Automated Identification and Categorization of Plant Leaf Diseases

Utilizing Image Analytics

Dr. Mani Devi, maanisingh38@gmail.com
Assistant Professor, CSE, SRM University, Haryana, India

Abstract
Plant diseases pose a serious risk to food security and agricultural productivity. Early
detection and accurate diagnosis of these disorders are essential for efficient disease
management. In this work, we describe a novel method for employing convolutional neural
networks (CNN) to identify plant leaf disease. The objective is to develop a dependable and
effective system that can identify and classify various plant leaf diseases using digital
photographs. The suggested CNN-based method is broken down into multiple phases. First, a
sizable dataset of photos of plant leaves showing different disease symptoms is collected and
appropriately annotated. The dataset is primarily divided into test and train data, with the test
data further subdivided into test and validation, respectively. Pre-processing methods
including picture augmentation, normalization, and cropping are employed to enhance the
dataset and the network's generalization capabilities. After building and training a CNN
architecture on the training set, discriminative features are extracted from leaf images.
Convolutional, pooling, and fully linked layers are the three types of layers that make up the
architecture, which is designed to effectively capture disease-related patterns and structures.
By feeding a big dataset, like ImageNet, into a network with pre-trained weights, transfer
learning is used to increase model performance. Throughout the training phase, a variety of
optimization techniques, including stochastic gradient descent and adaptive learning rate
algorithms, are employed to decrease errors such as classification. Experimentation is used to
fine-tune hyperparameters including learning rate, batch size, and regularization strategies to
attain peak performance. The suggested model is tested using data from an unknown test set
in order to determine the efficacy of the suggested methodology. Various metrics, including
accuracy, precision, recall, and F1-score, are used to assess the model's performance. The
CNN-based strategy is demonstrated to be superior in terms of computing efficiency and
accuracy of disease diagnosis when compared to existing methods. The experimental findings
show how accurately plant leaf diseases may be identified and categorized using the
recommended CNN-based method. The system is resilient against a wide range of disease
kinds, leaf deformations, and variations in lighting. Numerous practical uses for the
established system exist, including timely intervention strategies, precision crop management,
early disease warning systems, and agricultural sustainability.
Index Terms: Convolutional Neural Networks (CNN), Image augmentation, Transfer
learning, Optimization algorithms, Hyperparameter tuning
 I. Introduction
Agriculture is vital to India's economy, rural development, and employment. It ensures food
security for a large population and provides livelihoods, especially in rural areas, while
contributing significantly to export revenue. Agriculture also plays a key role in poverty
reduction and environmental sustainability. Thus, supporting and developing this sector is
crucial for India's growth and stability.
A plant leaf disease detection system is critical for India due to its heavy reliance on
agriculture for food security, livelihoods, and economic growth. Early disease detection helps
farmers take prompt action to prevent disease spread, protecting crops, farmers' incomes, and
food availability. This system promotes precision farming, resource optimization, and
environmentally friendly practices, enhancing productivity and food safety.
Detection techniques include visual inspection, spectroscopy, machine learning (ML),
computer vision (CV), sensor-based technologies, and smartphone applications. These
methods can be combined or customized for efficient and accurate detection.
1. Digital Image Processing Overview:
Digital image processing is the process of modifying and analyzing photographs using
computer algorithms,essential for automated tasks like object detection and pattern
recognition. Advances in algorithms and machine learning enhance computer vision
applications. The process typically includes:
1. Image Acquisition: Capturing images via digital devices.
2. Preprocessing: Enhancing image quality by reducing noise and correcting distortions.
3. Image Enhancement: Improving visual appearance through techniques like contrast
adjustment.
4. Image Restoration: Correcting deterioration to restore image quality.
5. Image Compression: Reducing image size for efficient storage and transmission.
6. Image Segmentation: Dividing images into meaningful areas for further analysis.
7. Feature Extraction: Identifying specific attributes from images for analysis.
8. Classification: Using ML algorithms to classify images into categories based on extracted

features.
 Figure 1: Steps for Digital Image Processing

Disease Identification and Decision-Making: Once trained, the classification model predicts
the presence and type of diseases in new leaf images, assigning a probability score for each
disease class. This helps prioritize actions or interventions. The detection system aids
farmers, agronomists, or researchers in making informed decisions on disease management,
including targeted pesticide use, crop rotation, or other control measures.
Digital Image Processing Applications: Digital image processing is widely used across
various fields such as medicine, astronomy, surveillance, robotics, entertainment, and
agriculture. It advances with sophisticated algorithms, machine learning, and access to
computational power and large-scale image datasets.

