrooms) are poisonous or safe to eat.” (P13).
Typical information to look for could be nutritional content,
Active Pharmaceutical Ingredient (API) percentages,
allergens, and item quality. Multiple users would like to have
the machine in a supermarket, to scan products and retrieve
information about their quality (e.g., fruit ripeness). “Putting
it in the supermarket would be really helpful, I can see a lot
of people using it.” (P12). “It would be useful for dementia
patients. Pharmaceuticals can look the same, but have
strongly different effects.” (P04). Several participants
envisioned that in the future, NIRS could be a helpful sensor
embedded in their smartphones. Multiple participants also
expressed their desire to have such a device in their home for
a number of different reasons. Throughout the study, people
were enthusiastic about the technology, and commented
positively on its potential.
?/2)@22/>.(
!"#$%&'(A0&L@'-%(.-$%(/01%$%-&(23-4,%"'4"35((
Previous work on portable NIRS has mainly been concerned
with testing performance under ideal conditions with domain
experts [2,28,39]. However, there is a scarcity of research
that highlights the potential use cases, design challenges, and
solutions to enable non-experts to use NIRS. The effect of
user-induced errors on NIRS can pose significant challenges,
particularly for untrained users. Users currently have to
operate the instrument through complex software and require
knowledge about advanced analytical methods to understand
the result. We argue that in order to make the technology
more accessible and still produce reliable results, the scanner

and its software/hardware controls must be designed to guide
the user and avoid problematic scanning conditions. In this
paper, we report and evaluate a hardware-software design
that proved successful in guiding and facilitating the
scanning process for non-expert end users.
Our application provided simplified instructions for users
with no previous contact with the technology. The warnings
were designed to help users overcome probable scanning
errors, and were shown to help them overcome sample
misplacement. Our design effectively provides an interactive
way to train users of the equipment, and our user study shows
that participants were able to quickly start using the
technology reliably without any verbal instructions.
Furthermore, the 3D printed enclosure and sample holder
enabled, for the most part, correct scanning behaviour by
facilitating optimal object placement. Participants felt that
the enclosure effectively guided them in correctly placing the
samples, while at the same time protecting the device. One
element of the scanning process that was rated slightly lower
than others was the scanning time (M = 3.80, SD = 0.68).
One potential reason for this is that the technology is quite
new - it is possible that users have not yet formed an opinion
on how long a scan should take. The scan time depends on
the scan configuration (e.g., resolution, width, and SNR). By
increasing the precision or reliability of the scan, it takes a
longer time to complete. Some configurations are more
appropriate, depending on the nature of the analysis and the
scanning environment. Ultimately, what constitutes a
feasible and acceptable scan time will depend on the scenario
(e.g., supermarket, home) and the analysis requirements
(e.g., allergens, API, nutrition, ripeness, poisonous).
Moreover, one important challenge for end-user NIRS is the
collection of reference scans. With our system, users have to
indicate the type of objects that they are scanning so that the
application can optimise the scanning configuration. This
then loads the appropriate model depending on the user’s
choice. We envision that in the future, devices will come pre-
loaded with models for commonly scanned objects. In
addition, crowdsensing communities [24] can emerge to
establish a shared repository of sample fingerprints to
facilitate object identification. Furthermore, methods that
enable use of existing knowledge-bases built by benchtop
instruments can be used with acceptable performance [21].
/*3=+4$,+"0'(1"%(O%$4,+4-($0&(8-'-$%4:(
We tested a relatively cheap and miniaturised NIRS device
on a wide set of everyday objects. We have shown how the
device performs when scanning various objects with
different physical characteristics. By better understanding
these parameters, it is possible to design enclosures and
applications that facilitate the correct usage of the
technology. This would allow for a smartphone to be turned
into an advanced scientific instrument by simply installing
an application and coupling it to a NIRS device. As a result,
even novice users would be able to conduct experiments,
earlier limited only to trained lab personnel. For instance,
farmers can be better informed on the best time to harvest
their crops without relying on expensive or time-consuming
methods. About 125,000 people die every year due to
medication mismanagement and the estimated cost is around
$300 billion [25]. Miniaturised NIRS could be implemented
to help nurses or the users themselves (e.g., old adults)
administer medicine. Furthermore, a number of interesting
use cases that we did not envision were identified by our
participants, highlighting the potential of the technology for
a wide variety of everyday scenarios. For researchers, our
study opens multiple avenues to explore. Due to the small
form factor, our solution can be carried around by the user
for in situ sample analysis. Furthermore, a multitude of
everyday objects, from cups and dishes to shelves and
refrigerators can begin to be instrumented with NIRS
hardware to identify objects. Ubiquitous scanning stations
where the NIRS is bundled with tablets could be installed in
public locations and serve a broad group of users. As NIRS
devices decrease in size and price, it may become possible to
bind them to smartphones in the future. This could create a
range of opportunities for scanning “in-the-wild” without
having to carry around an additional device.
