Content uploaded by Manuel B. Garcia
Author content
All content in this area was uploaded by Manuel B. Garcia on Dec 01, 2019
Content may be subject to copyright.
978-1-7281-3044-6/19/$31.00 ©2019 IEEE
Tomayto, Tomahto: A Machine Learning Approach
for Tomato Ripening Stage Identification Using
Pixel-Based Color Image Classification
Manuel B. Garcia, Shaneth Ambat, Rossana T. Adao
College of Computer Studies
Institute of Technology, Far Eastern University
Manila, Philippines
hi@manuelgarcia.info
Abstract—The main enterprise of the Philippine agriculture
sector is crop cultivation where tomato is deliberated as one of
the major crops in the country. With the abundance on tomato
production, ripeness classification becomes fairly laborious and
challenging, not to mention the subjective visual interpretation
of human graders grounded from practical experience that is
easily influenced by the environment and prone to error. Thus,
this study proposes an automatic tomato ripeness identification
using Support Vector Machine (SVM) classifier and CIELab
color space via a machine learning approach. Dataset used for
modeling and validation experiment in a 5-fold cross-validation
strategy was composed of 900 images assembled from a farm
and various image search engines. Divided into six classes that
represent tomato ripening stages, experimental results showed
that the proposed method was successful with 83.39% accuracy
in ripeness classification detection. With this machine learning
approach and combination of image processing techniques, the
agriculture industry could benefit by automating the ripeness
estimation which then could save tomatoes from damage.
Keywords— Image Processing, Image Classification, Support
Vector Machine, Machine Learning, CIELAB color space
I. INTRODUCTION
Despite the transformation to an industrialized economy,
Philippines is still primarily an agricultural country [1] where
the gross value of agricultural production is amounted to PhP
429.7 billion during the first quarter of 2019 [2]. Employing
39.8% of the country’s labor force [3], the main enterprise of
the agriculture sector of the Philippines is crop cultivation [4]
such as tomatoes, rice, corn, coconut, sugarcane, pineapple,
mango, banana, and abaca, just to name a few. According to
Philippine Statistics Authority, tomatoes are one of the major
crops in the country [5]. In fact, tomato production reached
95.30 thousand metric tons from January to March 2019 [6].
The tomato (Solanum lycopersicum), known as “Kamatis” in
the Philippines, has attracted the interest of food markets due
to its medicinal impact such as decreasing the risk of various
health conditions such as cancer, cardiovascular disease, and
osteoporosis, and nutritive values such as phosphorus, iron,
calcium, and vitamin C [7]. Lycopene, the major carotenoid in
the fruit, is only amassed during the final ripening stage which
accounts for 80% of the total carotenoid content [8]. Hence,
tomato ripeness estimation has been viewed as a vital process
that influences its quality evaluation. In fact, consumers use
ripeness as a strand of what defines quality fresh fruits, which
is grounded from its visual appearance such as color, size, and
shape. Fruits color skin is not only an important determinant
of fruit selling price, but also as a feature of fruit ripeness [9]
for numerous agricultural products such as apples [10, 11],
bananas [12, 13], lime [14], mangoes [15-17], tomatoes [18,
19], and watermelons [20]. The abundance of harvest made it
challenging to consistently determine tomato ripeness, which
then becomes a problem when exported to a far place. Besides,
farmers and human graders are usually subjective in ripeness
visual interpretation grounding it from practical experience
and/or ripeness classification charts, and easily influenced by
the environment and prone to mistakes [21]. Thus, tomatoes
are either unripe or overripe when they arrive at the market.
This is the primary reason why tomatoes are often harvested
during the “green” stage (See Fig. 1) to endure transportation
[22]. Nevertheless, consumers are still less likely to purchase
tomatoes when they are not on the “red” stage of ripening.
Fig. 1. Maturity and Ripening Stages of Tomatoes Based on United States
Standards for Grades of Fresh Tomatoes.
