Report on Project-II (BIT-440)

PLANT DISEASE DETECTION

USING MACHINE LEARNING
Submitted in the fulfilment of the requirement for the degree of

Bachelor of Technology
In

“Information Technology”

PUSKAR PANDEY (2021071029)

NISHIT UJJAWAL (2021071024)
SHIVANI VERMA (2021071034)

Under the supervision of

Prof. Shiva Prakash Sir

Department of Information Technology & Computer Applications

MADAN MOHAN MALVIYA UNIVERSITY OF TECHNOLOGY
GORAKHPUR (273010)

1
 DECLARATION

We hereby declare that the Project titled "Plant Disease Detection using Machine Learning,"
submitted in partial fulfillment of the requirements for the Degree of BTech in Information
Technology, is the result of my own original work. All sources used in this project have been
duly acknowledged. We have adequately cited and referenced the original sources. We also
declare that we have adhered to all principles of academic honesty and integrity and have
not misrepresented or fabricated or falsified any idea/data/fact/source in my submission. We
understand that any violation of the above will because for disciplinary action by the
University and can also evoke penal action from the sources which have thus not been properly
cited or from whom proper permission has not been taken when needed.

Sign: Date:
Student Name: Puskar Pandey

Roll No.: 2021071029

Sign:
Student Name: Nishit Ujjawal

Roll No.: 2021071024

Sign:
Student Name: Shivani Verma

Roll No.: 2021071034

2
 CERTIFICATE

Certified that Puskar Pandey (2021071029), Nishit Ujjawal (2021071024) and Shivani Verma
(2021071034) have carried out the project work presented in this project report entitled “Plant
Disease Detection Using ML” for the award of Bachelor of Technology in Information
Technology from Department of Information Technology & Computer Application, Madan
Mohan Malaviya University of Technology, Gorakhpur under my guidance. The project report
embodies results of original work, and studies are carried out by the students themselves and
the contents of the project report do not form the basis for the award of any other degree to the
candidates or to anybody else from this or any other University/Institution.

Signature of Mentor Signature of HOD

Prof. Shiva Prakash Prof. Shiva Prakash
ITCA Department ITCA Department
MMMUT, Gorakhpur MMMUT, Gorakhpur

Date:
Place:

3
 ACKNOWLEDGEMENT

We are pleased to acknowledge our sincere thanks to the Department of Information

Technology& Computer Applications, MMMUT for their kind encouragement in this project and
for completing it successfully. We are grateful to them.
We convey our Thanks to Mr. Kamlesh Gautam, Dr. R.K. Dwivedi, Head of ITCA Department
Dr. Shiva Prakash for providing us necessary support and details at the right time during the
progressive work.
We would withal relish conveying our gratitude to all the faculty members and staff of the
Department of ITCA for their wholehearted cooperation to make this work turns into authenticity.

Sign: Date:
Student Name: Puskar Pandey Place:
Roll No.: 2021071029

Sign:
Student Name: Nishit Ujjawal
Roll No.: 2021071023

Sign:
Student Name: Shivani Verma

Roll No.: 2021071034

4
 ABSTRACT

Plant diseases significantly affect global agricultural productivity, posing a serious threat to food
security and the economy. Traditional methods of disease identification are time- consuming,
labor-intensive, and often require expert knowledge, making early detection and treatment
challenging. This project aims to leverage machine learning techniques to develop an efficient,
accurate, and automated system for plant disease detection using image data. The methodology
involves collecting and preprocessing a dataset of plant leaf images, including healthy and
diseased samples. A Convolutional Neural Network (CNN)-based approach is employed to
classify diseases based on visual patterns. The model is trained and validated using a well-curated
dataset, achieving high performance in metrics such as accuracy, precision, and recall.
Additionally, the system is designed to provide real-time predictions, making it suitable for
practical deployment in agriculture.
The results demonstrate that the proposed machine learning model outperforms traditional
methods in terms of speed and accuracy. This system offers a scalable solution for early disease
detection, helping farmers reduce crop losses and improve yield quality. Future work involves
integrating this model with mobile applications and IoT devices for real-world usage and
expanding its applicability to a broader range of crops and diseases.
This abstract provides a concise overview of the project, emphasizing its objectives, methodology,
results, and potential impact. Plants are very essential in our life as they provide source of energy
and overcome the issue of global warming. Plants now a daysare affected by diseases like bacterial
spot, late blight, Septoria leaf spot. These diseaseseffect the efficiency of crop yield. So, the early
detection of diseases is important in agriculture. Detection of diseases as soon as they appear is
vital step for effective disease management.
The use of machine learning (ML) in plant disease detection and management has emerged as a
transformative approach to enhancing agricultural productivity and sustainability. This report
explores the application of advanced ML techniques for identifying and diagnosing plant diseases,
leveraging image recognition, data analysis, and predictive modeling.
\

