TENCON 2023 - 2023 IEEE Region 10 Conference (TENCON) FriMo1J.

2
31 Oct - 3 Nov 2023. Chiang Mai, Thailand

Non-invasive Transport Tier Classification of

Banana ‘Señorita’ (musa acuminata) Using Machine
Learning Techniques
Robert G. de Luna Verna C. Magnaye Rose Anne L. Reaño
Polytechnic University of the Philippines Polytechnic University of the Philippines Polytechnic University of the Philippines
Sto. Tomas City, Batangas, Philippines Sto. Tomas City, Batangas, Philippines Sto. Tomas City, Batangas, Philippines
rgdeluna@pup.edu.ph vcmagnaye@pup.edu.ph ralreano@pup.edu.ph

Karina L. Enriquez Meghann Kim O. Dungca Darlene Rocel C. Filler

Polytechnic University of the Philippines Polytechnic University of the Philippines Polytechnic University of the Philippines
Sto. Tomas City, Batangas, Philippines Sto. Tomas City, Batangas, Philippines Sto. Tomas City, Batangas, Philippines
klenriquez@pup.edu.ph mkodungca@iskolarngbayan.pup.edu.ph drcfiller@iskolarngbayan.pup.edu.ph

Cailon Jullius V. Cabrera Charlene M. Reyes Seane Allen J. Saballa

Polytechnic University of the Philippines Polytechnic University of the Philippines Polytechnic University of the Philippines
Sto. Tomas City, Batangas, Philippines Sto. Tomas City, Batangas, Philippines Sto. Tomas City, Batangas, Philippines
cjvcabrera@iskolarngbayan.pup.edu.ph cmreyes@iskolarngbayan.pup.edu.ph sajsaballa@iskolarngbayan.pup.edu.ph

Nicole Anne O. Maligalig

Polytechnic University of the Philippines
Sto. Tomas City, Batangas, Philippines
naomaligalig@iskolarngbayan.pup.edu.ph

Abstract—The lack of a transport quality forecasting storage system. More specifically, crops such as vegetables
system in farming and sorting facilities of indigenous varieties and fruits tend to stale quickly. Making them a large
of bananas is aiding the increase of food waste generation in contributor to the world’s food waste [1].
the country. This in turn decreases agriculture sustainability
and imposes economic losses to farmers. Musa acuminata
‘Señorita’ are diploid cultivars of bananas originating in the An example of this is one of the most significant fruits
Philippines. This study aims to develop a machine learning- found in the Philippines which are bananas. Bananas are
based system that classifies Musa acuminata ‘Señorita’ extensively consumed for their nutritional value as they are
bananas into their transport tiers: (I) for interprovincial rich in numerous health benefits that serve as protection
distribution, (II) for intra provincial distribution, or (III) against cell damage and various human diseases [2].
subject for rejection. The model is trained and went through 7 Because of this, banana production in the country is
machine learning classifiers to identify which model is the most prevalent and competitive under both local trade, export
compatible with the system design. The application of external trade, and import substitution scenarios [3]. However, not
parameters such as size, girth, weight, maturity stage, and
all the available produce is sold as about 60% of banana
RGB parameters can be the foundation to develop a machine
learning-based banana transport tier classifier that accurately biomass remains as waste and around 114.08 million metric
monitors and determines how long they can travel based on tons (MMT) of waste-loss are generated after harvest,
their maturity level. Among the seven models, Logistic causing dilemmas in the environment such as emission of
Regression, Linear Discriminant Analysis, Decision Tree harmful greenhouse gases.[4] Therefore, it is important to
Classifier, Gaussian Naive Bayes, and Support Vector Machine identify the transport quality better fitted for a bananas’
attained a classification accuracy of 100 %. The development maturity stage.
of this system can aid in the proper distribution of produce and
help uplift the country’s agriculture and economic Thus, methods are curated to determine a bananas’ post-
sustainability.
harvest maturity and freshness. A recent study in 2022 used
Keywords—musa acuminata, señorita banana, machine two deep learning techniques namely, Convolutional Neural
learning, transport classification, fruit maturity detection Networks and AlexNet to determine the best deep learning
algorithm most compatible for the forecasting of banana
I. INTRODUCTION fruit maturity and shelf-life quality. The researchers
The world’s growing population requires agricultural concluded that CNN is the optimal algorithm for banana
sustainability for guaranteed food security and hunger fruit maturity and quality classification. [5]. Meanwhile, in
abolition. To feed the growing population, food production 2019 and 2020, two studies recognized the importance of
the evaluation or estimation of the changing process of
must be increased by 60-110 % globally by 2050. However,
this rapid production may also lead to the increasing bananas. It is explained that the Cavendish Banana has four
probability of food waste or food loss, which pertains to consecutive maturity stages which vary from a dark green
edibles that remained unconsumed due to certain banana, a yellow banana with hints of green, an all-yellow
circumstances such as spoilage, expiration, or unproper banana, and a purely yellow banana with brown spots,