The unceasing development of digital image processing,
computer vision, and machine learning have paved a way for
agriculture industry to further improve quality inspection and
defect sorting [23] of fruits [10-20]. In case of tomatoes, the
identification of ripening stage is achievable to categorize by
computers using physical parameters like color since there is
a positive correlation between color and ripeness of tomatoes
[24]. Thus, this study proposes an automatic tomato ripeness
identification by means of Support Vector Machine (SVM)
classifier and CIELAB (L*a*b*) color space via a machine
learning approach. Castro et al. [25] acknowledged SVM
classifier as the best machine learning technique compared to
K-nearest neighbor, artificial neural network, and decision
tree algorithms, and L*a*b compared to RGB and HSV as the
best color space for ripeness level classification of Cape
gooseberry; hence, the usage of SVM and L*a*b* color space
for this study. The succeeding sections of the paper covered
the review of existing research on ripeness classification of
fruits and vegetables (Section 2), the materials and methods
describing the logistics and phases of the development of the
proposed tomato ripeness classification (Section 3), results
from the experimental evaluations and its discussion (Section
4), and the conclusion and future research works (Section 5).
II. RELATED WORKS
Xiaobo, Jiewen, and Yanxiao [10], and Cárdenas-Pérez et
al. [11] developed a computer vision system to identify the
classification of apple maturity based on its color parameters.
The first study examined three hundred and eighteen apples
by using organization feature parameter by genetic algorithm
which is expressed by the equation of feature parameter (1).
The next study, on the other hand, analyzed one hundred and
fourteen apples by calculating the apple color variances (ΔΕ)
on CIELab color space using equation (2). Both studies were
successful on classifying apple ripeness level using their own
techniques where Cárdenas-Pérez et al. received a hundred
percent apple ripeness classification accuracy while Xiaobo,
Jiewen, and Yanxiao’s method was more accurate than back-
propagation artificial neural network (BP-ANN).
[26] x
√
p1 + p3
√
(ΔL*) 2 + (Δa*) 2 + (Δb*) 2
On the other hand, Mendoza and Aguilera [12], as well as
Paulraj et al. [13] used bananas for their image classification.
The former used RGB color space to investigate 100 ripe and
116 unripe bananas and classify the ripeness using Artificial
Neural Network while the latter used L*a*b* color space to
examine 49 banana samples and classify it according to the 7
ripening stages by instigating computer vision. Interestingly,
Mendoza and Aguilera performed more progressive analyses
such as brown spot, image texture, and chemical analysis for
a more accurate classification model. Perhaps, Mendoza and
Aguilera’s model was more accurate (98%) than the model
developed by Paulraj et al. (96%) because of these additional
analyses. Brown spot analysis was evaluated from binarized
images of banana peel where the number of brown spot was
identified from the a* color which signifies brown spots. The
texture analysis focused on segmenting the grayscale version
of the sample using the average of 4 directions that extracted
four textual features. Lastly, chemical analysis was done to
determine total soluble solids using a digital refractometer.
Furthermore, similar and more methodologies of machine
learning and image processing were performed on mangoes.
First, Vélez-Rivera et al. [17] utilized L*a*b* color space as
well but mixed with HSB. Moreover, Manila mangoes were
inspected through its physicochemical properties to estimate
its ripening index (RPI). Zheng and Lu [15] utilized CIELAB
color space to examine mangoes as well but the classification
was performed using a least-squares support vector machine
(LS-SVM) classifier. Fractal analysis was the foundation of
LS-SVM since it has been successfully used on classification
[27], appearance characterization [28], and quality prediction
[29, 30] of foods. Nandi, Tudu, and Koley [16] utilized SVM
as well but persisted on using RGB instead of converting to
other color space to avoid additional calculations. Moreover,
the classification performance was calculated and quantified
using k-fold (k=6) cross-validation technique via sensitivity,
specificity, predictivity, and accuracy measures.
Red-Green = mean(I(:,:,1)) / mean (I(:,:,2))
Red-Green = mean(I(:,:,1)) - mean (I(:,:,2))
The ripeness classification of tomatoes has also been part
of the proposed systems that utilized computer vision, digital
image processing, and machine learning algorithms. Polder,
van der Heijden, and Young [18] did compare RGB images
with hyperspectral images of tomato to classify its ripeness.
