Proceedings of the Second International Conference on Applied Artificial Intelligence and Computing (ICAAIC 2023)

IEEE Xplore Part Number: CFP23BC3-ART; ISBN: 978-1-6654-5630-2

Plant Leaf Disease Detection using Machine

Learning

KethsyPrabavathy, Associate Professor M okara Bharath, UG Student KambamSanjayratnam, UG Student

Department of Computer Science and Department of Computer Science and Department of Computer Science and
Engineering, Engineering, Engineering,
2023 2nd International Conference on Applied Artificial Intelligence and Computing (ICAAIC) | 978-1-6654-5630-2/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICAAIC56838.2023.10140367

Karunya Institute of Technology and Karunya Institute of Technology and Karunya Institute of Technology and
Sciences, Sciences, Sciences,
Coimbatore, India Coimbatore, India Coimbatore, India
kethsy_cse@karunya.edu mbharath@karunya.edu.in kambamsanjay@karunya.edu.in

N. Sri Sai Rahul Reddy, UG Student M . Sasidhar Reddy, UG Student

Department of Computer Science and Department of Computer Science and
Engineering, Engineering,
Karunya Institute of Technology and Karunya Institute of Technology and
Sciences, Sciences,
Coimbatore, India Coimbatore, India
rahulnarapareddy@gmail.com msasidhar@karunya.edu.in

Abstract: Plant leaf disease detection is a critical task in accuracy is improved by fine-tuning it on a lower
modern agriculture to ensure better crop yield and quality. dimension. CNNs are one of the most powerful tools for
This provides a unique strategy for detecting plant leaf disease image Partition and categorization. The main focus of this
using machine learning techniques. The proposed methodology
proposed system is to accurately classify leaf diseases so
consists of three main stages, followed by classification using
five different models, including KNN, S VM, Decision Trees, that it can solve the leaf world problems that are faced in
Random Forest, and CNN. The collected images are pre - crop fields by farmers. The model is less complex and has
processed to eliminate unwanted features, and the images are easy identification of plant diseases. The system aims to
resized to a standardized size of 256 x 256 pixels. The following improve the accuracy of the existing methodology by using
stage involves utilizing the pre-trained CNN model to extract CNNs for image segmentation and classification, and by
pertinent features. The extracted features are then utilized to incorporating transfer learning methods. The proposed
train the classification models. The performance of each model model is composed of four distinct stages, including data
is assessed using various metrics, to predict its effectivity and collection, preprocessing, segmentation, and extraction,
accuracy. This proposed methodology is expected to provide a culminating in the final stage of classification. Training
reliable and efficient diagnosis of plant diseases, helping
farmers to take timely measures to prevent disease outbreaks makes up 80 percent of the gathered dataset, while testing
and ensure healthy crop growth. The proposed system
makes up 20 percent.
achieved high accuracy, less complexity, and easy
identification. The experimental findings show that the During the data preprocessing stage, the system receives an
suggested paradigm is successful in identifying common input image of a leaf, which is taken with the help of a
diseases. The suggested method of early detection and digital camera and represents the leaf's color in RGB format.
diagnosis of crop diseases can result in timely treatment and The preprocessing is achieved through image resizing and
higher crop yield. rescaling. The collected data is partitioned into subsets with
the help of the validation_split property.
Keywords—Machine Learning, Image Segmentation, Support
Vector Machine, Plant leaf disease detection.
In the Image Segmentation phase, the input image is divided
I. INT RODUCT ION into different segments to make image analysis easier. CNN
is used for segmentation, which reduces processing time and
The improvement in the field of machine learning in overcomes the problem of underfitting. Data aggregation
previous years has revolutionized the field of image data is produced as a result of the picture segmentation step.
recognition, and classification. A model is proposed for the
detection and classification of disease using ML techniques In the Feature Extraction phase, different features of plants
or methodologies. The model is designed to accurately such as size, shape, and surfaces are considered to extract
identify the disease affecting a plant, enabling farmers and
different properties of leaves. The GLCM algorithm is used
researchers to take corrective measures accordingly. for surface feature examination. After generating the co-
occurrence matrices, the necessary features are extracted to
The suggested framework relies on ML techniques. The
classify them in the subsequent stage, which is commonly
classification model is initially fed with a huge dataset for referred to as classification.
training using the machine learning approaches, and then its

