PLANT DISEASE DETECTION VIA DEEP LEARNING

Poonam Jyoti Namnish Dogra Banshika Salaria

Chandigarh University Chandigarh University Chandigarh University
Punjab, India Punjab, India Punjab, India
[email protected] [email protected] [email protected]

Manvi Singh Tushar Sood Sahil

Chandigarh University Chandigarh University Chandigarh University
Punjab, India Punjab, India Punjab, India
[email protected] [email protected] [email protected]

ABSTRACT plant disease prevention and treatment, which in turn ensures food
security and agricultural sustainability. Disease identification
Plant diseases pose a serious threat to the world's food supply thrives on a person's cognitive capacity in the conventional ways
because they reduce crop yields and result in financial losses. In the
past, diagnosing diseases has frequently relied on regular people, of disease detection. It's the world's hardest thing. These
usually skilled professionals, and their ability to visually check plants. procedures are not only labour-intensive and time-consuming, but
But this is frequently wrong, time-consuming, and completely they also carry the risk of error, particularly when analysing tiny
inappropriate for situations requiring extensive monitoring. This study
or large amounts of visual evidence. Another difficult problem
examines the use of deep learning models, namely convolutional
neural networks (CNNs), to automatically and precisely identify plant occurs in a small or underdeveloped region where some people
disease images on leaves. To train and assess the network models, a lack access to the profession of disease identification, delaying
neural network is used to process a specific set of graphics (mostly the prompt identification of plant diseases and leading to
pictures of plant leaves), including both healthy and sick examples. The
models are in charge of first assimilating visual patterns and then using
erroneous treatment and increased crop damage. Rapid
that information to accurately identify the type of sickness or if a leaf advancements in computer vision and artificial intelligence,
is healthy. Among the methods used to enhance model performance especially deep learning, have made it possible to build
and generalize the models include data augmentation, transfer autonomous models that can correctly identify and classify plant
learning, and hyperparameter tuning. Visual models in the suggested
deep learning approach are very accurate, have deep robustness, and illnesses from photos of leaves. Convolutional Neural Networks,
operate in real-time, making them ideal for the issue of early plant or CNNs, are a particular deep learning model in machine
disease identification. The model introduces new possibilities, such as learning that has shown promise in image identification tasks and
stopping farmers and the agricultural sector from getting by with the
may be adjusted to extract subtle patterns and nuances from
same and giving crops a fresh lease on life, with sufficient but
unidentifiable assistance from the AI. More precisely, the deep photos of plant leaves. The goal of this research is to use the deep
learning system discussed in this paper is far superior to deep learning learning model that has been constructed to leaf photos in order
systems for agricultural applications and serves as the foundation for to identify plant illnesses without the need for human interaction
the integration of AI-based solutions into precision farming tools, such
as drone-based surveillance systems and mobile applications. and, as a result, to accurately distinguish between healthy and
diseased leaves. Its main goal is to give a solution that is both
Keywords: Deep Learning, Convolutional Neural Networks realistic and practical in terms of its quality, portability, and
(CNN), Crop Health Monitoring, Transfer Learning, Smart Farming,
widespread use. This will allow all farmers, agronomists, and
Computer Vision in Agriculture.
even those who supply agricultural services to identify diseases
I. INTRODUCTION more rapidly. The system can also be incorporated into mobile
applications or smart agricultural devices. It can provide a timely
A. Problem Definition: and real-time diagnosis of diseases, lowering the need for human
It is common knowledge that agriculture is essential to the intervention and facilitating prompt preventive measures, which
survival of the world's population and economy. However, plant ultimately improve crop health and agricultural efficiency.
diseases continue to rank among the primary variables that B. Problem Overview
adversely impact agricultural output, raising the possibility of Locally and globally, plant diseases pose a serious challenge to
achieving lower crop yields and quality. Up to 40% of the world's reaching the maximum levels of agricultural output and food
food supply lost each year due to plant disease , according to the security. They have the ability to impact not only the harvest's
Food, good supply is lost each year due to plant diseases, quantity but also its quality, which results in significant financial
according to the Food and Agriculture Organization (FAO). Early losses for farmers and other industry participants. Therefore,
detection and accurate diagnosis are key components of effective being able to promptly and precisely identify plant diseases is
 essential for both crop management and preventing crop loss in standard for the past 20 to 30 years. The days of cultivating by
a timely manner. The majority of disease diagnosis techniques hand are long gone. a reality that everyone now acknowledges.
