Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ij isrt/IJISRT24MAR1459

Exploring Deep Learning Approaches for Citrus

Diseases Detection and Classification: A Review
Abdullahi Lawal Rukuna1
Department of Mathematical Science
Abubakar Tafawa Balewa University, Bauchi State Nigeria

Co-Authors:

Dr. F. U. Zambuk2 Dr. A. Y. Gital3

Department of Mathematical Science Department of Mathematical Science
Abubakar Tafawa Balewa University, Bauchi State Nigeria Abubakar Tafawa Balewa University, Bauchi State Nigeria

Umar Muhammad Bello4 Kaje Danladi Shemang5

Department of ICT Department of Mathematical Science
Federal University Gusau, Zamfara State Nigeria Abubakar Tafawa Balewa University, Bauchi State Nigeria

Nahuru Ado Sabongari6

Department of Computer Science
Kaduna Polytechnic, Kaduna State Nigeria

Abstract:- Citrus diseases pose significant threats to USA, Mexico, and Spain. Including numerous other tropical
global agriculture, impacting crop yield and quality. In and subtropical areas across the globe. (Agusti et al., 2013).
recent years the integration of deep learning models has
surfaced as a hopeful method for classifying and Contemporary technologies have empowered human
detecting diseases. This review critically analyzes and society to generate sufficient food to satisfy the needs of
synthesizes 25 research works that explore various deep over 7 billion individuals. Nevertheless, food security
learning models applications in citrus disease detection continues to face threats from various factors, with climate
and classification. The methodology involves a change being a significant concern (mohanty et al., 2016).
systematic literature search, filtering based on relevance,
publication date, and language. The selected works are The current challenges in citrus production include;
categorized, and each is analyzed for contributions and adverse weather conditions or climate changes, pest and
limitations. The review identifies limitations, notably the diseases, availability of arable land, access to technology,
reliance on limited datasets leading to issues of lack of government support etc (Christopher & Udoh, n.d.,
generalization and class imbalance. Data augmentation, 2020)
while employed, lacks comprehensive evaluation. Citrus diseases pose significant threats to global
Practical implementation in real-world agricultural agriculture, affecting crop yield and quality. Over the past
settings remains a challenge, demanding scalable, few years, the incorporation of deep learning models has
adaptable, and robust solutions. Future research surfaced as a promising method for detecting and classifying
directions are proposed to address limitations. Emphasis these diseases. this literature review aims to critically
is placed on curating larger and diverse datasets, actively analyze and synthesize 25 research works that delve into the
mitigating class imbalance, and rigorously evaluating utilization of diverse deep learning models in the context of
data augmentation techniques. detecting and classifying citrus diseases.

Keywords:- Citrus Diseases, Deep Learning, Disease Deep learning finds widespread application not only in
Detection, Disease Classification. image processing, recognition, and classification but also
extends its influence to various domains, including
I. INTRODUCTION agriculture. In comparing to earlier artificial neural network
approaches, deep learning demonstrates superior accuracy in
Presently, citrus stands as the predominant fruit tree recognition and effectively tackles challenges related to
crop globally, accounting for 104 million tons over 7.1 image classification and visualization. Consequently, it is
million hectares. This production is predominantly currently recognized as the most promising technology in
concentrated in countries such as Brazil, China, India, the the contemporary agricultural sector. (liu et al., 2021).

IJISRT24MAR1459 www.ijisrt.com 1821

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ij isrt/IJISRT24MAR1459

In this review, our objective is to illuminate the  Relevance:

strengths and weaknesses of various approaches, ultimately Articles were selected if they specifically addressed
contributing to the development of effective and reliable citrus disease detection and classification using deep
solutions for the detection of citrus diseases. The subsequent learning models.
sections will delve into the methodologies employed,
summarize key findings, and critically assess the limitations  Publication Date:
of each reviewed research work. Preference was given to recent publications between
the year 2019-2024 to capture the latest advancements in the
II. RESEARCH METHODOLOGY field.