Fisher’s linear discriminant analysis (LDA) was used as the
classification technique which is normally implemented for
spectroscopic image classification. An imaging spectrograph
with a spectral range of 396 to 736 nm and a 13 µm slit size,
was used in the experiment to obtain spectra. Based from the
findings of the study, RGB images showed substantial errors
when classifying images with little feature variance – another
reason why this study used other color space. RGB was used
by Goel and Sehgal [19] as well but they had more success
with it by using red-green color difference instead of separate
red, green, and blue values. To obtain the red-green ratio and
red-green difference, equations (3) and (4) were used. As far
as the researchers are concerned, they proposed this approach
because existing literature on tomato ripeness classification
did not take into consideration the light spots (highlights) on
tomato surface brought by natural environment illumination
[26]. To minimize such case on this study, a computer vision
system was setup as shown on Fig. 2. Nevertheless, the white
colored pixels were interpolated by the ‘tomato colors’ from
its neighboring segments during background removal. Taofik
et al. [31] claimed to have a new data acquisition approach in
this kind of classification quandary. On developing a model
for their ‘smart system’ designed to detect ripeness of tomato
and chili, the image dataset was taken periodically, at 65, 75,
83, and 90 days of planting. Yet, this kind of data acquisition
approach was not elaborated in terms of how it helped to
increase the classification accuracy. Arakeri and Lakhsmana
[21] developed an automated grading system as well with the
help of a fruit handling system that was used for moving the
tomatoes on the conveyor belt. With the integration of image
processing, tomatoes are moved to the respective bins based
from its classification: defective or non-defective, and ripe or
unripe. The capacity of the machine is 300 tomatoes per hour
which is a lot of development on task automation. However,
further works are still needed to increase the speed, even the
accuracy, especially for image with high specular reflection.
LOOKING FOR THE FULL PAPER?
manuelgarcia.info/publication/tomayto-tomahto
III. MATERIALS AND METHODS
A. Experimental Setup for Computer Vision System
The image acquisition system for collecting the dataset of
tomato images, as an addition for pictures downloaded from
image search engines, was a custom-built photography studio
designed for food photo shoot. Figure 2 shows this computer
vision system (CVS) and the setup of the lighting and studio
equipment which was arranged in such a way that the image
of tomatoes could avoid unwanted illumination [26]. Instead
of a green screen backdrop, a blue screen backdrop (Westcott
131 Wrinkle-Resistant Chroma-Key Backdrop) was used due
to the color features of tomato that includes green color. Two
45-watts light bulbs (R20 Incandescent Flood Light 2700K
E26 Base) with diffuser (24x36” Softbox Bowens Speedring
Kit) were placed at an angle of 45 degrees for both sides to
ensure uniform illumination system. Tomatoes are placed on
a 24x29 inches table wrapped with blue paper to complement
the blue screen backdrop. A digital camera (Nikon D3100)
was used to capture tomato images using both automatic and
manual settings (f = 3.6; speed = 1/60 or 1/125; no zoom; no
flash). The maximum resolution of 14.2 megapixels (4,608 x
3,072) was used to store JPEG images in a desktop computer
(Core i5-9600k, DDR3, 1600MHz, 1TB hard drive, 256GB
SSD, 8GB RAM) connected to the camera via a USB cable.
Fig. 2. Computer Vision System for Collecting Tomato Dataset.
B. Preparation of Tomato Images as a Dataset
1) Collection of Tomato Samples
Two image sources of tomato were used to assemble the
dataset: image search engines and CVS. Google images and
Adobe Stock were utilized for downloading images from the
Internet. For CVS, a sample of tomato fruits was collected
from Marulas Public Market (14.6738° N, 120.9837° E) for
four consecutive weeks (50-100 tomatoes/week) sold by a
vendor who imports from a plantation on Laguna. The fruits
were then manually classified according to its ripeness stage,
and stored inside 250C container storages [32]. Image search
engines and CVS combined, the total number of sample was
963 tomatoes with different colors and ripeness levels.
C = Fg.*mask + Bg.*(~mask)
2) Preprocessing of Tomato Images
Following the findings of Castro et al. [25] that revealed
L*a*b color space as the recommended option to work with
fruit classification when compared to RGB and HSV, tomato
samples were converted first from RGB to L*a*b color space
(rgb2lab in Matlab). Wu and Sun [33] also argued that this
color space provides uniform color distribution which makes
it appropriate for food color measurement. The use of L*a*b
color space, where L* (black to white) is the luminance and
a* (green to red) and b* (blue to yellow) are the chromatic
components, is also considered essential because it matches
the colors as perceived by human eyes. After the color space
conversion, the digital images were pre-processed using the
techniques employed by Garcia et al. [34] before segmenting
skin color pixels. First, given the large size of images (4,608
x 3,072), each sample was scaled to 1/8th of its image size to
speed up the calculations. Next, each sample taken by CVS
was applied with a modified Chroma-key method inspired by
Sang and Vinh’s technique [35] grounded on coarse and fine
filter. Using this processing method, the Chroma-key effect
was performed on both foreground and background of image
using equation (5). On the other hand, downloaded tomato
images underwent histogram equalization, noise reduction,
lighting correction, and sharpening [34] as the preprocessing
techniques before going to image segmentation stage. These
were performed in order to have a more accurate and smooth
mask (see Figure 3 1b, and 2b) for removing the background.