979-8-3503-0219-6/23/$31.00 ©2023 IEEE 1039

respectively [6]. Another method is the usage of transfer Señorita banana to the interprovincial and intra provincial
learning which established the connection between the trade market will improve the Philippine’s economic
freshness of a fruit and its storage dates. The model was able growth, more specifically Region-IV Calabarzon, as said
to detect banana freshness with 98.92% accuracy, which is species of the fruit is mainly produced and transported from
significantly higher than the human capacity [7]. the area. The project will serve as a basis and
encouragement to pioneer fruit quality inspection in the
Other published papers put concern about Banana Strain community to aid in preventing food loss, which is directly
detection in conjunction with identifying its quality. In correlated to global hunger.
2021, five strains of different bananas are classified and
identified with 96.8% performance accuracy. These are the C. Assumptions, Scope, and Delimitations
cavendish, shabri, red, lady finger and green banana. [8]. There are many claims that the use of Machine Learning
algorithms has a positive impact in the economy,
However, there is a lack of variety within the species of agricultural management, efficient monitoring of resources,
banana used in the studies. Only a specific type of banana, and improved environmental state [9]. Specifically, the
which is the Cavendish, was frequently used as subject for banana supply chain has become a recent area of focus for
fruit maturity and quality monitoring as it is the most research that applies Machine Learning in smart agriculture
common and available globally. The parameters manually [10] Due to bananas having a high rate of deterioration,
taken such as size were also recorded dependent on a single farmers must determine the fruit’s maturity stage to
banana from a whole hand, leading for the raw data to be minimize losses during the post-harvest process [11].
less accurate.
The scope of the study primarily focuses on a specific
With this in consideration, the researchers chose Musa species of banana called Senorita, as there is limited
acuminata Señorita cultivar as the focus of the study to information available that is dedicated to said species.
monitor a systematic model that predicts the maturity and Parameters included in the selection of features that have a
trading grade by incorporation of machine learning sufficient correlation to the transportability of said bananas
techniques based on several parameters such as RGB color are size, weight, maturity, girth, R, G, and B color spectrum.
indexes, measured finger size, weight, and mean
circumference. These parameters will be used as a guide to Other species of bananas are not included in the study.
classify Señorita bananas, which has a significantly fragile The fruit is classified in only 3 transport tiers namely:
peel that is dependent on their maturity level, into different interprovincial, intra provincial, and rejected respectively.
tiers: (TIER I) for interprovincial distribution (Up to 5
days), (TIER II) for intraprovincial distribution (Up to 3 II. REVIEW OF RELATED LITERATURE AND WORKS
days), and (TIER III) for rejected class. The parameters chosen to determine the transportability
of Senorita bananas are maturity, girth, weight, size, and R,
TABLE I. TRANSPORT TIER
G, and B colors. The maturity of bananas is a leading factor
Transport Life to accomplish the aim of this research as it has a big effect
Transport tier Distribution
(Days)
on the fruit's quality, marketability, and shelf life after it has
I Up to 5 Interprovincial been harvested [12]. Meanwhile, size and RGB are the most
II Up to 3 Intraprovincial frequently observed qualities when evaluating banana
quality. The Red, Green, and Blue color system enhances
III 0-1 Rejected the accuracy of image processing and determines hue
frequency values that can visually predict a banana’s
A. Objectives of the Study maturity level when incorporated to machine or deep
This project aims to classify the transportability of learning techniques [13]. Similarly, a change in the size and
bananas through image processing and multiple machine girth of a banana is a great indicator of its species and aids
learning techniques by considering their maturity, size, in age-bunch control [14]. This is showcased in a study
girth, weight, R, G, and B values. Specifically, the study which classified post-harvest grading based on the
aims to: parameters mentioned along with surface analysis [15].
Lastly, fruit yield is represented by hand weight which is
A.1. identify the best combination of features to
significant to the cost evaluation of its distribution [16].
determine the transportability of bananas.
With these in mind, the selected parameters are further
A.2. accurately predict the transportability of bananas
investigated on the following studies:
using selected parameters and identify which machine
learning technique is best suited for the system. A. External Parameters
The stage of maturity is vital to the quality, marketability,
B. Significance of the Study and shelf life of bananas after they have been picked. Color,
The study will contribute to the United Nations size, and defects are the most frequently observed qualities
Sustainable Development Goals (UN SDG) as the when evaluating banana quality. These are either currently
monitoring and classification of bananas’ transportability classified manually or individually studied by past
based on their maturity level will ensure responsible researchers.
production and consumption of goods, which would then
The study [17] conducted a study in the Philippines that
limit excessive food waste pollution. The introduction of resulted in a system that accurately sorts bananas by cultivar