To ensure a comprehensive review, an exhaustive  Language:

literature search was conducted across reputable academic Articles published in the English language were
databases including the Institute of Electrical and included to facilitate a standardized review process.
Electronics Engineers (IEEE), Association for Computing
Machinery (ACM), Science Direct, ResearchGate, Google Upon completing the initial search, the identified
Scholar, PubMed, Scopus, SpringerLink, and Tech Science articles underwent a thorough screening process. Relevant
that addressed citrus disease detection using deep learning information from each selected research work was
models, we eliminated any repetitive entries that resulted systematically extracted, including details on deep learning
from database searches. The following criteria were used for models employed, datasets used, experimental setups, and
the selection of research works: key findings. 25 research work were selected and were
categorized based on common themes and methodologies.
This categorization aimed to facilitate a structured review by
identifying patterns and variations across the studies. Each
category was then analyzed in terms of its contributions,
strengths, and limitations.

Fig 1 Research Selection Process

III. RELATED WORK (Firdaus et al., 2023) proposed a convolutional neural

network (CNN) model employing the DenseNet-169
(Syed-Ab-Rahman et al., 2022) Introduced a dual stage architecture in two scenarios include one utilizing original
deep convolutional neural network (CNN) model designed features and another incorporating a combination of
for the detection and classification of citrus diseases through features. The study focuses on classifying four categories:
the analysis of leaf images. The model comprises a feature canker, greening, blackspot and healthy plants, all sourced
extractor and Region Proposal Network, ROI pooling, and a from the same dataset as (Syed-Ab-Rahman et al., 2022).
classifier. It attains a disease detection accuracy of 94.37% Augmentation techniques were applied to the dataset to
and an average precision of 95.8%. The dataset includes augment the quantity of training data in other to overcome
three categories of citrus diseases: black spot, canker, and overfitting and improve the overall detection accuracy. The
Huanglongbing and a healthy class. The proposed model DenseNet-169 model incorporating feature combination
uses Faster R-CNN with ResNet101 as the feature extractor. achieves a 96.66% accuracy rate, achieving superior
The RPN extracts potential diseased areas, and ROI pooling performance compared to the model utilizing only original
and the classifier classify the regions. features 91.33%.

(Raaj & Selvy, n.d.) Introduced a detection system for (Çeti̇ ner et al., 2023) introduced a three deep learning
citrus fruit and leaf diseases using the YOLO classifier. The models employing three distinct approaches for the
proposed YOLO classifier model targets common citrus classification of citrus leave diseases, proposed denseNet,
diseases and demonstrates superior performance compared basic denseNet201 and 21-layer convolutional neural
to other deep learning approaches. The methodology network models were used. The study utilizes a dataset
involves a CNN, Gaussian Feature extraction and YOLO containing Images depicting diseases in citrus leaves,
Classifier, with training images undergoing grayscale including HLB (Huanglongbing), CBS (Citrus Black Spot),
conversion and binary image classification. and CBC (Citrus Canker) with preprocessing steps like

IJISRT24MAR1459 www.ijisrt.com 1822

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ij isrt/IJISRT24MAR1459

resizing, standardization, and data augmentation. An Results show improved accuracy through data
accuracy of 95% was achieved for the proposed augmentation, and MobileNetV2 proves effective for real-
denseNet201 while 99% accuracy was achieved for the time disease identification. The model was able to achieve
CNN model. an accuracy oof 87.28%.