currently in use include manual plant examination by agricultural People are now employing faster and more advanced methods of
professionals, which is essentially costly, time-consuming, and weather forecasting, and it is widely acknowledged that the old
prone to mistakes, particularly when dealing with big farmed methods of identifying plant diseases are no longer
areas. Many farmers, particularly those from remote and valid. Traditional scouting techniques are not consistent with the
underdeveloped areas, lack the necessary personnel and area and are likely to be impacted by observer bias, according to
equipment, such as diagnostic tools, which can lead to Bock et al. (2010). Nevertheless, these manual examinations need
inconsistent and erroneous disease identification. Similar to this, not just the observer's entire day, but also a significant degree of
certain illnesses may present with extremely mild visual signs skill to distinguish between visual signs of related
that are difficult to detect until the damage has become quite illnesses. Furthermore, Mahllein (2016) described in detail how
severe. Thus, the current circumstances are presented here as imaging technologies and spectrum sensors are used to improve
issues that can only be resolved by a solution that is accurate and disease prediction. They may undoubtedly function well in a
efficient enough to be implemented in a small- or large-scale controlled setting, but these systems are always expensive and
agricultural setup. the beginnings and growth of artificial somewhat complex, which ultimately prevents small farmers
intelligence (AI), with a particular emphasis on computer vision from using them. Rule-based and machine-learning methods have
and deep learning? results in the development of methods that been developed in response to the obstacles. Although Camargo
promise extremely precise and dependable automation for plant and Smith (2009) proposed a method for classification using
disease diagnosis. The concept of using drones to deal with this colour and form characteristics present in illness images, the
circumstance should be considered since enemy identification outcomes were unsatisfactory. It is difficult in the actual world
may be completed quickly. In particular, leaf analysis using because of the different sounds and light coming from different
photos of healthy and sick plants was described as a very directions, which leads to confusion. The introduction of plant
effective yet straightforward method of identifying the disease. disease detection applications for mobile devices, such as Plant or
In order to stress the plants or make microbes more aggressive in Leaf Doctor, was significant. These applications share cloud-
areas where they already reside, farmers can expose the infected based databases for diagnosis and are classified as very simple
plants to high, unusual temperatures (such as extremely low image classification algorithms. However, Ferentinos (2018)
nighttime temperatures) that are employed for crops. However, asserts that the applications have yet to achieve the required level
this has no physiological effect on the crop. However, the of accuracy due to a number of shortcomings, such as the limited
technical and practical challenges limit the potential for self- data and learning that is limited to a single domain, which makes
regulation. The objective of this research is to implement a deep it challenging to generalize information about plants regardless of
learning-based smart leaf model that can be used in the field to their species and the constantly changing
detect plant diseases. Using sets of open-source data, the model environments. Additionally, these apps require a fast and
will be trained and evaluated under real-world settings, and uninterrupted internet connection to function, so the systems rely
optimization techniques will be used to further improve heavily on a stable internet connection. Deep learning-based
performance. As the farming business enters a new era, the models that can overcome these particular limits by learning from
implementation of such a system will be clearly seen in drones, image datasets have been further explored in research laboratories
smartphone apps, and IoT-enabled gadgets. The goal of these as a result of the development of more sophisticated, potent, and
new technological innovations is to help small-scale farmers so real-time detection technologies.
that they may use agriculture not only to grow food but also to
make money when they sell their greens at the market. B. Drawback of Current System:
II. LITERATURE REVIEW
1. Manual Dependency: Farmers or agricultural experts still
A. Current System
mostly inspect crops by hand in order to identify diseases,
The current system shows the latest and present version of the data
which is time-consuming and subjective.