3) Image Segmentation
For further analysis and understanding of digital images
[36] like fruit ripeness classification, image segmentation is
customarily a fixed image stage in order to extract a feature or
an object of interest which is performed by thresholding,
boundary detection or region dependent techniques. Among
the aforesaid methods, the simplest and most widely used in
image segmentation is thresholding [37]. The techniques in
performing thresholding is classified as global (traditional,
iterative, and multistage thresholding) and local (Niblack,
Sauvola, Bernsen, and Yanowitz and Bruckstein’s Method)
[38]. Niblack thresholding algorithm, eq. (6), was validated as
the better approach at removal of background noise [38].
Hence, tomato image dataset was binarized using Niblack
technique to produce better segmentation results. Because of
complex and sometimes undistinguishable background and
tomato color, the binary-segmented images from CVS (Fig. 3,
2b) were smoother, clearer, and more precise when compared
to downloaded images (Fig. 3, 1b). Specular reflection or glare
on the tomato surface due to natural lighting conditions was
already deciphered by Goel and Sehgal [19] by means of
interpolating the neighboring tomato color pixels. As such, the
same method was performed for downloaded images. In the
case of images from CVS, this was not an issue because the
setup of lighting condition was controlled. The resulting
binarized images were secluded from the background pixels
to eliminate unnecessary and similar colors with tomatoes as
these colors could only confuse the classification model and
lower the accuracy result. The pixels for tomato were shown
in Figure 3 1c and 2c which evidently captured not only the
shape feature of the fruit but also its color that will be used for
building the tomato ripeness classification model.
T(x, y) = m(x, y) + k*(x,y)
LOOKING FOR THE FULL PAPER?
manuelgarcia.info/publication/tomayto-tomahto
(1a) Image from Search Engine
(1b) Masked Image Using Threshold
(1c) Segmentation Results
(2a) Original Image from CVS
(2b) Masked Image Using Chroma Key
(2c) Segmentation Results
Fig. 3. Background Removal of Tomatoes Captured by CVS and Downloaded from Image Search Engines.
C. Classification of Tomato Ripeness Level
Features of tomatoes extracted from the previous process
were classified according to ripeness level based from United
States standards for grades of fresh tomatoes [39]. As shown
on Figure 1, the color classification that indicates the stage of
tomato maturity of a red fleshed variety of tomatoes includes
six ripeness levels: green, breakers, turning, pink, light red,
and red. At any ripeness stage, however, tomatoes normally
have a mix and color gradient (instead of single solid color)
between the neighboring levels. As such, the model accuracy
of any color-based image classification could suffer and may
produce incorrect predictions. Amirulah, Mokji, and Ibrahim
[40] experienced the same dispute for starfruit color maturity
classification and proposed a solution by overlapping the hue
of adjacent levels where Ci is the fruit ripeness level and Ci+1
is the adjacent fruit ripeness level, and quantified the area of
fruit by dividing the number of pixels of class i that has a
color value of less than or equal to H to total pixel of class i,
and multiply the quotient to 100. The same approach was
performed for the classification of tomatoes grounded from
the pixel colors on this study. Apart from the color gradient
to the overall skin of the fruit, color gradient spot of the next
or previous ripeness level is also part of the classification; for
instance, the breakers ripeness stage which is a green-colored
gradient but also has a pink or red gradient in certain spots or
the pink ripeness stage that still has a green gradient color in
certain tomato surface. This is critical to ensure that the color
for classification determinant is not focused on single mix.
To classify tomatoes based from its surface color, a non-
linear SVM classifier was trained and validated. Castro et al.
[25] acknowledged SVM classifier with L*a*b* as the best
machine learning technique and color space for ripeness
classification of a fruit. Nonetheless, SVM was designed for
two class problems only. Hence, the application of SVM to a
multi-class classification requires a reduction of classes to
produce binary problems. Two approaches could achieve the
reduction of classes namely “one against all” or “one against
one”. The first approach is training the binary classifiers to
separate each class from others, which is a fast method but
usually suffer from marginal errors [41]. The second method
is much like the first one but the “one against one” utilizes
one optimization problem to obtain the N decision functions
which is why it was selected and utilized for the multi-class
of tomato ripeness level needed for the classification model.