(Luaibi et al., 2021) introduce a two ways conventional (Elaraby et al., 2022) suggested AlexNet and VGG19,
neural network, AlexNet and ResNet architectures both with using transfer learning to enhance efficiency. The process
and without the application of data augmentation, which involves scaling and augmenting disease images, The
focuses on the early detection of citrus leaf diseases. A self- proposed approach is assessed using both the citrus disease
compiled dataset comprising 200 images encompassing image gallery dataset and the combined dataset, which
healthy leaves and diseases such as nutrient deficiency, includes citrus image datasets of infested scale and Plant
Phyllocnistis citrella and scale insects is used. The Results Village. The approach achieves 94% total performance.
reveal that models trained with the incorporation of data
augmentation achieve the highest accuracy, with ResNet at (Dai et al., 2023) Proposed an improved FastGAN
95.83% and AlexNet at 97.92%. (FastGAN2) for image generation and an Enhanced
EfficientNet-B5 (EfficientNet-B5-pro) for classification.
(Khattak et al., 2021) employed a convolutional neural FastGAN2 is designed to generate diverse and realistic
network model with the aim to distinguish between healthy citrus disease images, overcoming issues of small and
fruits and leaves and those affected by prevalent citrus unevenly distributed datasets. The proposed EfficientNet-
diseases such as black spot, greening, canker, scab, and B5-pro incorporates Arcface loss and adversarial weight
melanose. The convolutional neural network architecture perturbation, enhancing its performance in disease
includes preprocessing, two convolutional layers, max- classification.it demonstrates that FastGAN2 can be trained
pooling, flattening, and classification. With a Citrus and with only 50 images, producing 8000 images with improved
PlantVillage datasets, the model demonstrated a quality. Classification networks, trained exclusively on
commendable test accuracy of 94.55%. images generated by FastGAN2, achieve high accuracy rates
exceeding 93%. EfficientNet-B5-pro exhibits superior
(Mudholakar et al., 2022) introduced a CNN model for performance compared to EfficientNet-B5, with accuracy,
the detecting citrus fruit and leaves diseases. The proposed scores of 97.04%.
system integrates various image processing steps and
employs a convolutional neural network to differentiate (Yadav et al., 2022) employed hyperspectral imaging
healthy citrus plant and the ones with diseases, such as along with an artificial intelligence algorithm to classify
canker, citrus blight and black spot. The convolutional eight distinct peel conditions on citrus fruit. The method
neural network model, inspired by AlexNet architecture, involved selecting discriminative bands using Principal
achieves promising training and validation accuracies above Component Analysis (PCA) and training a custom CNN
85% and 90%, respectively. The system was implemented modeled on the VGG-16 architecture designed for
using the Django framework for a user-friendly interface. classification. The PCA-selected bands achieved high
accuracy (99.84%), sensitivity (99.84%), and specificity
(Çetiner, 2022b) introduce A distinctive CNN-based (99.98%). Additionally, a CNN using randomly selected
architecture developed for the detection and classification of bands demonstrated slightly lower but still substantial
black spot, greening and canker diseases. Using a publicly performance achieving an accuracy of 98.87%.
available Citrus Leaves Prepared dataset and employs
preprocessing techniques, including histogram equalization (Dhiman et al., 2023) introduces an efficient model
and data augmentation, the proposed model achieved high design for predicting diseases in citrus fruit by fusing LSTM
average values of 95% for F1-score, 96% for Precision, 95% and CNN with edge computing. Using citrus dataset form
for Recall, and 96% for Accuracy. Kaggle with 2950 images, the proposed CNN-LSTM model
demonstrates superior disease detection capabilities. With
(Senthilkumar & Kamarasan, 2021) proposed a Magnitude Based Pruning and Post Quantization technique
segmentation process based on Otsu using an Inception the model attained an accuracy of 97.18% and 98.25%,
ResNet V2 based approach to detect and classify citrus The surpassing the baseline CNN method. The multi-phase
methodology was evaluated on the dataset from citrus image approach involves data collection, pre-processing, re-
gallery, showcasing a notable accuracy of 99.13%. The key scaling, data augmentation, and image segmentation.
processes involved in the proposed model are preprocessing, Evaluation parameters like accuracy, recall, precision and F-
segmentation, extracting features through the Inception score highlight the model's effectiveness across various
ResNet v2 model and performing classification using a disease classes, ranging from 94.65% to 99.05%.
Random Forest (RF) classifier.
(Faisal et al., 2023) introduced an automatic system
(Liu et al., 2021) used a mobileNet V2 as the main aimed at detection and classification disease of citrus plant,
network and conducting a comparative analysis with other utilizing deep learning models, particularly focusing on
networks based on model accuracy, speed and speed. The transfer learning. Four pre trained CNN models
study involves dataset enhancement techniques and (EfficientNetB3, MobiNetV2, ResNet50 and InceptionV3)
compares MobileNetV2's performance with other networks. were utilized to the citrus dataset, demonstrating improved