Although laboratory testing has occasionally been used to help,
2. Lack of Accuracy: Human error might occur while using
eye observation has always been the primary method of plant
traditional procedures. Similar-symptom diseases are
disease identification throughout its conventional stage, carried
frequently misdiagnosed, leading to inappropriate treatment.
out mostly by farmers or other agriculture specialists. In the
3. Expert Unavailability: Trained agricultural specialists are
meantime, the demands of contemporary large agriculture are no
difficult to find in the majority of rural and remote areas of
longer consistent with these two practices, which have been the
 the nation, which causes delays in the detection and Res-Net, Inception, VGG, or Mobile-Net is used. The model
management of diseases. is trained and validated using a sizable and diverse dataset,
4. High Cost of Laboratory Testing: For small farmers, sending such as Plant-Village or custom-collected datasets.
samples to labs for analysis can be expensive and difficult. 4.Classification and Prediction of Diseases: The trained model
5. Scalability Issues: In vast agricultural plots or farms, it is not correctly predicts the type of illness (such as blight, rust, or
feasible or feasible for an expert to perform a manual leaf spot) or state of health. Additionally, a confidence score
inspection or diagnosis. that indicates the outcome's validity is returned.
6. Late Detection: The disease may have spread by the time 5.Offline Capability (Optional): By optimizing the system for
symptoms are apparent and manually identified, which on-device inference, it can be used offline in places with poor
reduces the likelihood of recovery. connectivity, such as rural areas.
Existing rule-based systems and mobile apps rely on basic Early detection reduces crop damage and increases productivity
algorithms and frequently fail in real-world scenarios due to by enabling prompt treatment: The objectives of precision
complicated image backgrounds, illumination, or ambient agriculture and sustainable farming methods are in line with this
noise. suggested approach, which turns plant disease detection into an
7. Internet Dependency: The majority of mobile-based intelligent, automated, and trustworthy procedure.
solutions require continuous internet access, which isn't
usually available in rural agricultural areas. FLOWCHART OF PROPOSED SYSTEM
8. Absence of Real-Time Monitoring: Existing systems are
unable to continuously monitor plant health due to a lack of
automatic or real-time monitoring solutions.
9. Low Generalization Capability: When exposed to new or
unexplored data, the majority of current models or apps
perform poorly because they were trained on certain crops
or disease kinds.
C. Proposed System:
We present an automated, scalable, and intelligent system that
uses deep learning algorithms, specifically Convolutional Neural
Networks (CNNs), to automatically detect and identify plant
diseases based on leaf images, thereby overcoming the drawbacks
of traditional and current digital plant disease diagnosis
approaches. The system is perfect for practical agricultural
applications, especially in remote or resource-constrained areas,
because it is designed to be accessible through cellphones or field
devices with IoT support. The planned system's main goal is to
develop a reliable and accurate model that can identify plant
illnesses early on, allowing for prompt intervention and treatment.
By providing farmers, agronomists, and other agricultural
stakeholders with an affordable, real-time diagnostic tool, the
system will reduce the need for expert time and physical work. FIGURE 1
1.Obtaining Images: Mobile phones, drones, or field-based The "Traditional Image Processing Recognition" diagram shows
stationary IoT devices are used to collect high-resolution leaf how plant disease detection systems typically operate. Image
photos. The system supports real-time image input, allowing acquisition, image processing, feature extraction, and
for ongoing monitoring. identification and classification are the four primary steps in this
method's sequential pipeline. The procedure starts with image
2.Image Preprocessing: To enhance model accuracy and
acquisition, where the main input is pictures of plant leaves taken
improve image quality, the obtained images go through a with cameras or smartphones. To separate the leaf from the
number of preprocessing stages. This comprises: background and boost image quality, these photos are
3.Deep Learning Model (CNN Architecture): To extract subsequently subjected to image processing, which includes
hierarchical features from photos and categorize them as processes like noise reduction, segmentation, and enhancement.
either healthy or disease-specific, a CNN model customized The next step is feature extraction, which involves manually
locating and extracting pertinent characteristics including colour,
for the application or a pre-trained transfer learning model like
shape, texture, and edge patterns that aid in differentiating dataset, the model then allocates the input image to the
between plants that are healthy and those that are unhealthy. The appropriate class label, which may include certain illnesses like
last step is identification and classification, where the disease is rust, mosaic virus, bacterial blight, powdery mildew, or others.
categorized using the retrieved features using conventional This automated, end-to-end procedure reduces human error and
machine learning methods like Support Vector Machines (SVM), allows for real-time diagnosis while improving speed and
k-Nearest Neighbours (k-NN), or decision trees. accuracy in the identification of plant diseases.