978-1-6654-5630-2/23/$31.00 ©2023 IEEE 378

Authorized licensed use limited to: Malnad College of Engineering. Downloaded on November 04,2024 at 16:05:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Applied Artificial Intelligence and Computing (ICAAIC 2023)
IEEE Xplore Part Number: CFP23BC3-ART; ISBN: 978-1-6654-5630-2

In the Classification phase, Support vector machines and technique that is commonly used for pattern identification
other classification algorithms are used. The input images applications. The suggested method detected plant illnesses
for the neural network are reduced in dimensions to reduce with a remarkable precision rate of 96%, demonstrating the
processing time and overcome the problem of underfitting. promise of machine learning methods in the context of
The model undergoes training for a specific number of recognizing plant diseases. This study has important
epochs, after which the classification model's performance implications for enhancing agricultural production and
metrics are evaluated and compared. The results eventually leading to higher yields and better-quality crops.
demonstrate that the proposed model accurately classifies In their study, Ramesh and co-authors [7] introduced a
plant leaf diseases with high accuracy. The proposed system cutting-edge system using advanced machine learning
can be utilized to classify all kinds of diseases easily, which methodologies, namely (CNN) and (SVM), for the purpose
can lead to the early uncovering of plant diseases and the of detecting plant diseases. System yielded a remarkable
subsequent prevention of their spread. accuracy rate of 91%, indicating the high potential of
machine learning in plant disease detection. This outcome is
II. LITERATURE SURVEY significant as it suggests that the advanced transfer learning
Plant diseases are a significant threat to crop methodologies could significantly enhance the ability to
productivity, and the traditional methods of manual detect and mitigate plant diseases, ultimately leading to
observation and identification of diseases are time- improved crop yields and food security.
consuming and often not efficient. With the rapid
development of ML techniques, researchers are able to Kumar [8] proposed a system using machine learning
develop automated systems to detect and diagnose diseases. methodologies or techniques to detect a leaf disease. The
Some of the existing research is reviewed. proposed system achieved 96.5% accuracy, demonstrating
its effectiveness in detecting plant leaf disease.
One of the recent studies was done by Reddy et al. [1], who
created a method utilizing deep learning models to detect Varshney along with colleagues [9] suggested an approach
plant illnesses. They employed pre-existing models such as for identifying plant diseases that utilize machine learning
VGG16 and InceptionV3 to extract features and methods such as SVM and Random Forest. The suggested
subsequently trained classifiers such as SVM, KNN, and method obtained a precision of 92.7%, indicating the
Decision Tree. The outcome was an accuracy rate of potential for machine learning in detecting plant diseases.
97.35%, indicating the vast possibilities of ML in
identifying plant diseases. Kaur and Kang [10] proposed an advanced classifier, which
utilizes a SVM model to enhance the detection process.
Another study on plant disease detection was done by Their suggested approach achieved an impressive accuracy
Panchal [2], who proposed a method using CNNs to detect rate of 95.33%. This accomplishment confirms the efficacy
leaf diseases. The proposed model achieved a high accuracy of machine learning in accurately detecting plant diseases.
of 97.67%, demonstrating the effectiveness of deep learning This innovative approach can lead to improved management
in detecting leaf diseases. of plant diseases, which can ultimately lead to better crop
yields, increased food production, and a more sustainable
De Luna [3] presented an autonomous mechanism for environment.
acquiring photographs of tomato plant foliage and applied
deep-learning algorithms to categorize them as either In their research study, Khirade and Patil [11] introduced an
healthy or diseased. They obtained a precision of 98.4% innovative system for detecting plant diseases by leveraging
using the ResNet-50 architecture, indicating the possibility the power of advanced image processing techniques. Their
of automatic systems in detecting plant diseases. system demonstrated remarkable accuracy of 93.33%. The
researchers' findings shed new light on the efficacy of this
A study by Yusoff[4] proposed a disease identification approach and suggest its feasibility in practical settings.
system thatdetects Hevea leaves using the Sobel edge With further research and development, this technology
algorithm on FPGA. The proposed system achieves 92.3% could revolutionize the way people monitor and manage
accuracy and a processing speed of 40 frames per second, plant health, offering a more precise and efficient alternative
demonstrating the potential of FPGA-based systems for to traditional methods.
plant disease detection.
Sharif and associates [12] suggested a segmentation
Srikaew and collaborators [5] created a plant illness approach that is optimized along with an approach of
identification system using an artificial intelligence selecting a feature for identifying and categorizing pomelo
algorithm and a co-occurrence matrix. The system proposed diseases. The suggested method attained a precision rate of
by them attained a precision of 92.5%, showcasing the 97.27%, showcasing the great prospects of machine learning
capabilities of artificial intelligence in identifying plant in detecting plant diseases.
diseases.
Chandavale et al. [13] developed a comprehensive strategy
In their study, Thirumalesh [6] and colleagues presented a for preventing, identifying, and treating plant diseases. To
K-Nearest Neighbor (KNN) classifier that was used to identify and categorize plant diseases, the suggested method
identify disorders in paddy leaves, which was a unique combines machine learning algorithms with image
method. The KNN method is a supervised machine-learning processing techniques. The scientists reported a 95%