B+*+,$,+"0'(
The work presented in this paper has several limitations.
First, we conducted the user-induced errors experiments
using visual alignment. While both careful precision and
measuring tools were used to align the samples, there may
still exist some human error. Second, we did not compare
participant usage of the technology with and without our
solution. It would be unreasonable to give the hardware with
its default elements and expect a participant to be able to
conduct any given step of the scan and analysis process.
Also, discussion about classification models is considered to
be out of scope for this paper. Finally, we only utilise one
NIRS scanner in this paper. While our findings can be used
as guidelines for a larger array of hardware, it would be
interesting to test multiple devices from different vendors for
a more thorough investigation.
)>.)B@2/>.(
Our work has systematically investigated the effect of user-
induced errors and the complexity of the sampling process in
the context of an affordable and miniaturised NIRS. We
subsequently investigated the magnitude of these parameters
to derive their potential impact on scanning quality and
usability. Previous work has taken for granted that the
instrument is used under ideal conditions by trained
personnel. We show that NIRS can be successfully used by
novice end-users, if the challenges explored in this paper are
taken into consideration. This is validated through a user
study where non-expert users test a NIRS equipped with a
3D printed enclosure and smartphone application. The
results indicate that the accuracy and user experience when
using NIRS for object detection and analysis is adequate
when use is facilitated with our proposed non-expert
assistance. In our ongoing work, we intend to deploy an in
situ scanner in a variety of different scenarios.

8AIA8A.)A2(
1. Nawaf Abu-Khalaf. 2015. Sensing tomatos pathogen
using Visible/Near infrared (VIS/NIR) spectroscopy
and multivariate data analysis (MVDA). Palestine
Technical University Research Journal 3, 1: 12-22.
2. T. Akinaga, Munehiro Tanaka and Sheishi Kawasaki.
2008. On-tree and after-harvesting evaluation of
firmness, color and lycopene content of tomato fruit
using portable NIR spectroscopy. Journal of food,
agriculture & environment.
3. Dimitris Alexandrakis, Gerard Downey and Amalia G.
Scannell. 2012. Rapid non-destructive detection of
spoilage of intact chicken breast muscle using near-
infrared and Fourier transform mid-infrared
spectroscopy and multivariate statistics. Food and
Bioprocess Technology 5, 1: 338-347.
4. Katherine A. Bakeev. 2010. Process analytical
technology: spectroscopic tools and implementation
strategies for the chemical and pharmaceutical
industries. John Wiley & Sons.
5. Ravi Bhoraskar, Nagamanoj Vankadhara, Bhaskaran
Raman and Purushottam Kulkarni. 2012. Wolverine:
Traffic and road condition estimation using smartphone
sensors. In 2012 Fourth International Conference on
Communication Systems and Networks (COMSNETS
2012), 1-6.
6. M. Blanco and I. Villarroya. 2002. NIR spectroscopy: a
rapid-response analytical tool. TrAC Trends in
Analytical Chemistry 21, 4: 240-250.
7. L. Bokobza. 1998. Near infrared spectroscopy. Journal
of Near Infrared Spectroscopy 6: 3-18.
8. Emmanuel J. Candès and Michael B. Wakin. 2008. An
introduction to compressive sampling. IEEE signal
processing magazine 25, 2: 21-30.
9. A. Candolfi, R. De Maesschalck, D. L. Massart, P. A.
Hailey and A. C. E. Harrington. 1999. Identification of
pharmaceutical excipients using NIR spectroscopy and
SIMCA. Journal of pharmaceutical and biomedical
analysis 19, 6: 923-935.