FLOWCHART DIAGRAM OF THIS MODEL D. Literature Review Summary in table

TABLE I. SUMMARY OF LITERATURE REVIEW:

Conferences Year Title Findings
Introduced a large
IEEE
PlantVillage: dataset
Conference
A Benchmark (PlantVillage),
on Computer
Dataset for used CNNs for
Vision and 2018
Image-Based image
Pattern
Plant Disease classification,
Recognition
Detection open-source
(CVPR)
dataset.

Deep Learning Review of deep

IEEE
for Plant learning models
International
Disease (CNN, RNN),
Conference 2019
Detection and comparisons,
on Big Data
Diagnosis – A challenges, and
(BigData)
Review future directions.

FIGURE 2
Using pictures of plant leaves, the provided architecture
demonstrates a deep learning-based system for identifying plant Tomato Leaf
diseases. As the main source of data for analysis, the procedure International Disease Used CNNs on
begins with taking or uploading an input image of a leaf. After Conference Detection tomato leaf
that, the picture goes through a feature extraction procedure that on Intelligent using dataset, achieved
identifies important visual elements like texture, colour patterns, 2020
Sustainable Convolutional high accuracy,
leaf form, and edge structures. These collected characteristics are Systems Neural trained using
fed into a Convolutional Neural Network (CNN) model and are (ICISS) Networks Keras/TensorFlow.
essential for differentiating between healthy and sick leaves. The
central component of the system is the CNN model, which is
trained on a carefully selected dataset of annotated photos of both
healthy and sick plant leaves. Through a sequence of
convolutional and pooling layers, this training allows the model
to automatically learn hierarchical representations of the input
features. Low-level to high-level patterns, such spots, IEEE Real-Time
Mobile
discolouration, or shape abnormalities frequently linked to plant International Plant Disease
deployment, edge-
illnesses, are gradually captured by these layers. The image is Conference Detection
device optimized
converted into a rich collection of feature maps that contain both on Machine 2021 Using Mobile-
CNNs, real-time
structural and spatial information as it passes through the CNN Learning and Based Deep
detection and
layers. A dense (completely connected) layer receives the output Applications Learning
classification.
from the convolutional layers after that. In order to further refine (ICMLA) Models
the extracted characteristics and get them ready for the
classification stage, this layer is essential. The dense layer creates
a condensed but useful high-level representation by combining
and weighing many leaf characteristic elements. They are saved
A key component in enhancing the precision, effectiveness, and
as distinct "Leaf Features," which are necessary for precise
responsiveness of deep learning-based plant disease detection
categorization. Using these leaf characteristics, the system
systems is feature selection. The model's capacity to correctly
performs categorized detection in the last step, identifying
identify and diagnose plant illnesses is significantly impacted by
whether the leaf is disease-free or healthy. Depending on the
the selection of the most crucial features in agricultural settings 2. The manual feature extraction used in current image
where vast volumes of leaf image data are analyzed. Selecting processing techniques is inaccurate and not flexible
features well enables deep learning models to detect plant enough for real-world field scenarios.
abnormalities with reduced computational overhead and 3. An efficient, scalable, and real-time deep learning-
enhanced processing efficiency. This is especially crucial when based system that can autonomously detect plant
using these models in low-resource, real-time settings, such as diseases in a variety of species and settings while using
mobile or edge devices used in agricultural fields. Feature little computational resources is needed.
selection lowers the input to convolutional neural networks Therefore, the main goal of
Four types of disease in our Dataset
(CNNs) by removing redundant or unnecessary information, problem formulation is to create
which improves model interpretability, speeds up training, and an accurate and lightweight deep learning model that can
uses less memory. In addition to enhancing detection automatically detect and categorize plant diseases in order to
performance, this makes practical application easier in actual improve crop yield, enable early diagnosis, and support
agricultural settings, where environmental variability and sustainable agriculture practices through clever and
hardware limitations are major problems. By selecting the right understandable solutions.