5
 TABLE OF CONTENTS
TOPIC PAGE NO.
Chapter 1 – Introduction
1.1 Background and Motivation 8
1.2 Objectives 9
1.3 Scope of this Project 10
Chapter 2 – Literature Review
2.1 Introduction to Plant Disease 11
2.2 Existing Methods for Plant Disease Detection 12
2.3 Analysis of Current Challenges and Gaps in 13
Conventional Technique
2.3 Cutting-Edge Machine Learning and Imaging Techniques in 14
Agriculture
Chapter 3 – Problem Statement
3.1 Problem Statement 17
Chapter 4 – Proposed Methodology
4.1 Description of the Datasets 18
4.2 Model Architecture 18
4.3 Flow Diagram 19
4.4 Data Preprocessing 20
4.5 Model Training and Optimization 20
4.5 ML Algorithms and Techniques 20
4.6 Model Evaluation Metrics 21
Chapter 5 Implementation
5.1 Software Tools, Libraries, Frameworks 22
5.2 Dataset Setup, Data Augmentation, and Preprocessing 23
5.3 Displaying Sample Images from the Training Set 23
5.4 Building the Convolutional Neural Network (CNN) Model 24
5.5 Compile and Train the CNN Model 24
5.6 Plotting Training and Validation Accuracy and Loss 25
Chapter 6 – Result and Analysis
6.1 Performance Metrices 26
6.2 Results and Discussion 27
6.3 User Interface 29
Chapter 7 – Conclusion and Scope for future research 31
References 32

6
 LIST OF FIGURES

FIGURE TOPIC PAGE NO.

NO.
1 Training Image 8
2 Table of Literature Review 16
3 Flow Diagram 19
4 Implementation code 23
5 Accuracy graph 27
6 Epoch Images 28
7 User Interface 29

7
 CHAPTER 1

INTRODUCTION

1.1 Background and Motivation

Plant diseases are a significant threat to global agriculture, causing considerable damage to crops
and leading to substantial economic losses. These diseases can be caused by various pathogens,
including fungi, bacteria, viruses, and environmental stress factors, affecting a wide range of crops
such as fruits, vegetables, and grains. For instance, fungalinfections like blight in potatoes or mildew
in wheat can devastate entire fields if not detected and managed promptly. The impact of plant
diseases is not only felt economically but also in terms of food security, as crop failures due to
uncontrolled disease outbreaks can lead to food shortages and increased prices. Traditional methods
of plant disease detection often involve manual inspections by experts or laboratory tests, both of
which are time-consuming, expensive, and prone to human error. As the global demand for foodrises
and agricultural practices become more intensive, there is a growing need for more efficient and
scalable solutions for early disease detection.

8
 This has led to increased interest in leveraging machine learning and computer vision technologies.
By automating the disease detection process, machine learning models can quickly analyze large
volumes ofdata, such as images of plant leaves, to accurately identify diseases at an early stage. The
motivation behind this research is to harness the power of machine learning to provide farmers with
faster, more accessible tools to detect and manage plant diseases, ultimatelyreducing crop losses and
improving food security worldwide. Fungal diseases, such as blight, mildew, and rust, are among
the most common and destructive. For example, late blight in potatoes and tomatoes, caused by the
fungus Phytophthora infestans, led to the Irish Potato Famine in the mid-19th century, highlighting
the devastating potential of plant diseases. Similarly, wheat rust diseases (e.g., Puccinia graminis)
continue to threaten wheat production globally, leading to losses in grain quality and yield. Bacterial
diseases, like bacterial wilt and leaf spot, can spread rapidly, affecting entire fields and making
disease management particularly challenging. Viruses, such as the Tobacco Mosaic Virus (TMV) or
Tomato Yellow Leaf Curl Virus (TYLCV), can severely hinder plant growth, often leading to stunted
plants and reduced yields. Nematode infestations can damage plant roots, reducing water and nutrient
uptake, further weakening crops.