10. Zhenghua Chen, Han Zou, Hao Jiang, Qingchang Zhu,
Yeng C. Soh and Lihua Xie. 2015. Fusion of WiFi,
smartphone sensors and landmarks using the Kalman
filter for indoor localization. Sensors 15, 1: 715-732.
11. Fiona C. Clarke, Stephen V. Hammond, Roger D. Jee
and Anthony C. Moffat. 2002. Determination of the
information depth and sample size for the analysis of
pharmaceutical materials using reflectance near-
infrared microscopy. Applied spectroscopy 56, 11:
1475-1483.
12. E. Cleve, E. Bach and E. Schollmeyer. 2000. Using
chemometric methods and NIR spectrophotometry in
the textile industry. Analytica Chimica Acta 420, 2:
163-167.
13. Anil K. Deisingh. 2005. Pharmaceutical counterfeiting.
Analyst 130, 3: 271-279.
14. David L. DiLaura, Kevin W. Houser, Richard G.
Mistrick and Gary R. Steffy. 2011. The lighting
handbook: Reference and application.
15. DLP® NIRscanTM Nano EVM User's Guide.
Retrieved 21/03/2016 from
http://www.ti.com/lit/ug/dlpu030f/dlpu030f.pdf
16. A. Durand, O. Devos, C. Ruckebusch and J. P.
Huvenne. 2007. Genetic algorithm optimisation
combined with partial least squares regression and
mutual information variable selection procedures in
near-infrared quantitative analysis of cotton-viscose
textiles. Analytica Chimica Acta 595, 1: 72-79.
17. M. Dyrby, S. B. Engelsen, L. Nørgaard, M. Bruhn and
L. Lundsberg-Nielsen. 2002. Chemometric quantitation
of the active substance (containing ≡ N) in a
pharmaceutical tablet using near-infrared (NIR)
transmittance and NIR FT-Raman spectra. Applied
Spectroscopy 56, 5: 579-585.
18. Christopher D. Elvidge, David M. Keith, Benjamin T.
Tuttle and Kimberly E. Baugh. 2010. Spectral
identification of lighting type and character. Sensors
10, 4: 3961-3988.
19. Ming Gao, Xihong Hu, Bo Cao and Dianxin Li. 2014.
Fingerprint sensors in mobile devices. In 2014 9th
IEEE Conference on Industrial Electronics and
Applications, 1437-1440.
20. Mayank Goel, Eric Whitmire, Alex Mariakakis, T. S.
Saponas, Neel Joshi, Dan Morris, Brian Guenter,
Marcel Gavriliu, Gaetano Borriello and Shwetak N.
Patel. 2015. HyperCam: hyperspectral imaging for
ubiquitous computing applications. In Proceedings of
the 2015 ACM International Joint Conference on
Pervasive and Ubiquitous Computing, 145-156.
21. C. V. Greensill and K. B. Walsh. 2000. Optimization of
Instrumentation Precision and Wavelength Resolution
for the Performance of NIR Calibrations of Sucrose in
a Water-Cellulose Matrix. Applied Spectroscopy 54, 3:
426-430.
22. Mathew J. Gregoski, Martina Mueller, Alexey
Vertegel, Aleksey Shaporev, Brenda B. Jackson, Ronja
M. Frenzel, Sara M. Sprehn and Frank A. Treiber.
2012. Development and validation of a smartphone
heart rate acquisition application for health promotion
and wellness telehealth applications. International
journal of telemedicine and applications 2012: 1.
23. G. M. Hieftje. 1972. Signal-to-noise enhancement
through instrumental techniques. II. Signal averaging,
boxcar integration, and correlation techniques.
Analytical chemistry 44, 7: 69A-78a.
24. Simo Hosio, Jorge Goncalves, Niels van Berkel and
Simon Klakegg. 2016. Crowdsourcing Situated &

Subjective Knowledge for Decision Support. In
International Joint Conference on Pervasive and
Ubiquitous Computing Adjunct, ACM, 1478-1483.
http://dx.doi.org/10.1145/2968219.2968591.
25. Is Mom Taking Her Medicine? An Entrepreneur Takes
On A $300 Billion, Life-And-Death Challenge.