2. Plant Leaf Disease Overview: Plant leaf-diseases, due to pathogens, fungi, bacteria,
viruses, environmental factors, & nutrient deficiencies, significantly damage crops
and ornamental plants.
Fungal-diseases: for example, powdery mildew, downy mildew, leaf-spot, and anthracnose
often appear as discoloration, patches, lesions, or powdery growth on leaves and other plant
parts.

Fig.2 Fungal Plant Leaf Disease

Bacterial illnesses: Leaf diseases like bacterial canker, bacterial blight, and bacterial spot can
all be caused by bacteria. Leaf spots, wilting, yellowing, or blackening of the leaf tissue are
possible symptoms. Water and insects are common vectors for the propagation of bacterial
infections.

Fig.3 Bacterial Plant Leaf Disease

Plant illnesses caused by viruses: Depending on the particular virus and host plant, symptoms
of many plant diseases might vary greatly. Stunted growth, mosaic patterns on leaves,
yellowing, necrosis, or distortion of leaf tissue are all possible outcomes of viral infections.
Usually, infected plant material, contaminated instruments, or insect vectors are used to
transmit them.

Fig. 4 Viral Plant Leaf Disease

Early detection and accurate diagnosis are essential for plant leaf disease treatment. A range
of methods, including as visual examination, laboratory examinations, and contemporary
automated image-based detection systems, can be utilized to pinpoint the precise ailment and
direct suitable management actions.
To lessen the effects of leaf diseases and preserve plant health, integrated pest management
(IPM) techniques are frequently used. These tactics include cultural practices, biological
control agents, and the sparing application of pesticides.
 Transmission of
Infected Grains disease by
infected seeds

The spore
germination and leaf
infection Disease spore germination
and enters into seeding roots

Spore formation on
leaf spots. Blown to
leaves of other plants.

Fig.5 Disease Cycle

The series of occurrences that take place during the emergence and propagation of a plant
disease is known as the "plant disease cycle" (Figure 5). It deals with how the pathogen, host
plant, and surroundings interact. For illness management techniques to be effective, an
understanding of the disease cycle is necessary. However, the details of the disease cycle can
change based on the plant and the pathogen.