1.2 Objectives

The purpose of this study is to develop a machine learning-based system for the early detection of
plant diseases using image analysis techniques. With the growing challenges in agriculture,
particularly in the context of climate change and increased pest pressures, timelyidentification of plant
diseases is crucial to minimizing crop loss and ensuring food security.Traditional diagnostic methods,
which often rely on expert knowledge and manual inspection, are not always feasible or efficient for
large-scale farming. This project aims toaddress this gap by leveraging machine learning, particularly
deep learning techniques suchas Convolutional Neural Networks (CNNs), to automatically detect
and classify plant diseases from images of leaves. The system is designed to provide accurate, real-
time predictions that can assist farmers in identifying diseases early, allowing for quicker
interventions and better disease management strategies. Through this work, we aim to createa scalable,
accessible tool that can help mitigate the economic impact of plant diseases and improve overall
agricultural productivity The objective of this study is to develop an effective and automated system
for the early detection of plant diseases using machine learning techniques, specifically focusing on
9
 image-based classification. The aim is to createa model that can accurately identify diseases in plant
leaves by analyzing visual patterns inimages, enabling timely intervention to prevent widespread
crop damage. By leveraging advanced machine learning algorithms, particularly Convolutional
Neural Networks (CNNs), the system is designed to automatically classify plant diseases,
distinguishing between healthy and diseased leaves. The ultimate goal is to provide farmers and
agricultural professionals with a tool that offers fast, reliable, and scalable disease detection,which can
be integrated into mobile applications or other monitoring systems. This will notonly reduce the
reliance on manual inspections but also improve crop yield management and help mitigate the
economic impact of plant disease.

1.3 Scope of this Project

The scope of the plant disease detection model using machine learning focuses on the identification
and classification of diseases in a selected range of crops, specifically [specifycrops, e.g., tomato,
potato, corn, etc.]. The model is designed to recognize a variety of common plant diseases, including
fungal, bacterial, and viral infections, that affect these crops. Examples of diseases addressed include
[specific diseases, e.g., early blight, late blight, powdery mildew, bacterial spot, etc.]. The model
will be trained on image datasets containing both healthy and diseased plant leaves, ensuring that it
can accurately differentiate between healthy foliage and those exhibiting symptoms of disease. The
primary focus is on developing a robust system that can handle real-time detection of multiple
diseases within a single crop or across multiple crop types, providing a comprehensive solution to
plant disease management. While this project focuses on a specific set of plants and diseases, the
methodology and model can be extended to other crops and additional disease types with further
refinement and training.

10
 CHAPTER 2

LITERATURE REVIEW

Plant disease detection is a crucial aspect of precision agriculture, focusing on enhancingcrop health
and yield by enabling the early and precise identification of plant diseases. This literature review
examines the different techniques and technologies used in this domain, emphasizing significant
progress, existing challenges, and potential future developments.

2.1 Introduction to Plant Disease

An examination of plant diseases highlights the variety of pathogens, environmental influences, and
plant responses that affect agricultural output globally. These diseasescan be caused by numerous
agents, such as fungi, bacteria, viruses, nematodes, and abiotic stressors like nutrient imbalances,
extreme temperature fluctuations, humiditychanges, and soil conditions. Symptoms of plant diseases
often include leaf spots, wilting, yellowing, lesions, stunted growth, and reduced crop yields. The
financial consequences of plant diseases are significant, resulting in considerable yield losses, lower
food quality, higher production costs due to disease control measures, and eventrade barriers. To
manage plant diseases, a range of strategies are employed, includingcultural practices, plant rotation,
breeding for genetic resistance, chemical treatmentswith fungicides or pesticides, biological control
methods, and precision farming technologies. Early identification and accurate diagnosis of plant
diseases are vital forsuccessful management, often assisted by advanced tools like remote sensing,
image analysis, machine learning algorithms, and sensor networks. Emphasis is increasinglyplaced
on sustainable, integrated disease management techniques to reduce the environmental impact of
chemicals and support resilient farming systems.

In conclusion, effectively understanding and managing plant diseases is key toensuring global food
security and sustainable agricultural practices, especially in the face of challenges such as climate
change, global trade dynamics, and the growing demand for agricultural production

11
.
2.2 Existing Methods for Plant Disease Detection

Various methods have been developed for detecting plant diseases, each offering uniquebenefits and
facing certain challenges. Below are some of the widely used techniques.

1. Manual Inspection: Farmers and agricultural specialists often rely on visualinspections to

identify plant diseases by observing symptoms such as leaf discoloration, spots, lesions,
wilting, or irregular growth patterns. While this method is simple and low-cost, its accuracy
can be compromised, especially when detecting early or subtle symptoms, and it is highly
dependent on the skill and experience of the observer

2. Laboratory Diagnostics: Diseased plant samples are collected and analyzed in laboratories
using techniques like microscopy, culture methods, serological tests, and molecular
diagnostic tools such as Polymerase Chain Reaction (PCR) and Enzyme-Linked
Immunosorbent Assay (ELISA). Though laboratory testing provides precise identification of
pathogens, it is a time-consuming, expensive process that requires specialized equipment and
expertise.