Retrieved 02/01/2017 from
http://www.forbes.com/sites/elizabethmacbride/2016/0
9/27/is-mom-taking-her-medicine-an-entrepreneur-
takes-on-a-300-million-life-and-death-
challenge/#210eb96054b3
26. Meltem Izzetoglu, Ajit Devaraj, Scott Bunce and Banu
Onaral. 2005. Motion artifact cancellation in NIR
spectroscopy using Wiener filtering. IEEE
Transactions on Biomedical Engineering 52, 5: 934-
938.
27. Simon Klakegg, Chu Luo, Jorge Goncalves, Simo
Hosio and Vassilis Kostakos. 2016. Instrumenting
Smartphones with Portable NIRS. In International
Joint Conference on Pervasive and Ubiquitous
Computing Adjunct, ACM, 618-623.
http://dx.doi.org/10.1145/2968219.2971590 .
28. Jeroen Lammertyn, Bart Nicola, Kristine Ooms, V. De
Smedt and Josse De Baerdemaeker. 1998. Non-
destructive measurement of acidity, soluble solids, and
firmness of Jonagold apples using NIR-spectroscopy.
Transactions of the ASAE 41, 4: 1089.
29. Jeroen Lammertyn, Ann Peirs, Josse De Baerdemaeker
and Bart Nicola. 2000. Light penetration properties of
NIR radiation in fruit with respect to non-destructive
quality assessment. Postharvest Biology and
Technology 18, 2: 121-132.
30. Kristiyan Lukanov, Horia A. Maior and Max L.
Wilson. 2016. Using fNIRS in usability testing:
understanding the effect of web form layout on mental
workload. In Proceedings of the 2016 CHI Conference
on Human Factors in Computing Systems, 4011-4016.
31. Chu Luo, Angelos Fylakis, Juha Partala, Simon
Klakegg, Jorge Goncalves, Kaitai Liang, Tapio
Seppänen and Vassilis Kostakos. 2016. A Data Hiding
Approach for Sensitive Smartphone Data. In
International Joint Conference on Pervasive and
Ubiquitous Computing Adjunct, ACM, 457-468.
http://dx.doi.org/10.1145/2971648.2971686.
32. David Luxton, Russell A. McCann, Nigel E. Bush,
Matthew C. Mishkind and Greg M. Reger. 2011.
mHealth for mental health: Integrating smartphone
technology in behavioral healthcare. Professional
Psychology: Research and Practice 42, 6: 505.
33. Horia A. Maior, Matthew Pike, Sarah Sharples and
Max L. Wilson. 2015. Examining the reliability of
using fNIRS in realistic HCI settings for spatial and
verbal tasks. In Proceedings of the 33rd Annual ACM
Conference on Human Factors in Computing Systems,
3039-3042.
34. Desiré L. Massart, B. G. M. Vandeginste, S. N.
Deming, Y. Michotte and L. Kaufman. 1988.
Chemometrics: a textbook.
35. D. A. McNamara, C. W. I. Pistorius and J. A. G.
Malherbe. 1990. The Uniform Geometrical theory of
diffraction. Artech House, London.
36. Shree K. Nayar, Katsushi Ikeuchi and Takeo Kanade.
1991. Surface reflection: physical and geometrical
perspectives. IEEE Transactions on Pattern Analysis &
Machine Intelligence 7: 611-634.
37. Bart M. Nicolai, Katrien Beullens, Els Bobelyn, Ann
Peirs, Wouter Saeys, Karen I. Theron and Jeroen
Lammertyn. 2007. Nondestructive measurement of
fruit and vegetable quality by means of NIR
spectroscopy: A review. Postharvest Biology and
Technology 46, 2: 99-118.
38. Brian G. Osborne and Thomas Fearn. 1986. Near
infrared spectroscopy in food analysis. Longman.
39. Andrew J. ONeil, Roger D. Jee, Gad Lee, Andrew
Charvill and Anthony C. Moffat. 2008. Use of a
portable near infrared spectrometer for the
authentication of tablets and the detection of
counterfeit versions. J. Near Infrared Spectrosc 16:
327.
40. Evan M. Peck, Daniel Afergan, Beste F. Yuksel,
Francine Lalooses and Robert J. Jacob. 2014. Using
fNIRS to measure mental workload in the real world.