979-8-3503-0219-6/23/$31.00 ©2023 IEEE 1040

and grade. Image processing with the main algorithms K-
Means clustering and HOG feature extraction makes this
possible. In 2019, a group of researchers took a similar
approach. Using optical imaging techniques, the peel color
characteristic is more suitable for determining the various
phases of banana fruit maturity due to its simplicity and
precision [18].
For determining which type of machine learning model is
best for determining banana ripeness or quality, Zhang et
al.’s [19] study stated that the CNN model provides a crucial
indicator to show the banana ripeness that works well for
both a coarse and fine-grained classification of the fruit's
ripening stages.
B. RGB Parameter
A common parameter integrated into existing studies and
observations about fruit quality is the RGB system. The RGB
colors, red, green, and blue, can determine the specification
of hues in a standardized way which can be easily recognized
by human eyes. The combination of these three shades
creates a three-dimensional coordinate system wherein a
single point represents each color. RGB, HSV, HSI, and Fig. 1. Methodological Framework that Summarizes the Entire Systematic
L*a*b color models are the most commonly available color Process of the Study
spaces [20].
Relatively, to achieve a more accurate RGB-based LED bar light illumination against a plain white
prediction, Chen, et. al. [21] wrote a study comparing background, the raw images of the Señorita bananas are
different tobacco leaves’ skewness distribution as well as recorded with 50 samples for each class of banana (class 1,
computing the mean, median, mode, and kurtosis, which class2, and class 3) giving us 150 samples per class. Image
showed greater accuracy. augmentation is performed by photographing in two
The benefits of both RGB and hyperspectral imaging in orientations (front and back). The total number of bananas
the classification of bananas can also be used to model the amounted to 300 samples which were photographed in a
classification of other horticultural crops [22]. With regards way that the overall values and conditions of the sample are
to farm postharvest procedures, the multi-input model's rapid captured.
processing time may be more efficient. Another study
integrated an image recognition system using a hyperspectral
camera. The corresponding images taken are rendered with a
set of RGB images under it to build and feed the neural
network model that will be the basis of prediction for the
sample’s quality [23].
In 2022, an RGB-based system was used for classifying
fruit stages. Stages are as follows: young, swelling, white-
mature, premature, and mature. The results of the
observation were based on the quantified pigmentation and (a)
key color changes of the fruit [24]. In the same year, a study
performed image segmentation and color transformation
where the input image’s RGB color space is converted into
Lab color space, which is a color system with dimension for
lightness (black to white) on a scale (zero to 100) [25].
III. METHODOLOGY
Application of image processing in agriculture has been
a growing research area during the past years. In this study,
Image processing techniques used concerned the following (b)
steps: image acquisition, pre-processing of raw data and
image segmentation, and extraction of features. Fig. 1 shows
Fig. 2. (a) Actual photographing of samples in two orientations (Front and
the methodological framework used in the study. Back) and (b) raw image samples of Señorita bananas.
A. Image Acquisition
The subject of this study is the Musa acuminata (AA The following information in Table II regarding
Group) 'Señorita'. The data gathering process started with maturity time frames and visual cues of the bananas
the preparation of the bananas to be used for image according to the seller and supplier’s perspectives was also
acquisition. Using the 12 MP camera of an iPhone 8 and an acquired.