3. Image Processing:
Image-processing is said to be the study and utilization of mathematical operations and
techniques for the modification and interpretation of digital images. It offers a wide selection
of approaches or strategies that are perfect for improving, altering, evaluating, and also
obtaining data from pictures.
The first step in the procedure is obtaining digital photos using tools like scanners and
cameras. The quality of these photos is then increased by applying preprocessing techniques
like noise reduction and contrast enhancement. While feature extraction algorithms locate and
extract pertinent data, such as edges, textures, or shapes, filtering techniques are used to
improve particular image features.
To extract useful information from the images, image analysis techniques like object
detection, segmentation, and pattern recognition are used. After picture processing, the results
are analyzed to help with decision-making and action.
Post-processing procedures improve the analysis's findings and present the data in a more
understandable way. Image processing is used in many different areas, such as biometrics,
robotics, computer vision, remote sensing, medical imaging, and surveillance. Because it is
necessary for tasks like tracking, object detection, image restoration, and measuring, it
contributes to 10 advancements in several industries.
4. Image Processing in Agriculture:
Image processing, which has transformed traditional agricultural practices and improved
crop management through sustainability, is a major contributor to modern agriculture.
By using advanced image analysis technologies, agricultural practitioners can get
important insights into crop health, disease identification, yield estimation, and resource
optimization.
Mainly the primary applications of image processing in agriculture is the diagnosis of
plant disease. Algorithms can identify visual clues linked to pests, diseases, or dietary
deficiencies by analyzing photographs of plants collected in the field. By implementing
pertinent treatments or preventive measures in a timely way, early diagnosis helps
farmers avoid crop damage and productivity losses.
4.1 Applications
Plant Disease Detection: Image processing can help in the early identification and
diagnosis of plant diseases by analyzing photographs of leaves, stems, or entire plants.
Using algorithms to identify visual cues, patterns, or color changes suggestive of disease,
farmers can opt corrective action to stop the spread of infections & minimize crop
wastage.
Weed Management and Detection: By utilizing visual cues such as color, shape, or
texture, image processing algorithms may distinguish between weeds and crops. Farmers
can reduce the competition between weeds and crops by implementing targeted weed
management strategies, such as selective herbicide spraying or mechanical removal, by
identifying and mapping weed-infested regions.
Crop Growth and Development Monitoring: By evaluating photos taken at regular
intervals, image processing allows for the continuous monitoring of crop development
and growth. Algorithms can quantify variables such as plant height, leaf area, or canopy
cover, revealing information about the health, growth patterns, and responses of crops to
their surroundings. Farmers can use this information to optimize their fertilization,
irrigation, and resource allocation techniques.
Precision Agriculture: By facilitating field management that is site-specific, image
processing is essential to precision agriculture. Algorithms are capable of identifying
spatial heterogeneity in crop growth, insect pressure, and soil parameters by evaluating
high-resolution photographs. By adjusting input inputs (such as water, fertilizer, or
pesticides) based on particular field circumstances, farmers can maximize crop output and

resource efficiency.

Harvesting and Yield Mapping: To maximize harvesting processes and produce yield
maps, image processing techniques might be used. Algorithms that analyze photos taken
during harvesting can determine which crops are ripe or mature, allowing for automated or
efficient harvesting. In order to support precision agricultural methods and the creation of
geographic maps of crop yield variability, yield mapping approaches employ image
analysis.
Soil Analysis and Mapping: Spectral analysis and image processing work together to
support soil analysis and mapping. Algorithms are capable of evaluating soil parameters
like moisture content, nutrient levels, and organic matter content through the analysis of
photos obtained by remote sensing platforms.
This data facilitates the administration of fertilizer at the appropriate spot and directs soil
management techniques.

II. Literature Review on Deep-Learning in Agriculture for Plant

Disease-Detection
Deep learning has revolutionized agricultural practices, particularly in domain of plant
disease-detection. The application of deep learning techniques enables farmers and
researchers to make precise forecasts, optimize resources, and enhance productivity while
reducing environmental impact. This review synthesizes recent research exploring various
deep learning models and methodologies applied to agricultural data, particularly for
identifying plant diseases.

Convolutional-Neural-Networks (CNNs)
CNNs have evolved as a predominant technique in plant disease detection because of
their efficacy in handling image-based data. Various studies highlight the superiority of
CNNs in precisely classifying and diagnosing plant-diseases:

Shelar et al. utilized the VGG19 network to detect diseases in potato leaves, achieving
effective binary classification between healthy and unhealthy leaves by passing data
through 19 convolutional layers .

Anand krishnan et al. developed a deep CNN model making use of 18,160 tomato-leaf
pictures, achieving an impressive accuracy of 98.40% in identifying tomato leaf diseases .

Rehan et al. emphasized the transition from traditional machine learning to deep learning,
particularly CNNs, in improving disease identification processes. Their work suggests
that CNN techniques significantly enhance accuracy and speed in detecting plant
diseases. Uday Pratap et al. applied a multi-layer CNN to detect Anthracnose disease in
mango leaves, achieving 97% accuracy. The study involved data augmentation and
visualization techniques to improve model performance .

Specialized CNN Architectures: Several studies have proposed advanced CNN

architectures tailored to specific plant disease detection tasks:

GPD-CNN: A specialized architecture designed by researchers for cucumber leaf disease

detection, utilizing dilated convolutions and global pooling to capture contextual
information from images, achieving high precision and accuracy .