Fig. 4. Visualization of SVM Classification Plot.
LOOKING FOR THE FULL PAPER?
manuelgarcia.info/publication/tomayto-tomahto
TABLE I. DETAILED ACCURACY RESULTS OF THE TOMATO RIPENESS CLASSIFICATION MODEL
Ripeness
Stage
Tomato Samples
Correctly Classified
Accuracy (%)
Precision
Recall
F-Measure
Training
Testing
Training
Testing
Training
Testing
Green
105
45
101
43
96.19
95.55
0.9600
0.9114
0.9351
Breakers
105
45
94
38
89.52
84.44
0.8800
0.8980
0.8889
Turning
105
45
84
36
80.00
80.00
0.8000
0.8276
0.8136
Pink
105
45
91
38
86.67
84.44
0.8600
0.8600
0.8600
Light Red
105
45
76
35
72.38
77.78
0.7400
0.7450
0.7425
Red
105
45
75
37
71.43
82.22
0.7467
0.7417
0.7442
Total
630
270
521
227
82.70
84.07
0.8311
0.8306
0.8307
IV. EXPERIMENTAL EVALUATION
Using a one-against-one multi-class SVM, a 5-fold cross-
validation strategy was performed to 900 colored-images of
tomatoes (150 samples per ripeness level) divided into 70:30
ratios for testing and training per ripeness class. Each tomato
sample was analyzed based from its L*a*b values to build the
classification model. Area Under Curve (AUC), denoting the
aggregate measure of performance across all possible tomato
ripeness classification thresholds, shows a close value to 1.0
which means that the prediction of the model is close to being
correct. In classifying a non-linear dataset, the SVM linear
classifier is still a thinkable algorithm so long that the given
dataset would be projected into a higher dimension by adding
a new dimensionality to transform the dimension into 3D in
which the data is linearly separable but dimensionality
reduction is through Principal Component Analysis (PCA).
The multi-class SVM classifier for tomato ripeness level
yielded a mean accuracy of 83.39%, which is higher than the
accuracy achieved by Taofik et al. [31] (accuracy = 80%) for
the classification of tomato ripeness stage using fuzzy logic
technique and RGB color space. The experimental results of
Goel and Sehgal [19] (accuracy = 94.29%) utilizing Fuzzy
Rule-Based Classification approach (FRBCS), on the other
hand, produced a more accurate classification. Therefore, the
classifier and color space used, SVM and L*a*b, may not be
the most precise and accurate when it comes to this particular
dataset, unlike the results of Castro et al. [25] when tested on
cape gooseberry dataset. In classifying the ripening stage of
tomatoes using this model, the most accurate is on the green
stage which could be explained through the color percentage
(almost 100%) of green shades present on the surface. On the
other hand, the least accurate ripeness classification is on the
light red stage where 60% of tomato surface shows pinkish-
red while the remaining is a mix of red and pink shades. The
confusion matrix presented a hard evidence that the tomato
ripeness classification model had a problem with adjacent
neighboring levels especially on the pinkish-red, light red and
red stages. Nevertheless, the classification model yielded an
acceptable precision, recall, and f-measure (precision =
0.8311; recall = 0.8306; f-measure: 0.8307).
V. CONCLUSION
With the critical role of agriculture in global economy, it
is inevitable for technologists and agriculturists to find a way
on how to mount machine learning on agricultural sector to
drive agrarian productivity and harvest quality which can be
observed on today’s modern agricultural system operations.
The automated process of ripeness estimation, for instance, is
not only beneficial for increasing sustainable crop production
but also for decreasing pre- and post-harvest waste. Machine
learning models built for ripeness classification can be surely
beneficial for different domains and applications like sorting
system based from maturity level and crop system to ensure
timely harvest, and on the digital agriculture in general. This
is the main research motivation of the proposed approach on
classifying the tomato ripeness using SVM algorithm.
As indicated earlier, Philippines is an agricultural country
where tomatoes are considered as one of the major crops of
the country. Therefore, another motivation of this study is to
develop an image classifier using machine learning process.