3. Remote Observation: Technologies like satellite imagery, drones, and unmanned aerial
vehicles (UAVs) are employed to capture high-resolution images of crop fields. These
images are analyzed to detect changes in crop health and identify disease outbreaks through
spectral data. Remote sensing allows for extensive monitoring and early disease detection,
though it may face limitations such as cloud interference, resolution constraints, and high
cost.

4. Advanced Imaging and Pattern Recognition: High-quality digital images of crops are
taken using cameras or mobile devices, then analyzed with computervision techniques to spot
disease symptoms. By examining features like color,texture, shape, and size, these methods
help in classifying plants as healthy or diseased. While this approach offers fast and non-
invasive disease detection in the field, it requires substantial computational power and
expertise to develop and optimize the algorithms.

12
.

5. Remote Sensing with IoT: A network of sensors deployed across crop fields monitors
environmental factors such as soil moisture, temperature, humidity, and pH levels. Changes
in these parameters can serve as early indicators of disease outbreaks or plant stress. The use
of Internet of Things (IoT) technologies enables continuous monitoring and supports data-
driven decisions, although it may require substantial infrastructure and ongoing maintenance.

6. Machine Learning and Deep Learning: Machine learning methods, bothsupervised and
unsupervised, are trained on labeled datasets of crop images to identify diseases and assess
their severity. Particularly, deep learning models like Convolutional Neural Networks
(CNNs) have demonstrated significant success in image classification by automatically
extracting relevant features from raw data. These methods provide scalable, automated
disease detection, but they depend on large, diverse datasets for effective training and
validation.

7. Smartphone Apps and Decision Support Systems: Mobile applications and decision
support systems provide farmers with tools for real-time disease diagnosis, management
recommendations, and pest alert notifications. These apps may utilize image recognition,
sensor data, and expert knowledge to assist farmers in making informed decisions about
disease control measures.

2.3 Analysis of Current Challenges and Gaps in Conventional Technique

 Time and Cost Constraints: Laboratory-based methods and advanced imaging technologies
often require significant time, expertise, and financial resources.

 Accuracy Limitations: Visual inspections can be unreliable, particularly when distinguishing

between diseases with similar symptoms or when symptoms are subtle.

 Scalability Issues: Traditional methods are difficult to scale for large farms or diverse crop
types, limiting their practicality in industrial agriculture.

 Data Scarcity: Many conventional methods lack comprehensive datasets for regional, crop-

13
 specific, or disease-specific diagnosis.

 Inability to Predict: Most traditional methods focus on detecting diseases after symptoms
appear, with limited capability for early prediction or prevention

2.4 Cutting -Edge Machine Learning and Imaging Technique in Agriculture

Recent breakthroughs in machine learning and computer vision have transformed the agricultural
landscape, paving the way for innovative solutions to a wide rangeof agricultural challenges. These
advancements have led to significant improvements in crop monitoring, disease detection, yield
forecasting, and precision farming. Below are some of the notable developments in these fields.

1. Image-based Plant Monitoring

• High-Resolution Imaging: Advancements in imaging technologies, such as drones,satellites,
and high-resolution cameras, now allow for the capture of detailed imagesof crop fields. These
images provide essential insights into plant health, growth stages, and environmental stress
factors.

• Computer Vision Algorithms: Advanced computer vision techniques analyzethese images

to extract key features like leaf color, texture, and shape. These algorithms can identify
abnormalities, detect diseases, and evaluate plant growth parameters factors.

2. Plant Disease Detection

• Deep Learning Models: Convolutional Neural Networks (CNNs) have shownexceptional
performance in identifying plant diseases from images. These models can learn intricate
patterns and features from large datasets, enabling accurate, automated disease detection.

• Transfer Learning: Transfer learning utilizes pre-trained CNN models from large image
datasets (like ImageNet) and fine-tunes them for specific agricultural needs with smaller,
more targeted datasets. This approach reducesthe need for vast amounts of labelled data and
enhances detection accuracy.

14
3. Yield Prediction and Plant Management
• Data-driven Models: Machine learning algorithms, including regression models and
ensemble methods, analyze diverse data sources like weather conditions, soil health, and
historical yield records to predict future crop yields. These models provide farmers with
actionable insights for plant management, resource allocation, and planning harvest
schedules.