DenseNet-121: In a study by Andrew J. et al., DenseNet-121 outperformed other models

like Inception V4, ResNet-50, and VGG-16, achieving 99.81% classification accuracy on
the Plant Village dataset .

Vision Transformers: Yasamin Borhani et al. explored the use of vision transformers for
plant disease classification, leveraging self attention techniques to get global relationships
in pictures, resulting in accurate disease categorization .

GANs: Sook Yoon et al. incorporated Generative Adversarial Networks (GANs) to

augment data and improve disease detection accuracy. GANs generate synthetic data to
enhance model training, addressing data scarcity issues .

Hybrid and Ensemble Approaches Combining different techniques can lead to improved
performance in plant disease detection:

EM-BP-ANN: Thushara et al. proposed a hybrid model combining Expectation

Maximization (EM) for characteristics extraction & Backpropagation Artificial-Neural-
Network (BP-ANN) for classifying, achieving 96% accuracy in real-time disease
detection .

OMNCNN: Ashwin Kumar et al. utilized a MobileNet-based CNN with hyperparameter

optimization and extreme learning machine (ELM) for classification, showing enhanced
performance through the use of bilateral filtering and segmentation techniques .

Data and Methodological Considerations High-quality datasets and robust preprocessing

techniques are crucial for the success of deep learning models:

Dataset Quality: Studies emphasize the importance of large, labeled datasets for training
deep learning models. For example, Anandkrishnan et al. used the Plant Village dataset,
while Jiang et al. created an augmented Apple Leaf Disease dataset to ensure diverse and
representative training examples .

Preprocessing and Augmentation: Techniques such as data augmentation, thresholding,

and bilateral filtering are commonly employed to enhance model generalization and
accuracy .

III. PLANT-LEAF-DISEASE-DETECTION

In this research work, input is the images taken from plant leaves of various categories. Since
the data used as input is an image so image processing techniques have been used in the
proposed methodology. The plant leaf image samples are collected from an open database
named plant village on Kaggle. The images are colored and in jpg format. There isa total of
15 directories, each directory contains a different variety of plant leaf images with a different
number of data samples. The collection contains 20,123 labeled leaf images which contain
healthy as well as unhealthy plant leaf images. There are a total of 15 different classes 40 of
plant leaf including the healthy leaves. Majorly three different types of plants like pepper bell,
potato and tomato are included in the dataset with 12 different disease classes including a
healthy class from each three types of plant, so it becomes 15. The proposed methodology
uses this dataset which consists of 15 classes among which 3 directories contain healthy leaf
images from each type of plant namely pepper bell, tomato, and potato. The sample diseased
and healthy leaves from all three categories are shown below in Figures 3.1 and 3.2. Reason
for selection of plant village dataset is that, it is only one dataset in which images are captured
in lab, and images are separated from plant, rest all like Bing, AES all are taken in mixed
environment and have either white bag round or busy background with other green elements
in and not separated from plant as well.
 Fig. 6 Varieties of tomato leaf disease used

Fig. 7 Varieties of Potato and Pepper Bell leaf disease

Fig. 8 Frequency Histogram of different Class

 Fig. 9 Workflow for the classification process

IV. Splitting Training and Testing Dataset

The 256×256×3 input image data were gathered for the training set. Following preprocessing,
the final dataset is bifurcated into train set & test set data in an 80% to 20% ratio. 44
augmentation is then carried out. For testing and training purposes, neural networks often
follow this split ratio guideline. To fit into the model, each dataset image has 256 by 256
pixels. The network model is trained using the hyperparameters, which include the number of
epochs, data size, batch size, and model training rate. The model goes through the training
phase using a total of 25 epochs in order to improve accuracy. For training, the model is
applied with a batch size of 32. Additionally, the model is trained using CNN-VGG16
classifier.

Deep Learning Model

The proposed model consists of multiple sets of Convolutional (Convs.), Pooling, and
activation layer (ReLu) which is most commonly used in neural networks. The Maxpooling
layer is helping down sample the total of the feature taken from the dataset. Which helps to
reduce overfitting issues.