978-1-6654-5630-2/23/$31.00 ©2023 IEEE 379

Authorized licensed use limited to: Malnad College of Engineering. Downloaded on November 04,2024 at 16:05:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Applied Artificial Intelligence and Computing (ICAAIC 2023)
IEEE Xplore Part Number: CFP23BC3-ART; ISBN: 978-1-6654-5630-2

accuracy after testing their method on a variety of crops,

including tomato, wheat, and soybean.

Kaur and Kang [14] proposed an intensification in the SVM

classifier to revamp detection. The scientists employed
several pre-processing approaches to improve picture
quality and reached an efficiency of 97.8% in detecting and
classifying rice illnesses.

Khirade and Patil [15] developed a system that is effective

and efficient. The authors used a combination of color
segmentation, morphological operations, and feature
extraction methods to detect and classify plant diseases. The
proposed system was tested on various crops such as
tomato, potato, and chili, and achieved an accuracy of 96%.

To sum up, the application of machine learning for

identifying plant diseases has received substantial interest in
Fig.2. Schematic flow of the proposed System.
the last few years due to its precision and effectiveness.
Multiple experts have suggested diverse methods and
formulas for detecting and categorizing plant ailments.
A. Data Acquisition
Nonetheless, a few hurdles persist, including the absence of
a standardized dataset and the requirement for additional Data Acquisition is the starting step of the proposed
investigation in the realm of immediate recognition methodology where the required dataset of plant leaf images
is collected. The dataset that has been used for this system is
III. M ET HODOLOGY taken from online datasets which contain 40,305 input
images that are classified into different sections based on the
disease. The dataset has been partitioned into two where the
The methodology is explained in detail by segregating it major portion is utilized to train the data and the remaining
into four different stages . This proposed system gives an is utilized to test the data after the model is trained
accurate classification, Less complexity, and Easy completely.
identification. Similar to other machine learning models this
proposed system also follows the three-step architecture i.e.
Data Collection, Data Pre-processing, followed by Testing, B. Pre-Processing Phase
and Training. Fig(1) and Fig(2) describes the proposed
system architecture and the schematic flow. All the phases The acquired images need to be preprocessed to eliminate
are described in detail below. all unnecessary elements. information present in the images.
In this study, the preprocessing step involves image resizing
and normalization. The images are resized to a uniform size
of 256 x 256 pixels. The pixel values are then normalized to
lie between 0 and 1.

C. Feature Extraction Phase

The next step in the proposed methodology is
feature extraction where the relevant features are going to be
extracted from the preprocessing phase. The features are
extracted using the VGG16 model, which is a pre-trained
The CNN architecture has already undergone training on the
pre-existing ImageNet processing file. To extract features,
the VGG16 model, which was pre-trained, is utilized, and
the Result generated by the antecedent stratum is utilized as
the extracted feature vector by eliminating the final
Fig.1. Architecture connected layer. As a result, the extracted features are then
presented to the classification model for the subsequent
phases of training and testing.