characteristics, such as texture, color patterns, lesion forms, and
spot distribution, deep learning models can be made to be reliable IV. OBJECTIVES/FUTURE SCOPE
and accurate even in the face of changing lighting, leaf
orientations, and background clutter. Therefore, feature selection Building a clever, reliable, and approachable deep learning model
is a key component of the design of intelligent, scalable, and that can automatically recognize and categorize plant illnesses
reasonably priced plant disease detection systems that can support from leaf photos is the main goal of this project. The goal is to
smart and sustainable agriculture. assist farmers and agricultural specialists in recognizing issues
early on, before they result in significant harm. This method aims
III. PROBLEM FORMULATION to improve plant disease detection speed and availability, even in
remote or resource-poor areas, by combining the power of deep
Plant diseases continue to pose a serious threat to global learning and computer vision. In the end, this work promotes
agriculture, resulting in massive crop losses and affecting food better harvests, healthier plants, and more intelligent, sustainable
security. Traditional methods of diagnosing diseases rely mostly farming.
on visual inspection and human experience, which are typically 1. To create an intelligent and accurate deep learning model for
labor-intensive, inconsistent, and unsuitable for large-scale automatically identifying plant diseases from photos of
farming. Additionally, traditional image processing methods are leaves.
ineffective because to their reliance on human feature extraction 2. To minimize agricultural losses and increase output by
and their limited capacity to adapt to real-world field settings, assisting farmers in identifying plant diseases early on.
such as varying leaf textures, crowded backgrounds, and 3. To create a scalable and user-friendly system that can be
changing lighting. Plant disease identification and categorization implemented in real farming environments despite resource
using leaf photos may become automated with the advancement limitations.
of deep learning, especially convolutional neural networks 4. Increase the model's scale to allow for real-time detection
(CNNs). Although deep learning has potential, existing through mobile applications for use in the field.
approaches suffer from a variety of issues, such as the V. EXPERIMENTAL RESULT
computational cost of models, non-real-time detection, and the
difficulty of deploying on devices with minimal resources, such
as smartphones or edge devices. Furthermore, developing a strong
and generalizable detection system is made more difficult by the
diversity of plant species, disease presentations, and
environmental factors. Therefore, the main challenge is to create
a deep learning-based model that is accurate, scalable, and
efficient so that it can automatically identify plant diseases with
high accuracy even under challenging circumstances. To enable
farmers to identify diseases early and take timely action, the
model should require less human participation, be designed for
low-power devices, and operate in real-time, resulting in more
intelligent and sustainable farming methods.Extremely precise
and robust in field settings
1. Traditional plant disease detection techniques are labor-
intensive, time-consuming, and not scalable for large-
scale agricultural monitoring.
Five types of Detection Result
Five different categories' contributions to the total are graphically agriculture and offers a viable way to accurately identify a variety
segmented in the pie chart on the left. As can be observed, the of plant illnesses.
blue sector is the most prevalent category in the data, taking up VI. CONCLUSION
the most portion at 43%. This means that this one group accounts In this study, we presented a hybrid deep learning framework for
for almost half of whatever is being summed. After that, the the effective and precise identification of plant diseases using
orange segment has 29%, which is still a significant amount but photos of leaves. The model effectively and more accurately
clearly less than the blue. These two together make up more than identifies illness classes by combining the benefits of CNN and
70% of the chart, indicating that these two categories alone hold Bi-GRU network architectures in deep feature extraction with a
the majority of the attention or data points. The yellow and gray Support Vector Machine (SVM) for classification. The Bi-GRU
sections, which each make up only 14%, are the same size at the determines sequential linkages between the visual information,
same moment. Since these are the smallest components, it follows whereas the CNN determines local structures and spatial
that these categories have less of an impact on the dataset or information from the leaf textures. The model can learn powerful
analysis that is being given. Since it shows that the data is not and expressive feature representations because to this
distributed uniformly, this type of distribution is essential to combination. According to our experimental findings, the hybrid
understand. If the illness frequency in plants were represented by model outperforms both independent machine learning and deep
this chart, we would know that certain diseases are far more learning models in terms of accuracy, recall, and precision. The
common than others and that our models or systems might need approach has the potential to be used in actual agricultural
to be specially configured to account for these imbalances. systems, giving farmers and agronomists a useful tool for tracking
crop health and early disease detection. In order to build a more
TABLE 1: Evaluation matrix for result comprehensive plant disease diagnosis system, next work will
ML Precision Recall F1 Score Accuracy focus on improving the model's generalization with more diverse
Classifier datasets, enabling real-time detection through mobile devices, and
SVM 81.45 78.36 79.67 78.97 including other contextual information like weather and soil
RF 79.42 75.66 77.02 76.25 conditions.