979-8-3503-0219-6/23/$31.00 ©2023 IEEE 1041

TABLE II. MATURITY TIME FRAMES WITH INDICATED
PARAMETERS
Banana Parameters
Maturity Maturity Days in
Level Color
in Days Transport
Unmature 1-2 Up to 5 days Green
Partially Green with some
3-5 Up to 3 days
Mature yellow parts
Reject / Yellow to Dark
Mature 6-7 immediate Yellow with Black
dispatch and Brown Specks

B. Data Preprocessing and Segmentation

Fig. 4. RGB values extraction via OpenCV
The images will go through pre-processing, where
image enhancement takes place for photographs that have
existing noises or interference. The raw images were labeled RGB color space is the most common color space where
“1uf.jpg 1ub.jpg, 1pmf.jpg, 1pmb.jpg, 1mf.jpg, 1mb.jpg, colors are represented by the three primary colors (red,
etc.” indicating the maturity stage (u,pm,m), and banana green, and blue). For each component, the value ranges
orientation(front, back). In this stage, image segmentation, between 0 and 255, where black is represented with the
or the separation of the background from the actual sample, tuple (0, 0, 0) while (255, 255, 255) are the values for the
is done using binary thresholding. Hidden details are white color. After going through the process of elimination
extracted to increase low contrasts in some images to of noise and background via Binary Thresholding, the RGB
produce an output whose pixels are significantly enhanced values of the Señorita banana images were extracted.
for RGB recognition, and mechanism.
The approach was to feed the images using cv2.imread()
function. This function reads the image then a NumPy array
representing the image is returned. The cv2.split() function
is then used to split the image into its three-color channels.
This function takes an input image as an argument and
returns a tuple containing the three-color channels.
(a) (b) E. Feature Selection
Several methods of feature selection, such as recursive
feature elimination, univariate selection, and feature
importance are performed to further identify the best
features to be incorporated in the machine learning model.
Table III summarized the results of the three featue selection
(c) (d) (e) methods. It suggests that variables which are significant in
predicting the class are maturity, girth_mm, size_mm, RGB
features.
Fig. 3. Process of Image segmentation via Binary Thresholding:
(a) Original image, (b) grayscale conversion, and blurring to denoise (c) TABLE III. HOLD-OUT VALIDATION USING SELECTED FEATURES
histogram creation of the gray scaled and blurred image (d) create a binary (OPTIMIZED PARAMETERS)
mask based on the threshold values (e) Background Subtraction using the
binary mask Feature
Univariate Recursive
Importance
maturity
C. Feature extraction girth_mm maturity
G
weight_kg girth_mm
A total of 7 parameters are acquired namely: size_mm, B
maturity size_mm
weight, girth_mm, orientation, and the individual R, G, B Features Selected R
B B
girth_mm
color indexes. Banana farmer’s mode of banana size_mm
G G
classification is typically based on banana size, making it a R weight_kg
significant parameter along with its color indexes. Related
literature limitations concerning dimension extraction via To justify that only these features are relevant for the
image processing of banana images, manual measurement prediction of class, the researchers provided supporting
was implemented. The size parameter is measured in related literature and works and used a correlation matrix to
millimeters (mm), the weight of the bananas was measured evaluate the correlation between the features that were
using a standard weighing scale and recorded in kilograms selected, as highly correlated features can lead to problems
(kg), the girth was measured by measuring the average such as multicollinearity in some models. Weight is
circumference of each finger in a bunch in millimeters correlated to maturity and size, as shown in the matrix in
(mm). Finally, the extraction of the RGB values for each Fig. 5. Weight may not have a direct relationship with the
raw image sample is executed using the OpenCV library as target variable or class, and its high ranking in the plot may
shown in Fig. 4. be due to its correlation with other features that are more