IJISRT24MAR1459 www.ijisrt.com 1823

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ij isrt/IJISRT24MAR1459

accuracy and reduced computational complexity. The Whale Optimization Algorithm, and outperforms individual
suggested methodology involves data collection, data models. The experiments involve a dataset of 279 images,
augmentation, image pre-processing, extracting deep CNN six disease classes, and various classifiers, demonstrating
features through transfer learning, and final classification. the efficacy of the proposed framework in enhancing citrus
The study identifies EfficientNetB3 as the most effective disease classification accuracy for practical agricultural
model, achieving training, validating, and testing accuracies applications.
of 99.43%, 99.48%, and 99.58%, respectively.
(Shireesha & Reddy, 2022) Suggested a DenseNet-121
(Shermila et al., 2024) Proposed a customized CNN- model with the objective of discerning between healthy
based model, integrating CNN with LSTM, to automatically citrus leaves and fruits and those afflicted by diseases such
detect and classify diseases in citrus fruits and leaves. The as black spot, greening, scab, and canker. The dataset
research focuses on identifying illnesses such as fruit blight, included images from the Plant Village Dataset and Citrus
fruit greening, fruit scab, and melanoses. The suggested Dataset, classified into black spot, canker, greening, scab,
model demonstrates excellent performance, achieving a 96% and healthy categories. By employing the DenseNet-121
accuracy in identifying and classifying citrus fruit and architecture and pre-training with ImageNet over 50 epochs,
leaves. The dataset encompasses both healthy and diseased the model achieved a commendable accuracy of 96%,
citrus leaves and fruits from diverse origins. The effectively tackling challenges like degradation and
architecture of CNN-LSTM incorporates convolutional vanishing gradient.
layers, max-pooling layers, fully connected layers, and an
LSTM layer. (Kalim et al., 2022) Introduced a hybrid model,
integrating a CNN for feature extraction and a Random
(Janarthan et al., 2020) Presented is a compact deep Forest (RF) for classification. The dataset comprised 58
metric learning architecture tailored for precise detection of images of healthy and 536 images of unhealthy instances
citrus diseases, specifically crafted for devices with limited involving Black Spot, Canker, and Greening. The VGG16-
resources such as mobile phones. The suggested framework Random Forest algorithm demonstrated optimal
comprises a classification network based on patches, performance, achieving an accuracy of 87%, outperforming
featuring an embedding module, cluster prototype module, ResNet50-Random Forest (83%) and InceptionV3-Random
and a neural network classifier. The process involves pre- Forest (80%). The model's flow involved CNN as a feature
processing, embedding calculation, K-Means clustering, and extractor, utilizing VGG16, ResNet50, and InceptionV3,
patch embedding computation. Evaluation on citrus and tea followed by RF for classification.
leaf datasets demonstrates the model's efficiency, achieving
a 95.04% detection accuracy. (Subramani et al., 2023) focuses on identifying and
categorizing diseases in citrus leaves using deep learning
(Jasim et al., 2022) Introduce a CNN model for architectures, including ResNet50, ResNet101, VGG16, and
recognizing citrus diseases using a dataset that comprises of VGG19. With a dataset of 609 images featuring Black Spot,
2450 sample of images with seven disease classes. The Greening, Canker, Melanose and Healthy classes, the pre
CNN structure comprises pooling, convolutional and fully trained ResNet50 achieves a very good accuracy of 96.8%
connected layers. The system achieves an 88% recognition and negligible loss (0.1573). The ResNet models, known for
accuracy for citrus diseases, and the study discusses the deep convolution and skip connections, prove powerful in
impact of optimization algorithms and data augmentation on addressing the vanishing gradient problem.
model performance.
(Challagundla et al., 2023) Proposed Inception V3,
(Huang et al., 2023) presented a proficient diagnostic VGG16, SqueezeNet and machine learning classifiers
system for citrus fruit diseases based on CNN. The model (Neural Network, Random Forest, KNN, Gradient Boosting,
integrates the Inception module with EfficientNetV2 to SGD). A dataset with 759 images (80 healthy, 679 diseased)
improve multi-scale feature extraction. VGG replaces the U- of citrus diseases is used. Image embedding is performed
Net backbone, resulting in enhanced segmentation using pre-trained CNN models, and machine learning
performance. Findings demonstrate a recognition accuracy classifiers are applied for disease prediction. The neural
exceeding 95%, with VGG-U-Net achieving the highest network utilizing Inception V3 attains the highest average
segmentation accuracy at 87.66%. The dataset encompasses accuracy of 96.6%.
800 images of citrus fruit affected by black spot and canker
diseases. Transfer learning is employed, utilizing pre-trained (Khan et al., 2021) proposed a deep learning approach
weights from ImageNet, to enhance classification to categorize six distinct citrus diseases affecting the yield
performance. and quality of citrus fruits. The dataset undergoes data
augmentation to enhance it, subsequently used for the
(Ur Rehman et al., 2021) proposed a pair of pre-trained retraining of two lightweight pre-trained deep learning
models, MobileNetV2 and DenseNet201 for the models: SqueezeNet and MobileNetV2. Features are
classification of citrus plant disease, image augmentation, extracted and optimized using the Whale Optimization
and feature fusion. The technique achieves a notable Algorithm (WOA). SqueezeNet surpasses MobileNetV2,
classification accuracy of 95.7%. It addresses data scarcity achieving a classification accuracy of 96%.
through image augmentation, optimizes features using the