Fig. 10 Proposed CNN model for detection and classification process

In the above figure, it is shown how these three layers are divided into three stages followed
by max pooling in the initial two sets of convolutions, then the dense layer and classification
layer that comes afterward to third stage. The filter is used as 3*3 and the stride is used as 1.
For max pooling the filter size is 2*2 and stride are kept as 2, padding is kept the same.
The convolutional layer defines a set of filters that apply convolution on the entire image. In
this system, every convolutional layer learns several features that correspond to arbitrary
patterns required to identify the type of plant leaf. In Deep Neural Networks, each gradient
update on a batch of data makes use of distinct feature information from the layer before it.
Given that during training the parameters of the previous layers are modified, the input
feature map's data distribution also varies significantly. This increases training speed greatly
and uses several methods to figure out parameter initialization. The ReLu activation function
that is extensively employed in creating neural networks. It is the function of identification,
f(x) = max (0, x). Max pooling is reducing the height and width of the feature vector. It uses a
filter size of 2 cross 2 and an ST value of 2. The neural network uses sets of pooling layers
and convolution layer stacks for feature extraction. To put it simply, it's a series of digital
filters. The image size is subsequently decreased by the pooling layer. The pooling layer
merges neighboring pixels into a single pixel. By normalizing the input of each layer in the
network, not just the input layer, batch normalization drastically cuts down on the amount of
time needed for training. Higher learning rates are made possible by this strategy, which
lowers the number of training steps required for the network to converge. The SoftMax
function serves as the output layer activation function for CNN model that forecasts a
multinomial probability distribution. The logistic SoftMax activation function is deployed at
last to classify data into, multi-class. This is used for multiclass classification. Small filter
size reduces computational costs. 3*3 size has been proven as the best fit so far.

Fig. 11 CNN Architecture Brief

Fig. 12 Proposed Convolutional Neural Network Architecture visualization

The proposed VGG16 neural networks outperform other neural networks when voice, visual,
or audio signals are provided as inputs. Three distinct kinds of levels are present in them:
 • FC (fully-connected) layer; • Layer of pooling; • Layer of convolution
V. Training and validation

Fig. 13 Training and validation accuracy of the proposed model with data augmentation

It can be seen from the above image that the validation accuracy is low initially when the
model was passed through the training process, initial epochs validation accuracy is low
which is as expected very natural that initially, the model is trying to learn from the dataset
and learning 53 algorithm. The number of times model passes through the training phase, the
validation accuracy increases and reaches approximately the level of training accuracy, we
can see that in the last few epochs out of a total of 25 epochs that goes up to 98%. That means
now the model has learned from the data after going through the same training phase 25
times.