D. Classification (Model Building)

The final step in the proposed methodology is to classify the
extracted features from the above phase into different
classes of plant diseases. This proposed system builds five
different classification models, namely KNN, SVM,

978-1-6654-5630-2/23/$31.00 ©2023 IEEE 380

Authorized licensed use limited to: Malnad College of Engineering. Downloaded on November 04,2024 at 16:05:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Applied Artificial Intelligence and Computing (ICAAIC 2023)
IEEE Xplore Part Number: CFP23BC3-ART; ISBN: 978-1-6654-5630-2

Decision Trees, Random Forest, and CNN. All the on computation time. The average computation time
classification models are trained on the extracted features required for processing a single image was found to be
which are extracted in the above phase and results are noted. 0.003 seconds on a standard desktop computer. This
The evaluation of the individual model is computed utilizing demonstrates the real-time capability of the proposed
diverse criteria such as exactness, meticulousness, retrieval, system, making it suitable for practical applications in the
and F1-measure. The decision tree algorithm is a tree-based field. In summary, the proposed system utilizes machine-
model that partitions the feature space into regions based on learning techniques to classify different plant leaf diseases
the values of the features and then assigns a class label to based on input images. The proposed system attains 91.93%
each region. The SVM algorithm is a binary linear classifier accuracy, which is higher than that achieved by previous
that maximizes the closeness among vectors of various studies. The feature importance analysis reveals that the
kinds in the feature space. The Random Forest model makes Mean Green Intensity is the most significant feature
use of numerous decision trees to improve the classification contributing to the classification performance. Overall, the
performance. proposed system efficiently predicts the disease with the
help of leaf images accurately thereby contributing to the
IV. RESULT development of sustainable agriculture. Some of the sample
The proposed methodology was implemented using Python results of the proposed system regarding the classification of
programming language and the TensorFlow machine different diseases are shown below.
learning framework. The performance metric results are
given below in Table 1 for each classification model.

Table 1: Experimental Results

Fig.3. Corn_(maize) Gray_leaf_spot classification

From Table 1, it can be observed that from all four
classification models, the Random Forest classification
model achieves an accuracy of 91.93% considered to be the
maximum, followed by the SVM model with 91.51%
accuracy. These results indicate the effectiveness of transfer
learning algorithms in detecting and classifying plant
diseases based on leaf images. Next, performed a feature
importance analysis to identify the most significant features
contributing to the classification performance. To enhance
the evaluation, the proposed system performance metrics are
compared with other approaches that had gone through
during the review. Table 2 shows the accuracy achieved by
the suggested system with the reviewed approaches .

Table 2: Comparison of Accuracy with Previous Studies

Fig.4. Corn maize Healthy Leaf

As shown in Table 2, the proposed system Fig.5. Potato_EarlyBlight classification

achieved the highest accuracy compared to the previous
studies, demonstrating its effectiveness in detecting and
classifying plant diseases. The efficiency is evaluated based

978-1-6654-5630-2/23/$31.00 ©2023 IEEE 381

Authorized licensed use limited to: Malnad College of Engineering. Downloaded on November 04,2024 at 16:05:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Applied Artificial Intelligence and Computing (ICAAIC 2023)
IEEE Xplore Part Number: CFP23BC3-ART; ISBN: 978-1-6654-5630-2