979-8-3503-0219-6/23/$31.00 ©2023 IEEE 1042

relevant. In such cases, weight may not provide any Support Vector Machine
C: 0.1, decision_function_shape:
additional information helpful for class prediction. ovo,kernel: linear
activation: logistic, alpha: 0.0001,
Multi-layer Perceptron hidden_layer_sizes: (100, 100),
Classifier learning_rate: constant,
random_state: 0, solver: adam

Table V and VI shows the performance of the models

with optimal parameters using selected features. Five out of
seven models obtained an accuracy of 100 % after the cross-
validation and hold-out validation namely: logistic
regression, linear discriminant analysis, decision tree,
Gaussian Naïve Bayes, and support vector machine.
TABLE V. CROSS VALIDATION USING SELECTED FEATURES
(OPTIMIZED PARAMETERS)
Fig. 5. Correlation between Variables
Machine Learning Model Accuracy Result
F. Machine Learning
Logistic Regression 1.000000
The machine learning process has two main parts: the
training and testing of dataset. During the training phase, 80 Linear Discriminant Analysis 1.000000
% of the dataset is trained to create a machine learning K-Nearest Neighbors 0.893333
model. Seven classification models are used: logistic
regression, linear discriminant analysis, K-nearest Decision Tree 1.000000
neighbors, decision tree classifier, gaussian naïve bayes,
Gaussian Naive Bayes 1.000000
support vector machine and multi-layer perceptron
classifier. Training validation ensures that there will be no Support Vector Machine 0.993333
overfitting of data and the process is generalized. The
remaining 20 % is treated as the test dataset which is used Multi-layer Perceptron Classifier 0.966667
for testing the model. The model prediction is evaluated
through performance metrics like precision, recall, f1-score, TABLE VI. HOLD-OUT VALIDATION USING SELECTED FEATURES
and accuracy. (OPTIMIZED PARAMETERS)
G. Optimization of Parameters Machine
Precision Recall F1-score Accuracy
Learning Model
Machine learning optimization refers to the process of Logistic
1.00 1.00 1.00 1.00
iteratively increasing the accuracy and decreasing the level Regression
of error of a machine learning model. Machine learning Linear
Discriminant 1.00 1.00 1.00 1.00
models learn to generalize and predict new live data based Analysis
on knowledge gained from training data. The GridSearchCV K-nearest
0.86 0.86 0.86 0.85
was used to find the best parameters for each machine Neighbors
learning model. Decision Tree 1.00 1.00 1.00 1.00
Gaussian Naïve
IV. RESULTS AND DISCUSSION 1.00 1.00 1.00 1.00
Bayes
Support Vector
Table IV summarizes the list of parameters and their Machine
1.00 1.00 1.00 1.00
optimal values for each machine learning model. These Multi-layer
0.97 0.97 0.97 0.97
parameters were determined using the GridSearchCV Perceptron
function. All parameters of each not included in the
optimization process are considered using their default
values. V. CONCLUSION
This section provides a summary of the project and a
TABLE IV. BEST PARAMETERS FOR EACH MACHINE LEARNING restatement of its primary outcome, i.e., what was learned
MODELS
and what was accomplished. At the conclusion of the test,
the deliverability prediction accuracy for bananas was
Machine Learning Model Optimized Parameters
determined given seven parameters, including size_mm,
Logistic Regression
C: 100, penalty: l1, random_state:0, weight, girth_mm, orientation, and the R, G, and B color
solver: liblinear indexes of bananas.
Linear Discriminant Analysis shrinkage: auto, solver: lsqr
A satisfactory classification using machine learning was
algorithm: auto, leaf_size: 10,
K-nearest Neighbors achieved from the seven machine learning models. Each
n_neighbors: 6, weights: distance
criterion: gini, max_depth: 2, model's best parameter combination is determined by
Decision Tree Classifier min_samples_leaf: 1, optimization. Among the seven models, Logistic
min_samples_split: 2 Regression, Linear Discriminant Analysis, Decision Tree
Gaussian Naïve Bayes var_smoothing: 1e-09 Classifier, Gaussian Naive Bayes, and Support Vector
Machine attained a classification accuracy of 100 %. The