IJISRT24MAR1459 www.ijisrt.com 1824

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ij isrt/IJISRT24MAR1459

IV. LIMITATIONS addressed a broader range of diseases. Additionally,

variations were observed in the choice of datasets, with
The primary goal of this study is to gain a some studies utilizing publicly available datasets, while
comprehensive grasp of the existing scenario in citrus others collected and curated their own datasets.
diseases detection and classification by employing deep
learning models. One notable limitation across the reviewed VI. CONCLUSION
research works is the reliance on limited datasets for training
and evaluation. Many studies faced challenges associated The significance of citrus cultivation in global
with dataset size, diversity, and representativeness. The use agriculture cannot be overstated, and the challenges posed
of small datasets may hinder the ability of deep learning by climate change, pests, and diseases underscore the need
models to generalize effectively to the wide range of citrus for innovative solutions. this literature review has provided a
disease scenarios encountered. comprehensive analysis of 20 research works focused on
citrus disease detection using deep learning models. By
Many studies faced challenges related to class examining the methodologies, key findings, and limitations,
imbalance in the datasets, where certain citrus diseases were several common themes and variations have been identified.
underrepresented. This imbalance can impact the model's the common themes observed across the research highlight
capability to sufficiently learn and distinguish between the versatility of deep learning in citrus disease detection,
different disease classes, leading to biased predictions and along with persistent challenges related to datasets and the
reduced overall performance. need for models to transition effectively from controlled
settings to real-world agricultural contexts. These findings
While data augmentation techniques were employed in provide insights for future research endeavors and the
some of the research to enhance dataset diversity, the impact advancement of more resilient and applicable solutions for
and effectiveness of these techniques were often not the management of citrus diseases.
rigorously evaluated. The extent to which data augmentation
contributes to improved model generalization and ACKNOWLEDGEMENTS
performance remains an area of concern and requires further
investigation. We would like to express our gratitude to all the
researchers and authors whose work contributed to the body
A common limitation identified is the gap between of knowledge analyzed in this review. Their dedication and
experimental results and practical implementation in real- contributions have been invaluable in advancing the field of
world agricultural environments. The transition from citrus disease detection and classification using deep
controlled experimental setups to dynamic field conditions learning models.
poses challenges in terms of scalability, adaptability, and
robustness of the proposed models. We also acknowledge the support and assistance of the
academic and research communities, as well as the
V. DISCUSSION institutions and organizations that provided resources and
funding for this research endeavor.
One prominent similarity among the reviewed papers
is the widespread adoption of deep learning architectures for REFERENCES
citrus disease detection. Various models, including
convolutional neural networks (CNNs), DenseNets, [1]. Agusti, M., Mesejo, C., Reig, C., & Martinez-
Inception modules, and YOLO classifiers, were employed Fuentes, A. (2013). Citrus production. In
across different studies. These architectures demonstrate the Horticulture: Plants for People and Places (Vol. 1, pp.
versatility and effectiveness of deep learning in handling 159–195). Springer Netherlands.
complex image data and extracting meaningful features for https://doi.org/10.1007/978-94-017-8578-5_6
disease classification. [2]. Çetiner, H. (2022a). Citrus disease detection and
classification using based on convolution deep neural
Preprocessing techniques also emerged as a common network. Microprocessors and Microsystems, 95.
theme across the reviewed papers. Authors consistently https://doi.org/10.1016/j.micpro.2022.104687
applied resizing, standardization, and data augmentation to [3]. Çetiner, H. (2022b). Citrus disease detection and
enhance the quality and diversity of the input data. Data classification using based on convolution deep neural
augmentation, in particular, played a crucial role in network. Microprocessors and Microsystems, 95,
artificially increasing the diversity of the training dataset, 104687.
thereby addressing concerns related to overfitting and https://doi.org/10.1016/J.MICPRO.2022.104687
improving the generalization capability of the models. [4]. Çeti̇ ner, H., Yazar, S., & Author, C. (2023). CITRUS
DISEASE CLASSIFICATION WITH TRANSFER
However, despite these commonalities, notable LEARNING AND CNN BASED MODELS
differences were observed in the methodologies and TRANSFER ÖĞRENME VE CNN TABANLI
experimental setups employed by the reviewed studies. For MODELLER İLE NARENCİYE HASTALIĞI
instance, some studies focused on specific citrus diseases, SINIFLANDIRMASI. In KSU J Eng Sci (Vol. 26,
such as black spot, canker, or greening, while others Issue 1).