V. CONCLUSION [3]R. G. de Luna, E. P. Dadios and A. Bandala, "Automated Image

Capturing System for Deep Learning-based T omato Plant Leaf Disease
The method used in this research for identifying leaf Detection and Recognition" IEEE Region 10 Conference.
diseases involves leveraging machine learning, along with
processing such as image resizing, rescaling, and extraction [4]"Real-time Hevea Leaves Diseases Identification Using Sobel Edge
Algorithm on FPGA" IEEE Control and System Graduate Research
of features, to extract relevant information from photos of Colloquium, Yusoff, N. M., Abdul Halim, I. S., Abdullah, N. E., & Ab
plant leaves. To categorize the plant leaves photos into Rahim, A. A. (ICSGRC).
either positive or negative groups, the system makes use of
[5]A.Srikaew, K. Attakitmongcol, C. Khitthuk, & P. Kumsawat (2018)
different classification models such as KNN, DT, SVM, and
(IEEECON).
Random Forest(RF) models. From all the classification
models that are trained, the RF model achieved a maximum [6]T hirumalesh, B. V., Suresha, M., &Shreekanth, K. N. (2017) (I2CT )
accuracy of 91.93% in categorizing the leaf images
according to the results. [7]S. Ramesh et a., "Plant Disease Detection Using Machine Learning,"
2018 International Conference on Design Innovations for 3Cs Compute
The proposed approach is a non-invasive and efficient Communicate Control (ICDI3C).
method for detecting plant leaf diseases , which can be used
by farmers to diagnose the diseases immediately. By [8]S. Kumar, K. Prasad, A. Srilekha, T . Suman, B. P. Rao, and J. N.
identifying the disease of plants early, farmers implement Vamshi Krishna, "Leaf Disease Detection and Classification based on
Machine Learning," (ICST CEE).
essential actions aimed at halting the continued propagation
of the ailment, resulting in increased crop yield and reduced [9]D. Varshney, B. Babukhanwala, J. Khan, D. Saxena, and "Plant Disease
economic losses. Detection Using Machine Learning T echniques," 2022 3rd (INCET ).

[10]R. Kaur and S. Kang, "An enhancement in classifier support vector

machine to improve plant disease detection," 2015 IEEE 3rd (MIT E).
VI. FUT URE W ORK
Going forward, there are plans to enhance the suggested [11]S. D. Khirade and A. B. Patil, "Plant Disease Detection Using Image
Processing," 2015 International Conference on Computing Communication
methodology by integrating advanced deep learning Control and Automation.
methodologies., which can provide better accuracy and
efficiency in plant leaf disease detection. Also has the plan [12]Sharif, Muhammad, Khan, Muhammad, Iqbal, Zahid, Lali,
to expand the dataset to include more plant species and Muhammad Ikram & Javed, Muhammad. Detection and classification of
citrus diseases in agriculture based on optimized weighted segmentation
diseases, which can further enhance the accuracy and and feature selection. (2018)
effectiveness of the proposed approach.
Furthermore, the intention is to develop a web app relying [13]Anjali Chandavale, Suraj Patil, Ashok Sapkal (2017). Agri-Guide: An
on the suggested strategy that farmers may employ to Integrated Approach for Plant Disease Precaution, Detection, and
T reatment.
discover and evaluate plant leaf diseases in real-time. The
mobile application will be equipped with a user-friendly [14]R. Kaur and S. Kang, "An enhancement in classifier support vector
interface, which can provide farmers with the necessary machine to improve plant disease detection," 2015 IEEE 3rd (MIT E).
information about plant diseases and their treatment options.
[15]S. D. Khirade and A. B. Patil, "Plant Disease Detection Using Image
In addition, planning to collaborate with agricultural experts Processing," 2015 International Conference on Comput ing Communication
and researchers to further improve the proposed approach Control and Automation.
and explore new avenues for plant disease detection and
prevention. The proposed approach has the realm of
possibility to transfigure the field of Agronomy and
contribute to the sustainable growth of the agricultural
sector.
In conclusion, the proposed approach is a promising
solution for plant leaf disease detection, which can provide
farmers with an efficient and non-invasive method to
Identify and ascertain maladies in vegetation. Subsequent
endeavors will concentrate on enhancing the precision and
productivity of the proposed approach and developing a
mobile application for real-time plant disease detection and
diagnosis. The proposed approach can contribute
significantly to the sustainable growth of the agricultural
sector and help farmers to increase their crop yield and
reduce economic losses.

VII. REFERENCES
[1]P.C Reddy, R. M. S. Chandra, P. Vadiraj, M.A Reddy, T . R. Mahesh,
and G.S Madhuri, "Detection of Plant Leaf-based Diseases Using Machine
Learning Approach," 2021 IEEE (CSIT SS).

[2]P. Panchal, V. C. Raman, and S. Mantri, "Plant Diseases Detection and

Classification using Machine Learning Models,"(CSIT SS).

978-1-6654-5630-2/23/$31.00 ©2023 IEEE 382

Authorized licensed use limited to: Malnad College of Engineering. Downloaded on November 04,2024 at 16:05:05 UTC from IEEE Xplore. Restrictions apply.