Fig 14. Training and validation loss of proposed model with data augmentation

From the above image we can see that after passing the model through several epochs, the
validation loss started decreasing which is a good sign for the model, which shows the model
has learned well and started predicting the output correctly with minimum error or loss. The
validation loss reduced from approximately 2% to 5% after 20 epochs. So that shows the
model is performing well and can predict the output with high accuracy of 98% to 99%.
In this part, the proposed research methodology for automatic disease detection & bifurcation
of leaf using picture analytics is explained as well as reflected. Dataset used for this research
work is in image format (.jpg) with different classes that contain images from different types
of plant leaf diseases with 256*256 uniform size. Dataset is taken from the open-source
available database for this task. The work discusses the classification of plant leaf disease into
15 different classes including the healthy category. Three different categories of plant leaves
are included in the dataset such as potato, pepper bell, and tomato leaves. Data augmentation
is applied to the dataset before training the model for classification purposes. Images are used
to train the framework which is a neural network model that runs through data 25 times to
learn the pattern from it to correctly classify the test data based on the learning from train set
data. The detection of disease and classification in different categories are achieved using the
proposed framework and the obtained result is discussed in the chapter. The proposed model
succeeded in achieving high accuracy with the plant village dataset and also the time taken by
the model for predicting various diseases from such a huge dataset is relatively low.
VI. Result:
VGGNet is used that is VGG16 with DA for which the model is created accordingly. For this
step, verities of libraries of Python programming are imported especially tensor flow, and
from tensor flow keras is imported to import the model from the applications of the Keras
library. For the array of images which is to be created, NumPy is imported. Additionally, the
model module is imported to create VGG16 model layers. Model Training: Initialize the
CNN’s VGG16with DA model with random weights and train it using the training set. The
model gains the ability to identify pertinent characteristics from the images provided and
categorize them into several disease categories during training. During training phase, the
model's weights are kept updated by backpropagation utilizing Adam optimization, forward
propagation, and the computation of the loss using an appropriate loss function (such as
categorical cross-entropy). Model Evaluation, Prediction, and Visualization: Using the test
data set, the trained CNN model's performance is assessed once it has been trained. The
model's efficacy in categorizing leaf-diseases is evaluated by measures, like precision, recall,
F1-score and accuracy. New, unseen leaf images are subjected to a disease class prediction
using the trained model. To gain 96 insights into its performance and identify any potential
misclassification, the model's predictions along with true labels are visualized.
VII. Future Scope:
It's significant to remember that dataset's quality and diversity, architecture's design, the
hyperparameters' tuning, and the model's iterative refinement are all necessary for
classification of plant leaf diseases using CNN to be successful. Regular updates and upkeep
are required to account for emerging diseases and maintain the model's functionality. In
future research work, the diversity in the dataset can be increased and the proportion of
affected leaf area can be detected apart from the disease. The primary images can be captured
from farming land which could have various noise factors, different light effects, etc., so in
future the work can be done to deal with this challenge of capturing input image from
farming land using which further can improve the efficiency of prediction and classification.
Also, a future work scope could consider detection of the presence of more than one disease
in one plant leaf and detecting it correctly and predicting disease from whole plant rather
apart from leaf area localization as well as predicting presence of disease in opposite side of
leaf.
 REFERENCES

[1] Albhaar, M. (2023). A Survey on Deep Learning and Its Impact on Agriculture:
Challenges and Opportunities. *Journal of Agriculture, 13*(3), 540.

[2] Anandhakrishnan, T., & Jaisakthi, S. M. (2022). Deep Convolutional Neural Networks
for image-based tomato leaf disease detection. *Sustainable Chemistry and Pharmacy,
30*, 100793.

[3] Battala lakshmi thushara et al. (2020). Analysis of Plant Diseases Using Expectation
Maximization Detection with BP-ANN Classification. *Volume XIII, Issue VIII,
August, 2020*.

[4] Fuentes, A., Yoon, S., & Kim, S. C. (2017). A robust deep-learning based detector for
real-time tomato plant diseases and pests recognition. *Sensors, 17*(9), 2022.

[5] Ganatra, N., & Patel, A. (2020). Performance Analysis of Fine-Tuned Convolutional
Neural Network Models for Plant Disease Classification. *International Journal of
Control and Automation, 13*(3), 293-305.

[6] Garg, K., Bhugra, S., & Lall, B. (2021). Automatic quantification of plant disease
from field image data using deep learning. In *Proceedings of the IEEE/CVF winter
conference on applications of computer vision* (pp. 1965-1972).

[7] Haseeb Nazki, S., Yoon, S., et al. (2020). Maximization Detection with BP-ANN
Classification. *Volume XIII, Issue VIII, August 2020*.

[8] Khandelwal, P. M., & Chavhan, H. (2019). Artificial Intelligence in Agriculture: An

Emerging Era of Research.

[9] Khan, R. U., Khan, K., Albattah, W., & Qamar, A. M. (2021). Image-based detection
of plant diseases: from classical machine learning to deep learning journey. *Wireless
Communications and Mobile Computing, 2021*, 1-13.

[10] Kumar, S., et al. (2019). Disease detection of various plant leaf using image
processing techniques: A Review. In *5th International Conference on Advanced
Computing & Communication Systems (ICACCS)*.