IJISRT24MAR1459 www.ijisrt.com 1825

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ij isrt/IJISRT24MAR1459

[5]. Challagundla, Y., Chintalapati, L. R., Tunuguntla, T. [16]. Khan, E., Zia Ur Rehman, M., Ahmed, F., & Khan,
S. C., Sur, A., Roy, B., & Zhuo, E. R. (2023). M. A. (2021). Classification of diseases in citrus
Screening of Citrus Diseases Using Deep Learning fruits using SqueezeNet. ICAEM 2021 - 2021
Embedders and Machine Learning Techniques. 2023 International Conference on Applied and Engineering
3rd International Conference on Artificial Mathematics, Proceedings, 67–72.
Intelligence and Signal Processing, AISP 2023. https://doi.org/10.1109/ICAEM53552.2021.9547133
https://doi.org/10.1109/AISP57993.2023.10134971 [17]. Khattak, A., Asghar, M. U., Batool, U., Asghar, M.
[6]. Christopher, I. C., & Udoh, E. (n.d.). AN Z., Ullah, H., Al-Rakhami, M., & Gumaei, A. (2021).
ASSESSMENT OF CITRUS FARMING IN Automatic Detection of Citrus Fruit and Leaves
NIGERIA. In International Journal of Recent Diseases Using Deep Neural Network Model. IEEE
Research in Social Sciences and Humanities Access, 9, 112942–112954.
(IJRRSSH) (Vol. 7). www.paperpublications.org https://doi.org/10.1109/ACCESS.2021.3096895
[7]. Dai, Q., Guo, Y., Li, Z., Song, S., Lyu, S., Sun, D., [18]. Liu, Z., Xiang, X., Qin, J., Tan, Y., Zhang, Q., &
Wang, Y., & Chen, Z. (2023). Citrus Disease Image Xiong, N. N. (2021). Image recognition of citrus
Generation and Classification Based on Improved diseases based on deep learning. Computers,
FastGAN and EfficientNet-B5. Agronomy, 13(4). Materials and Continua, 66(1), 457–466.
https://doi.org/10.3390/agronomy13040988 https://doi.org/10.32604/cmc.2020.012165
[8]. Dhiman, P., Kaur, A., Hamid, Y., Alabdulkreem, E., [19]. Luaibi, A. R., Salman, T. M., & Miry, A. H. (2021).
Elmannai, H., & Ababneh, N. (2023). Smart Disease Detection of citrus leaf diseases using a deep learning
Detection System for Citrus Fruits Using Deep technique. International Journal of Electrical and
Learning with Edge Computing. Sustainability Computer Engineering, 11(2), 1719–1727.
(Switzerland), 15(5). https://doi.org/10.11591/ijece.v11i2.pp1719-1727
https://doi.org/10.3390/su15054576 [20]. Mohanty, S. P., Hughes, D. P., & Salathé, M. (2016).
[9]. Elaraby, A., Hamdy, W., & Alanazi, S. (2022). Using deep learning for image-based plant disease
Classification of Citrus Diseases Using Optimization detection. Frontiers in Plant Science, 7(September).
Deep Learning Approach. Computational Intelligence https://doi.org/10.3389/fpls.2016.01419
and Neuroscience, 2022. [21]. Mosavi, A., Ardabili, S., & Varkonyi-Koczy, A. R.
https://doi.org/10.1155/2022/9153207 (n.d.). List of Deep Learning Models.
[10]. Faisal, S., Javed, K., Ali, S., Alasiry, A., Marzougui, https://doi.org/10.20944/preprints201908.0152.v1
M., Khan, M. A., & Cha, J. H. (2023). Deep Transfer [22]. Mudholakar, S., G, K., K T, K. K., & G V, S. (2022).
Learning Based Detection and Classification of Automatic Detection of Citrus Fruit and Leaves
Citrus Plant Diseases. Computers, Materials and Diseases Using Deep Neural Network. International
Continua, 76(1), 895–914. Journal for Research in Applied Science and
https://doi.org/10.32604/cmc.2023.039781 Engineering Technology, 10(7), 4043–4051.
[11]. Firdaus, M. H., Utami, E., & Ariatmanto, D. (2023). https://doi.org/10.22214/ijraset.2022.45868
Detection And Classification of Citrus Diseases [23]. Raaj, M. S., & Selvy, D. (n.d.). Detection of Citrus
Based on A Combination of Features Using the Leaves and Fruit Diseases Using YOLO.
Densenet-169 Model. Sinkron, 8(4), 2592–2601. [24]. Senthilkumar, C., & Kamarasan, M. (2021). An
https://doi.org/10.33395/sinkron.v8i4.12974 Effective Citrus Disease Detection And Classification
[12]. Huang, Z., Jiang, X., Huang, S., Qin, S., & Yang, S. Using Deep Learning Based Inception Resnet V2
(2023). An efficient convolutional neural network- Model. In Turkish Journal of Computer and
based diagnosis system for citrus fruit diseases. Mathematics Education (Vol. 12, Issue 12).
Frontiers in Genetics, 14. [25]. Shermila, J. P., Victor, A., Manoj, S. O., & Devi, E.
https://doi.org/10.3389/fgene.2023.1253934 A. (2024). Automatic detection and classification of
[13]. Janarthan, S., Thuseethan, S., Rajasegarar, S., Lyu, disease in citrus fruit and leaves using a customized
Q., Zheng, Y., & Yearwood, J. (2020). Deep metric CNN based model. In Boletin Latinoamericano y del
learning based citrus disease classification with Caribe de Plantas Medicinales y Aromaticas (Vol. 23,
sparse data. IEEE Access, 8, 162588–162600. Issue 2, pp. 180–198). MS-Editions.
https://doi.org/10.1109/ACCESS.2020.3021487 https://doi.org/10.37360/blacpma.24.23.2.13
[14]. Jasim, W. N., Almola, S. A. S., Alabiech, M. H., & [26]. Shireesha, G., & Reddy, B. E. (2022). Citrus Fruit
Harfash, E. J. (2022). Citrus Diseases Recognition by and Leaf Disease Detection Using DenseNet. 2022
Using CNN Model. Informatica (Slovenia), 46(7), International Conference on Smart Generation
85–94. https://doi.org/10.31449/inf.v46i7.4284 Computing, Communication and Networking,
[15]. Kalim, H., Chug, A., & Singh, A. P. (2022). Citrus SMART GENCON 2022.
Leaf Disease Detection Using Hybrid CNN-RF https://doi.org/10.1109/SMARTGENCON56628.202
Model. AIST 2022 - 4th International Conference on 2.10083852
Artificial Intelligence and Speech Technology.
https://doi.org/10.1109/AIST55798.2022.10065093