The tomato dataset used for experiments were 900 images
assembled by downloading images from search engines and
capturing photos of tomatoes brought from a market which is
directly supplied by a farm using CVS. Divided into 70:30
ratios for testing and training per ripeness class, each tomato
sample was analyzed based from its L*a*b values. To ensure
high accuracy result, the proposed approach consists of three
phases: (1) the experimental setup of CVS, (2) preparation of
the tomato image dataset by undergoing pre-processing and
image segmentation, and (3) classification of ripeness level
using one-against-one multiclass SVM where a 5-fold cross-
validation was performed on training and testing dataset.
In conclusion, the proposed machine learning approach
using SVM and L*a*b color space applied on classifying the
ripeness maturity of tomatoes grounded from the pixel color
generated a mean accuracy of 83.39%. Although this is more
accurate than existing ones, there are also other classification
models and techniques that exceeded this result. Evidently,
color space and machine learning algorithm influenced the
accuracy of the classification model. Therefore, future works
should consider using other algorithms and color space, and
try to mix different combination (SVM and RGB, ANN and
HSL, to name a few) to determine which one will produce a
more accurate model. Additionally, other features than color
may also be considered for parameter extraction such as the
diameter and age as these are proven indices for maturity of
tomatoes [42]. Another direction of research is to assemble a
dataset without using artificial background as captured using
CVS to preserve natural lighting and background. Lastly, the
application of nondestructive testing and non-invasive tools
such as smart sensing, colorimetric [43], and hyper spectral
imaging camera and systems [44] may be integrated in the
machine learning process from dataset preparation to image
classification. For information system and mobile application
development standpoint, the machine learning model created
from this study may be used to develop a real-time detection
of tomato ripeness by either uploading a picture on a website
application or by using the inbuilt camera of smartphones. In
the manufacturing and agriculture perspective, the model can
be installed as a core of tomato handling and sorting system.
In closing, the image classification model proposed in this
paper is a contribution to the agriculture sector, particularly on
Philippine agriculture, in terms of providing a valid and
accurate method of tomato ripening stage identification.
LOOKING FOR THE FULL PAPER?
manuelgarcia.info/publication/tomayto-tomahto
CReferences
[1] E. O. Brown, R. V. Ebora, and F. L. C. Decena, "The Current State,
Challenges and Plans for Philippine Agriculture," FFTC Agricultural
Policy Platform (FFTC-AP), 2018.
[2] Philippine Statistics Authority (PSA). (2019). Performance of
Philippine Agriculture: January to March 2019.
[3] Nations Encyclopedia. Philippines Agriculture, Information about
Agriculture in Philippines.
[4] Food and Agriculture Organization of the United Nations. Fact Sheet
Philippines – Women in agriculture, environment and rural
production. Available: http://www.fao.org/3/ae946e/ae946e03.htm
[5] Philippine Statistics Authority (PSA). (2018). Crops Statistics of the
Philippines - National and Regional.
[6] Philippine Statistics Authority (PSA). (2019). Major Vegetables and
Root Crops Quarterly Bulletin, January-March 2019. Available:
https://psa.gov.ph/vegetable-root-crops-main/tomato
[7] D. Bhowmik, K. P. S. Kumar, S. Paswan, and S. Shweta, "Tomato -
A Natural Medicine and Its Health Benefits " Journal of
Pharmacognosy and Phytochemistry, vol. 1, pp. 33-43, 2012.
[8] W. Klunklin and G. Savage, "Effect on Quality Characteristics of
Tomatoes Grown Under Well-Watered and Drought Stress
Conditions," Foods, vol. 6, 2017.
[9] G. Ramanathan, "Using Color Quality to Indicate Ripeness in Fruits:
Creating a Better Product with Spectrophotometric Technology,"
Color in Food Industry, HunterLab, April 24, 2015 2015.
[10] Z. Xiaobo, Z. Jiewen, and L. Yanxiao, "Apple color grading based on
organization feature parameters," Pattern Recognition Letters, vol.
28, pp. 2046-2053, 2007/11/01/ 2007.
[11] S. Cárdenas-Pérez, J. Chanona-Pérez, J. V. Méndez-Méndez, G.
Calderón-Domínguez, R. López-Santiago, M. J. Perea-Flores, et al.,
"Evaluation of the ripening stages of apple (Golden Delicious) by
means of computer vision system," Biosystems Engineering, vol. 159,
pp. 46-58, 2017/07/01/ 2017.