NB 61.76 77.34 22.33 45.45 REFERENCES
DT 12.34 69.78 77.09 98.0
Using the feature map of dimension (18210, 300) as the input 1. Mohanty, S. P., Hughes, D. P., & Salathé, M. (2016). Using
layer for our deep learning models, the features utilized in this deep learning for image-based plant disease
study were extracted using pretrained picture embeddings. A DL detection. Frontiers in plant science, 7, 215232.
architecture designed especially for plant leaf image processing
2. Mohanty, Sharada P., David P. Hughes, and Marcel Salathé.
was fed these padded image vectors. We evaluated the accuracy
of several deep learning models, as shown in Table 3. Our final "Using deep learning for image-based plant disease
hybrid methodology combines the performance advantages of detection." Frontiers in plant science 7 (2016): 215232.
deep learning (DL) and traditional machine learning (ML). We 3. Ahmed, I., & Yadav, P. K. (2023). Plant disease detection
introduce our hybrid design, which combines a Bidirectional using machine learning approaches. Expert Systems, 40(5),
Gated Recurrent Unit (Bi-GRU) and a Convolutional Neural e13136.
Network (CNN). After concatenating the latent feature
4. Yadav PK. Plant disease detection using machine learning
representations of the two networks, we transmit the combined
network output to an SVM classifier. When it comes to detecting approaches. Expert Systems. 2023 Jun;40(5):e13136.
and categorizing plant diseases, our multi-phase workflow 5. Panchal, A. V., Patel, S. C., Bagyalakshmi, K., Kumar, P.,
significantly improves model performance. Even though Khan, I. R., & Soni, M. (2023). Image-based plant diseases
particular ML or DL models, such CNN or Bi-GRU, produce detection using deep learning. Materials Today:
good results, they all have disadvantages. ML models are less Proceedings, 80, 3500-3506.
effective with high-dimensional, geographical data, but they 6. Ihtiram Raza Khan, and Mukesh Soni. "Image-based plant
typically perform well with structured characteristics. Despite
diseases detection using deep learning." Materials Today:
their strength, DL models can be computationally demanding and
have a tendency to overfit in the absence of a large dataset. By Proceedings 80 (2023): 3500-3506.
merging the two, we leverage CNN's feature extraction 7. Chowdhury, M. E., Rahman, T., Khandakar, A., Ayari, M. A.,
capabilities and Bi-GRU's sequential analytic capabilities. This Khan, A. U., Khan, M. S., ... & Ali, S. H. M. (2021).
combination records temporal consistency among leaf datasets as Automatic and reliable leaf disease detection using deep
well as spatial patterns (such texture, color variations, and leaf learning techniques. AgriEngineering, 3(2), 294-312.
spots). While the Bi-GRU recovers interdependencies between 8. Sarkar, C., Gupta, D., Gupta, U., & Hazarika, B. B. (2023).
picture segments, the CNN layers extract fine-grained spatial
Leaf disease detection using machine learning and deep
information from the diseased areas. After being fused in a fusion
layer, these are put into the SVM, which uses optimal hyperplanes learning: Review and challenges. Applied Soft
to identify different illness categories. This hybrid pipeline Computing, 145, 110534.
contributes to the broader goal of intelligent and automated
9. Sarkar, C., Gupta, D., Gupta, U., & Hazarika, B. B. (2023). using plant village. Journal of Computer and
Leaf disease detection using machine learning and deep Communications, 8(6), 10-22.
learning: Review and challenges. Applied Soft 22. Manjula, K., Spoorthi, S., Yashaswini, R., & Sharma, D.
Computing, 145, 110534. (2021, November). Plant disease detection using deep
10. Ahmad, A., Saraswat, D., & El Gamal, A. (2023). A survey learning. In ICDSMLA 2020: Proceedings of the 2nd
on using deep learning techniques for plant disease diagnosis International Conference on Data Science, Machine Learning
and recommendations for development of appropriate and Applications (pp. 1389-1396). Singapore: Springer
tools. Smart Agricultural Technology, 3, 100083. Singapore.