979-8-3503-0219-6/23/$31.00 ©2023 IEEE 1043

maturity and size parameters are classified as good PLOS ONE 14(10):e0223906. DOI:
https://doi.org/10.1371/journal.pone.0223906
predictors of banana class for its deliverability. [14] Siddiq M, Ahmed J, Lobo MG (2020) Handbook of banana
production, postharvest science, processing technology, and nutrition.
Hoboken, Wiley, 284 p. DOI: https://doi.org/10.1002/9781119528265
VI. ACKNOWLEDGMENT [15] Ucat, R. C., & Dela Cruz, J. C. (2019). Postharvest Grading
Classification of Cavendish Banana Using Deep Learning and
The authors would like to acknowledge the support and Tensorflow. 2019 International Symposium on Multimedia and
Communication Technology (ISMAC).
resources provided by Polytechnic University of the doi:10.1109/ismac.2019.8836129
Philippines (PUP). The facilities, funding, and access to [16] Ogunsua, J., Saengrayap, R., Ullah, H., & Chaiwong, S. (2019).
Application of artificial neural networks in the prediction of fruit
relevant literature and research materials have greatly damages and hand weight in Cavendish banana. Journal of Food
contributed to the success of this project. Science and Agricultural Technology (JFAT), 5, 140-145.
[17] S. G. E. Brucal, H. O. Ditano, N. C. Labuson, S. L. D. Raton, L. A.
Samaniego and S. R. Peruda, "Postharvest Quality Assessment of
REFERENCES Non-Chemical Bananas Utilizing HOG Feature Extraction and K-
means Clustering," 2021 Fifth World Conference on Smart Trends in
[1] Bilska B, Tomaszewska M, Kołożyn-Krajewska D. Managing the Systems Security and Sustainability (WorldS4), London, United
Risk of Food Waste in Foodservice Establishments. Sustainability. Kingdom, 2021, pp. 165-170, doi:
2020; 12(5):2050. https://doi.org/10.3390/su12052050 10.1109/WorldS451998.2021.9514024.
[2] Pereira JAM, Berenguer CV, Andrade CFP, Câmara JS. Unveiling the [18] Zhuang, Jiajun & Hou, Chaojun & Tang, Yu & He, Yong & Guo,
Bioactive Potential of Fresh Fruit and Vegetable Waste in Human Qiwei & Miao, Aimin & Zhong, Zhenyu & Luo, Shaoming. (2019).
Health from a Consumer Perspective. Applied Sciences. 2022; Assessment of External Properties for Identifying Banana Fruit
12(5):2747. https://doi.org/10.3390/app12052747 Maturity Stages Using Optical Imaging Techniques. Sensors. 19.
[3] S. Chakraborty, F. M. J. M. Shamrat, M. M. Billah, M. A. Jubair, M. 2910. 10.3390/s19132910.
Alauddin and R. Ranjan, "Implementation of Deep Learning Methods [19] Zhang, Y., Lian, J., Fan, M. et al. Deep indicator for fine-grained
to Identify Rotten Fruits," 2021 5th International Conference on classification of banana’s ripening stages. J Image Video Proc. 2018,
Trends in Electronics and Informatics (ICOEI), Tirunelveli, India, 46 (2018). https://doi.org/10.1186/s13640-018-0284-8