[12] F. Mendoza and J. M. Aguilera, "Application of Image Analysis for
Classification of Ripening Bananas," Journal of Food Science, 2004.
[13] M. Paulraj, C. R. Hema, K. R. Pranesh, and M. R. Siti Sofiah, "Color
recognition algorithm using a neural network model in determining
the ripeness of a banana," in Proceedings of the International
Conference on Man-Machine Systems (ICoMMS) 2009, pp. 1-4.
[14] D. J. Damiri and C. Slamet, "Application of image processing and
artificial neural networks to identify ripeness and maturity of the
lime," Int. J. Basic Appl. Sci., vol. 1, pp. 171–179, 2012.
[15] H. Zheng and H. Lu, "A least-squares support vector machine (LS-
SVM) based on fractal analysis and CIELab parameters for the
detection of browning degree on mango (Mangifera indica L.),"
Computers and Electronics in Agriculture, vol. 83, pp. 47-51, 2012.
[16] C. S. Nandi, B. Tudu, and C. Koley, "A Machine Vision-Based
Maturity Prediction System for Sorting of Harvested Mangoes,"
IEEE Transactions on Instrumentation and Measurement, 2014.
[17] N. Vélez-Rivera, J. Blasco, J. Chanona-Pérez, G. Calderón-
Domínguez, M. de Jesús Perea-Flores, I. Arzate-Vázquez, et al.,
"Computer Vision System Applied to Classification of “Manila”
Mangoes During Ripening Process," Food and Bioprocess
Technology, vol. 7, pp. 1183-1194, April 01 2014.
[18] G. Polder, G. W. A. M. van der Heijden, and I. T. Young, "Spectral
Image Analysis for Measuring Ripeness of Tomatoes," ASAE, 2002.
[19] N. Goel and P. Sehgal, "Fuzzy classification of pre-harvest tomatoes
for ripeness estimation – An approach based on automatic rule
learning using decision tree," Applied Soft Computing, vol. 36, 2015.
[20] M. S. B. Shah Rizam, A. R. Farah Yasmin, M. Y. Ahmad Ihsan, and
K. Shazana, "Non-destructive watermelon ripeness determination
using image processing and artificial neural network," Int. J. Electr.
Comput. Eng., vol. 3, pp. 332–336, 2009.
[21] M. P. Arakeri and Lakshmana, "Computer Vision Based Fruit
Grading System for Quality Evaluation of Tomato in Agriculture
industry," Procedia Computer Science, vol. 79, pp. 426-433, 2016.
[22] F. Fernqvist, "Consumer experiences of tomato quality and the
effects of credence," Department of Work Science, Business
Economics and Environmental Psychology, Swedish University of
Agricultural Sciences, Swedish, 2014.
[23] A. Bhargava and A. Bansal, "Fruits and vegetables quality evaluation
using computer vision: A review," Journal of King Saud University -
Computer and Information Sciences, 2018/06/05/ 2018.
[24] A. Baltazar, J. I. Aranda, and G. González-Aguilar, "Bayesian
classification of ripening stages of tomato fruit using acoustic impact
and colorimeter sensor data," Computers and Electronics in
Agriculture, vol. 60, pp. 113-121, 2008/03/01/ 2008.
[25] W. Castro, J. Oblitas, M. De-la-Torre, C. Cotrina, K. Bazán, and H.
Avila-George, "Using machine learning techniques and different
color spaces for the classification of Cape gooseberry (Physalis
peruviana L.) fruits according to ripeness level," PeerJ Preprints, vol.
7, p. e26691v2, 2019/02/14 2019.
[26] R. Xiang, Y. Ying, and H. Jiang, "Research on image segmentation
methods of tomato in natural conditions," in 2011 4th International
Congress on Image and Signal Processing, 2011, pp. 1268-1272.
[27] F. Mendoza, N. A. Valous, P. Allen, T. A. Kenny, P. Ward, and D.-
W. Sun, "Analysis and classification of commercial ham slice images
using directional fractal dimension features," Meat Science, 2009.
[28] R. Quevedo, L.-G. Carlos, J. M. Aguilera, and L. Cadoche,
"Description of food surfaces and microstructural changes using
fractal image texture analysis," Journal of Food Engineering, 2002.