11. Ferentinos, K. P. (2018). Deep learning models for plant 23. Akila, M., & Deepan, P. (2018). Detection and classification
disease detection and diagnosis. Computers and electronics in of plant leaf diseases by using deep learning
agriculture, 145, 311-318. algorithm. International Journal of Engineering Research &
12. Jackulin, C., & Murugavalli, S. J. M. S. (2022). A Technology (IJERT), 6(7), 1-5.
comprehensive review on detection of plant disease using 24. Ouhami, M., Hafiane, A., Es-Saady, Y., El Hajji, M., &
machine learning and deep learning Canals, R. (2021). Computer vision, IoT and data fusion for
approaches. Measurement: Sensors, 24, 100441. crop disease detection using machine learning: A survey and
13. Li, L., Zhang, S., & Wang, B. (2021). Plant disease detection ongoing research. Remote Sensing, 13(13), 2486.
and classification by deep learning—a review. IEEE 25. Binnar, V., & Sharma, S. (2023, January). Plant leaf diseases
Access, 9, 56683-56698. detection using deep learning algorithms. In Machine
14. Türkoğlu, M., & Hanbay, D. (2019). Plant disease and pest Learning, Image Processing, Network Security and Data
detection using deep learning-based features. Turkish Journal Sciences: Select Proceedings of 3rd International Conference
of Electrical Engineering and Computer Sciences, 27(3), on MIND 2021 (pp. 217-228). Singapore: Springer Nature
1636-1651. Singapore.
15. Trivedi, J., Shamnani, Y., & Gajjar, R. (2020). Plant leaf 26. Jiang, P., Chen, Y., Liu, B., He, D., & Liang, C. (2019). Real-
disease detection using machine learning. In Emerging time detection of apple leaf diseases using deep learning
Technology Trends in Electronics, Communication and approach based on improved convolutional neural
Networking: Third International Conference, ET2ECN 2020, networks. Ieee Access, 7, 59069-59080.
Surat, India, February 7–8, 2020, Revised Selected Papers 27. Moupojou, E., Tagne, A., Retraint, F., Tadonkemwa, A.,
3 (pp. 267-276). Springer Singapore. Wilfried, D., Tapamo, H., & Nkenlifack, M. (2023).
16. Sujatha, R., Chatterjee, J. M., Jhanjhi, N. Z., & Brohi, S. N. FieldPlant: A dataset of field plant images for plant disease
(2021). Performance of deep learning vs machine learning in detection and classification with deep learning. IEEE
plant leaf disease detection. Microprocessors and Access, 11, 35398-35410.
Microsystems, 80, 103615. 28. Chug, A., Bhatia, A., Singh, A. P., & Singh, D. (2023). A
17. Saleem, M. H., Potgieter, J., & Arif, K. M. (2019). Plant novel framework for image-based plant disease detection
disease detection and classification by deep using hybrid deep learning approach. Soft Computing, 27(18),
learning. Plants, 8(11), 468. 13613-13638.
18. Shoaib, M., Shah, B., Ei-Sappagh, S., Ali, A., Ullah, A., 29. Ahmed, A. A., & Reddy, G. H. (2021). A mobile-based
Alenezi, F., ... & Ali, F. (2023). An advanced deep learning system for detecting plant leaf diseases using deep
models-based plant disease detection: A review of recent learning. AgriEngineering, 3(3), 478-493.
research. Frontiers in Plant Science, 14, 1158933. 30. Kotwal, J., Kashyap, R., & Pathan, S. (2023). Agricultural
19. Kumar, Y., Singh, R., Moudgil, M. R., & Kamini. (2023). A plant diseases identification: From traditional approach to
systematic review of different categories of plant disease deep learning. Materials Today: Proceedings, 80, 344-356.
detection using deep learning-based approaches. Archives of 31. Sanida, T., Tsiktsiris, D., Sideris, A., & Dasygenis, M. (2022).
Computational Methods in Engineering, 30(8), 4757-4779. A heterogeneous implementation for plant disease
20. Mustofa, S., Munna, M. M. H., Emon, Y. R., Rabbany, G., & identification using deep learning. Multimedia Tools and
Ahad, M. T. (2023). A comprehensive review on Plant Leaf Applications, 81(11), 15041-15059.
Disease detection using Deep learning. arXiv preprint 32. Plant disease detection using machine learning. International
arXiv:2308.14087. Research Journal of Engineering and Technology
(IRJET), 7(7), 2924-2929.
21. Mohameth, F., Bingcai, C., & Sada, K. A. (2020). Plant
disease detection with deep learning and feature extraction