2021, pp. 1207-1212, doi: 10.1109/ICOEI51242.2021.9453004 [20] S. Vanakovarayan, S. Prasanna, S. Thulasidass and M. V, "Non-
[4] Cicih Sugianti, Teppei Imaizumi, Manasikan Thammawong, Kohei Destructive Classification of Fruits by Using Machine Learning
Nakano, Recent Postharvest Technologies in the Banana Supply Techniques," 2021 International Conference on System, Computation,
Chain, Reviews in Agricultural Science, 2022, Volume 10, Pages Automation and Networking (ICSCAN), Puducherry, India, 2021, pp.
123-137, Released on J-STAGE May 15, 2022, Online ISSN 2187- 1-5, doi: 10.1109/ICSCAN53069.2021.9526541.
090X, https://doi.org/10.7831/ras.10.0_123 [21] Chen, Z., Wang, F., Zhang, P. et al. Skewed distribution of leaf color
[5] Aherwadi N, Mittal U, Singla J, Jhanjhi NZ, Yassine A, Hossain MS. RGB model and application of skewed parameters in leaf color
Prediction of Fruit Maturity, Quality, and Its Life Using Deep description model. Plant Methods 16, 23 (2020).
Learning Algorithms. Electronics. 2022; 11(24):4100. https://doi.org/10.1186/s13007-020-0561-2
https://doi.org/10.3390/electronics11244100 [22] Mesa, A., Chiang, J. Multi-Input Deep Skewed distribution of leaf
[6] Zhuang, J., Hou, C., Tang, Y., He, Y., Guo, Q., Miao, A., Zhong, Z., color RGB model and application of skewed parameters in leaf color
& Luo, S. (2019). Assessment of External Properties for Identifying description model | Plant Methods Learning Model with RGB and
Banana Fruit Maturity Stages Using Optical Imaging Techniques. Hyperspectral Imaging for Banana Grading (2021).
Sensors (Basel, Switzerland), 19. https://www.mdpi.com/2077-0472/11/8/687
https://www.researchgate.net/publication/334149536_Assessment_of [23] Zhao J, Kechasov D, Rewald B, Bodner G, Verheul M, Clarke N,
_External_Properties_for_Identifying_Banana_Fruit_Maturity_Stages Clarke JL. Deep Learning in Hyperspectral Image Reconstruction
_Using_Optical_Imaging_Techniques from Single RGB images—A Case Study on Tomato Quality
[7] J. Ni, J. Gao, L. Deng and Z. Han, "Monitoring the Change Process of Parameters. Remote Sensing. 2020; 12(19):3258.
Banana Freshness by GoogLeNet," in IEEE Access, vol. 8, pp. https://doi.org/10.3390/rs12193258
228369-228376, 2020, doi: 10.1109/ACCESS.2020.3045394. [24] Hongxia L. et. al. Quantification and identification analysis of
https://ieeexplore.ieee.org/document/9296756/ Ziziphus jujuba Mill. peel pigmentation at different developmental
[8] A. S. M. F. Al Haque, M. A. Hakim and R. Hafiz, "CNN Based stages. Food Chemistry. 2022;
Automatic Computer Vision System for Strain Detection and Quality https://doi.org/10.1016/j.fochx.2022.100470
Identification of Banana," 2021 International Conference on [25] Y. Mehendran, T. Kartheeswaran and N. D. Kodikara, "Banana