[29] S. Kerdpiboon, W. L. Kerr, and S. Devahastin, "Neural network
prediction of physical property changes of dried carrot as a function
of fractal dimension and moisture content," Food Research
International, vol. 39, pp. 1110-1118, 2006/12/01/ 2006.
[30] H. Zheng, L. Jiang, H. Lou, Y. Hu, X. Kong, and H. Lu, "Application
of Artificial Neural Network (ANN) and Partial Least-Squares
Regression (PLSR) to Predict the Changes of Anthocyanins, Ascorbic
Acid, Total Phenols, Flavonoids, and Antioxidant Activity during
Storage of Red Bayberry Juice Based on Fractal Analysis and Red,
Green, and Blue (RGB) Intensity Values," Journal of Agricultural
and Food Chemistry, vol. 59, pp. 592-600, 2011/01/26 2011.
[31] A. Taofik, N. Ismail, Y. A. Gerhana, K. Komarujaman, and M. A.
Ramdhani, "Design of Smart System to Detect Ripeness of Tomato
and Chili with New Approach in Data Acquisition," IOP Conference
Series: Materials Science and Engineering, vol. 288, 2018.
[32] M. Cantwell, "Ripening Tomatoes," Postharvest Technology Center,
Dept. Plant Sciences, UC Davis, 2015.
[33] D. Wu and D.-W. Sun, "Colour measurements by computer vision for
food quality control – A review," Trends in Food Science &
Technology, vol. 29, pp. 5-20, 2013/01/01/ 2013.
[34] M. B. Garcia, T. F. Revano, B. G. M. Habal, J. O. Contreras, and J.
B. R. Enriquez, "A Pornographic Image and Video Filtering
Application Using Optimized Nudity Recognition and Detection
Algorithm," in 2018 IEEE 10th International Conference on
Humanoid, Nanotechnology, Information
Technology,Communication and Control, Environment and
Management (HNICEM), 2018, pp. 1-5.
[35] N. T. Sang and T. Q. Vinh, "FPGA implementation for real-time
Chroma-key effect using Coarse and Fine Filter," in 2013
International Conference on Computing, Management and
Telecommunications (ComManTel), 2013, pp. 157-162.
[36] H. Zhang, J. E. Fritts, and S. A. Goldman, "Image segmentation
evaluation: A survey of unsupervised methods," Computer Vision
and Image Understanding, vol. 110, pp. 260-280, 2008/05/01/ 2008.
[37] S. U. Lee, S. Yoon Chung, and R. H. Park, "A comparative
performance study of several global thresholding techniques for
segmentation," Computer Vision, Graphics, and Image Processing,
vol. 52, pp. 171-190, 1990/11/01/ 1990.
[38] N. Senthilkumaran and S. Vaithegi, "Image Segmentation by Using
Thresholding Techniques for Medical Images," Computer Science &
Engineering: An International Journal (CSEIJ), vol. 6, 2016.
[39] Agricultural Marketing Service, "United States Standards for Grades
of Fresh Tomatoes," ed: Fresh Products Branch, Fruit and Vegetable
Division, United States Department of Agriculture, 1991.
[40] R. Amirulah, M. M. Mokji, and Z. Ibrahim, "Starfruit Color Maturity
Classification Using Cr as Feature," 2010 Sixth International
Conference on Signal-Image Technology and Internet Based Systems,
IEEE Computer Society, pp. 93-97, 2010.
[41] H. Chih-Wei and L. Chih-Jen, "A comparison of methods for
multiclass support vector machines," IEEE Transactions on Neural
Networks, vol. 13, pp. 415-425, 2002.
[42] P. Okiror, J. B. Lejju, J. Bahati, G. K. Rugunda, and C. I. Sebuuwufu,
"Maturity indices for tomato (Solanum lycopersicum L.), cv. Ghalia
281 in Central Uganda," African Journal of Agricultural Research,
vol. 12, pp. 1196-1203, 2017.
[43] R. L. Shewfelt, J. K. Brecht, and C. N. Tai, "Classification of tomato
ripeness and maturity by food colorimetry," HortScience: a
publication of the American Society for Horticultural Science, 1992.
[44] Y. Lu, Y. Huang, and R. Lu, "Innovative Hyperspectral Imaging-
Based Techniques for Quality Evaluation of Fruits and Vegetables: A
Review," Applied Sciences, vol. 7, p. 189, 2017.
LOOKING FOR THE FULL PAPER?
manuelgarcia.info/publication/tomayto-tomahto