Advances in physiological computing: 117-139.
41. Ling Pei, Robert Guinness, Ruizhi Chen, Jingbin Liu,
Heidi Kuusniemi, Yuwei Chen, Liang Chen and Jyrki
Kaistinen. 2013. Human behavior cognition using
smartphone sensors. Sensors (Basel) 13, 2: 1402-24.
http://dx.doi.org/10.3390/s130201402.
42. Quarter of the world will be using smartphones in
2016. Retrieved 29/08/2016 from
http://www.telegraph.co.uk/technology/mobile-
phones/11287659/Quarter-of-the-world-will-be-using-
smartphones-in-2016.html
43. Kiran B. Raja, Ramachandra Raghavendra, Vinay K.
Vemuri and Christoph Busch. 2015. Smartphone based
visible iris recognition using deep sparse filtering.
Pattern Recognition Letters 57: 33-42.
44. Theodore S. Rappaport. 1996. Wireless
communications: principles and practice. Prentice Hall
PTR New Jersey.
45. Sirinnapa Saranwong, Jinda Sornsrivichai and Sumio
Kawano. 2004. Prediction of ripe-stage eating quality
of mango fruit from its harvest quality measured
nondestructively by near infrared spectroscopy.
Postharvest Biology and Technology 31, 2: 137-145.

46. T. W. L. Scheeren, P. Schober and L. A. Schwarte.
2012. Monitoring tissue oxygenation by near infrared
spectroscopy (NIRS): background and current
applications. Journal of clinical monitoring and
computing 26, 4: 279-287.
47. Sensors Overview | Android Developers. Retrieved
04/03/2014 from
http://developer.android.com/guide/topics/sensors/sens
ors_overview.html
48. Heinz W. Siesler, Yukihiro Ozaki, Satoshi Kawata and
H. M. Heise. 2008. Near-infrared spectroscopy:
principles, instruments, applications. John Wiley &
Sons.
49. Nicoletta Sinelli, Sara Limbo, Luisa Torri, Valentina
Di Egidio and Ernestina Casiraghi. 2010. Evaluation of
freshness decay of minced beef stored in high-oxygen
modified atmosphere packaged at different
temperatures using NIR and MIR spectroscopy. Meat
science 86, 3: 748-752.
50. Walter Slavin. 1963. Stray Light in Ultraviolet, Visible,
and Near-Infrared Spectrophotometry. Analytical
Chemistry 35, 4: 561-566.
51. Erin T. Solovey, Audrey Girouard, Krysta Chauncey,
Leanne M. Hirshfield, Angelo Sassaroli, Feng Zheng,
Sergio Fantini and Robert J. Jacob. 2009. Using fNIRS
brain sensing in realistic HCI settings: experiments and
guidelines. In Proceedings of the 22nd annual ACM
symposium on User interface software and technology,
157-166.
52. Megan Strait, Cody Canning and Matthias Scheutz.
2014. Reliability of NIRS-Based BCIs: a placebo-
controlled replication and reanalysis of Brainput. In
CHI'14 Extended Abstracts on Human Factors in
Computing Systems, 619-630.
53. Patrick L. Walter. 2006. The History of the
Accelerometer: 1920s-1996 - Prologue and Epilogue.
Sound & Vibration 41, 84-90.
54. Beste F. Yuksel, Kurt B. Oleson, Lane Harrison, Evan
M. Peck, Daniel Afergan, Remco Chang and Robert J.
Jacob. 2016. Learn piano with BACh: An adaptive
learning interface that adjusts task difficulty based on
brain state. In Proceedings of the 2016 CHI Conference
on Human Factors in Computing Systems, 5372-5384.
55. Ying Zhou, Xiaping Fu and Yibin Ying. 2006.
Influence of humidity on spectral performance for
near-infrared detection of fruit. In Optics East 2006,
638119-638119. http://dx.doi.org/10.1117/12.686542.
56. M. Zoccola, N. Lu, R. Mossotti, R. Innocenti and A.
Montarsolo. 2013. Identification of wool, cashmere,
yak, and angora rabbit fibers and quantitative
determination of wool and cashmere in blend: a near
infrared spectroscopy study. Fibers and Polymers 14,
8: 1283-1289.