Automation, Control and Mechatronics for Industry 4.0 (ACMI), Freshness Identification Using Image Processing Techniques," 2022
Rajshahi, Bangladesh, 2021, pp. 1-6, doi: 7th International Conference on Business and Industrial Research
10.1109/ACMI53878.2021.9528269. (ICBIR), Bangkok, Thailand, 2022, pp. 347-352, doi:
https://ieeexplore.ieee.org/document/9528269/ 10.1109/ICBIR54589.2022.9786519.
[9] Almeyda, E. , Ipanaque, W. (2022) Recent Developments Of https://ieeexplore.ieee.org/document/9786519/
Artificial Intelligence For Banana: Application Areas, Learning [26] Aceveo, S., Carrillo, A., Florez-Lopez, E., Grande-Tovar, C. (2021)
Algorithms, And Future Challenges. Artificial Intelligence. Recovery of Banana Waste-Loss from Production and Processing: A
http://dx.doi.org/10.1590/1809-4430- Contribution to a Circular Economy. Molecules 26, 5282.
Eng.Agric.v42nepe20210144/2022 https://doi.org/10.3390/molecules26175282
[10] Benos L, Tagarakis AC, Dolias G, Berruto R, Kateris D, Bochtis D [27] Sugianti, C. , Imaizumi, T., Thammawong, M., Nakano T., (2022)
(2021) Machine learning in agriculture: A comprehensive updated Recent Postharvest Technologies in the Banana Supply Chain,
review. Sensors 21(11):1–55. DOI: Reviews in Agricultural Science, Volume 10, Pages 123-137,
https://doi.org/10.3390/s21113758 Released on J-STAGE Online ISSN 2187-090X,
[11] Mazen FMA, Nashat AA (2019) Ripeness classification of bananas https://doi.org/10.7831/ras.10.0_123,
using an artificial neural network. Arabian Journal for Science and https://www.jstage.jst.go.jp/article/ras/10/0/10_123/_article/-char/en
Engineering 44(8):6901–6910. DOI: https://doi.org/10.1007/s13369- [28] Alvindia, D. (2019) Postharvest Pathology of Fresh Horticultural
018-03695-5 Produce. eBook ISBN 9781315209180.
[12] Pu, Y.-Y., Sun, D.-W., Buccheri, M., Grassi, M., Cattaneo, T. M. P., https://taylorfrancis.com/chapters/edit/10.1201/9781315209180-
& Gowen, A. (2019). Ripeness Classification of Bananito Fruit ( 8/banana-plantain-dionisio-alvindia?context=ubx
Musa acuminata, AA): a Comparison Study of Visible Spectroscopy [29] Go, E. (2022) Overland and Oversea: Domestic Trade Frictions in the
and Hyperspectral Imaging. Food Analytical Methods. Philippines, Asian Development Review, Vol. 39, No. 2, pp. 75–118.
doi:10.1007/s12161-019-01506-7 DOI: 10.1142/S0116110522500123
[13] Neupane B, Horanont T, Hung ND (2019) Deep learning based [30] Esguerra, E., Rolle, R. (2018) Post-harvest management of banana
banana plant detection and counting using high-resolution red–green- for quality and safety assurance: Guidance for horticultural supply
blue (RGB) images collected from unmanned aerial vehicle (UAV). chain stakeholders. Food And Agriculture Organization Of The
United Nation. https://www.fao.org/3/I8242EN/i8242en.pdf

979-8-3503-0219-6/23/$31.00 ©2023 IEEE 1044