IJISRT24MAR1459 www.ijisrt.com 1826

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ij isrt/IJISRT24MAR1459

[27]. Subramani, S., Sakthivel, B., George, N. T., & Raj, T.

A. (2023). Citrus Leaf Disease Detection Using
Various Deep Learning Architectures. 2023
International Conference on Sustainable
Communication Networks and Application
(ICSCNA), 1425–1430.
https://doi.org/10.1109/ICSCNA58489.2023.103702
98
[28]. Syed-Ab-Rahman, S. F., Hesamian, M. H., & Prasad,
M. (2022). Citrus disease detection and classification
using end-to-end anchor-based deep learning model.
Applied Intelligence, 52(1), 927–938.
https://doi.org/10.1007/s10489-021-02452-w
[29]. Ur Rehman, M. Z., Ahmed, F., Khan, M. A., Tariq,
U., Jamal, S. S., Ahmad, J., & Hussain, I. (2021).
Classification of citrus plant diseases using deep
transfer learning. Computers, Materials and
Continua, 70(1), 1401–1417.
https://doi.org/10.32604/cmc.2022.019046
[30]. Yadav, P. K., Burks, T., Frederick, Q., Qin, J., Kim,
M., & Ritenour, M. A. (2022). Citrus disease
detection using convolution neural network generated
features and Softmax classifier on hyperspectral
image data. Frontiers in Plant Science, 13.
https://doi.org/10.3389/fpls.2022.1043712

IJISRT24MAR1459 www.ijisrt.com 1827