[11] Neetu Rani. (2017). Image Processing Techniques: A Review. *Journal on Today's
Ideas-Tomorrow's Technologies*.
[12] Nixon, M. S. (2020). Image Processing. In *Feature Extraction and Image Processing
for Computer Vision* (Fourth Edition), (pp. 83-139).

[13] Puspha Annabel, L. S., Annapoorani, T., & Deepalakshmi, P. (2019). Machine
Learning for Plant Leaf Disease Detection and Classification – A Review. In *2019
International Conference on Communication and Signal Processing (ICCSP)*.

[14] Ramesh, S., Niveditha, M., Pooja, R., Bhat, N. P., & Shashank, N. (2018). Plant
Disease Detection Using Machine Learning.

[15] Sasikala, R. (2022). Literature survey on digital image processing and its techniques.
*International Journal for Research in Science Engineering & Technology (IJRSET),
8*(9).

[16] Shelar, N., Shinde, S., Sawant, S., Dhumal, S., & Fakir, K. (2022). Plant Disease
Detection Using CNN. In *ITM Web of Conferences, 44*, 03049.

[17] Vijayalakshmi, B., & Mohan, V. (2016). Kernel based PSO and FRVM: An automatic
plant leaf type detection using texture shape and color features. *Computer and
Electronics in Agriculture, 125*, 99-112.

[18] Zhang, L., Zhang, L., et al. (2018). Application research of digital media image
processing technology based on wavelet transform. *EURASIP Journal on Image and
Video Processing*.

[19] K. P. et al. (2018). Deep learning models for plant disease detection and diagnosis.
*Computers and Electronics in Agriculture, 145*, 311-318.

[20] Mohanty, S. P., Hughes, D. P., & Salathé, M. (2016). Using deep learning for image-
based plant disease detection. *Frontiers in Plant Science, 7*, 1419.

[21] K. P. et al. (2020). Deep learning models for plant disease detection and diagnosis.
*Biosystems Engineering, 191*, 111-121.

[22] Zhang, X., Zhu, Q., Wang, Q., & Xu, Q. (2020). Deep learning for plant disease
diagnosis: A review. *Computers and Electronics in Agriculture, 173*, 105377.

[23] Sannakki, P. S., & Hiremath, P. S. (2020). Plant disease detection using deep learning
techniques: A review. *Computers, Materials & Continua, 63*(2), 537-554.

[24] Faithpraise et al. (2013). Automatic plant pest detection & recognition using k-means
clustering algorithm & correspondence filters. *International Journal of Advanced
Biotechnology and Research, 4*, 1052-1062.
[25] Pooja et al. (2017). Identification of plant leaf diseases using image processing
techniques. *IEEE Transactions on Instrumentation and Measurement, 66*(4), 130-
133.

[26] Mohanty et al. (2016). Using deep learning for plant leaf disease detection. *Frontiers
in Plant Science, 7*, 1419.

[27] Ökten, İ., & Yüzgeç, U. (2022). Rice Plant Disease Detection Using Image Processing
and Probabilistic Neural Network. In: Seyman, M.N. (eds) Electrical and Computer
Engineering. ICECENG 2022. *Lecture Notes of the Institute for Computer Sciences,
Social Informatics and Telecommunications Engineering, 436*, Springer.

[28] Vaidhehi, M., & Malathy, C. (2022). An unique model for weed and paddy detection
using regional convolutional neural networks. *Acta Agriculturae Scandinavica,
Section B — Soil & Plant Science, 72*(1), 463-475.

[29] https://nwdistrict.ifas.ufl.edu/hort/files/2012/06/cercospora_hydrangea.jpg
[30] https://www.canr.msu.edu/news/
signs_and_symptoms_of_plant_disease_is_it_fungal_viral_or_bacterial
 Pre-submission Comments and Declarations

Ethics Approval and Consent to Participate : I confirm that the plant materials used in the
study are not endangered.
Availability of Data and Materials: Plant leaf image samples are collected from an open
database named plant village on Kaggle.
Funding: No Funder is associated.
Conflict of Interest: No Conflict of Interest.