• Precision Agriculture: The integration of computer vision and machine learning

technologies in precision agriculture allows farmers to apply water, fertilizers, and pesticides
more efficiently by considering spatial and temporaldifferences within fields. This improves
crop yield, reduces environmental impacts, and lowers production costs.

4. Weed and Pest Detection

• Automated Weed Identification: Machine learning-powered computer visionsystems can
distinguish between crops and weeds in real-time, enabling targeted herbicide application
that minimizes chemical use and protects crop health.

• Pest Monitoring: Machine learning models process sensor data—such as temperature,

humidity, and insect counts—to forecast pest outbreaks and evaluate infestation levels. Early
detection allows for timely pest management interventions.

5. Robotic Farming

• Autonomous Vehicles: Machine learning algorithms drive autonomous vehicles and robots
capable of navigating crop fields to carry out tasks like planting, weeding, harvesting, and
data collection. These robots enhance operational efficiency, reduce labor costs, and enable
continuous monitoring of agricultural processes.

15
16
 CHAPTER 3

3.1 PROBLEM STATEMENT

Plant diseases pose a significant threat to global agriculture, leading to reduced cropyields, poor food
quality, and economic losses. Early detection and effective management of these diseases are crucial
to ensuring food security, minimizing losses, and maintaining agricultural sustainability. Detecting
plant diseases at an early stage can help prevent the spread of infections, reduce the need for chemical
treatments, and promote healthier, more resilient crops. Traditional methods of plantdisease detection,
such as visual inspection by farmers, laboratory testing, and the use of chemical indicators, have
been widely used for decades. However, these approaches face several challenges. Visual inspection,
while cost-effective, is ofteninaccurate, especially for subtle or early-stage diseases, and heavily
depends on theexpertise of the observer. Laboratory testing, though precise, can be time- consuming,
expensive, and requires specialized equipment and training. Additionally, methods like remote
sensing and sensor networks, while effective, often suffer from limitations such as high costs, poor
resolution, and the need for constant monitoring. Given these challenges, there is a growing need for
more efficient, automated, and scalable solutions for plant disease detection that can overcome the
limitations of traditional methods. This highlights the importance of integrating advanced
technologies, such as machine learning and artificial intelligence, into plant disease detection
systems to improve accuracy, speed, and accessibility, ultimately supporting more sustainable
agricultural practices.

3.2 Objective of the Report:

The primary goal of this report is to explore and address the challenges in detecting plant diseases
by leveraging machine learning (ML) technologies. Specifically, the objectives are :

• Development of an ML-Based Detection System

• Evaluation of Effectiveness

• Advancement of Agricultural Technology

• Feasibility Analysis for Practical Use

17
 CHAPTER 4

PROPOSED METHODOLOGY

This chapter outlines the methodology for plant disease detection, detailing the datasets used, the
architecture of the model, preprocessing steps, modeling techniques, machine learning (ML)
algorithms, and evaluation metrics. Each section provides a comprehensive understanding of the
approach employed to achieve accurate detection of plant diseases.

4.1 Description of the Dataset:

 Source: The dataset is sourced from custom-collected images from field environments.

 Content: It comprises high-resolution images of healthy and diseased leaves of various

plants, categorized by type of plant and disease (e.g., bacterial, fungal, or viral
infections).

 Format: All images are RGB and resized to a uniform dimension (e.g., 128x128 pixels)
for compatibility with the model input.

 Annotations: Labels are provided for each image, specifying the plant type and the
disease category.

4.2 Model Architecture:

 Base Model: Pre-trained models such as ResNet50, VGG16, or MobileNet are fine-
tuned to leverage transfer learning for better performance on limited data.

 Input Layer: Accepts preprocessed RGB images in a fixed size (128x128x3).

18
  Convolutional Layers: Extract spatial features through filters, with layers followed
by activation functions like ReLU.

 Pooling Layers: Apply MaxPooling to reduce dimensionality while preserving

important features.

 Dense Layers: Fully connected layers consolidate extracted features forfinal

classification.

 Output Layer: A SoftMax activation function outputs probabilities foreach class.

4.3 Flow Diagram:

19
4.4 Data Preprocessing:
 Image Resizing: All images are resized to a uniform dimension for consistent input size.
 Normalization: Pixel values are scaled to a range of 0 to 1 to accelerate model convergence
 Data Augmentation: Techniques like rotation, flipping, cropping, and brightness adjustments
are used to artificially expand the dataset andmitigate overfitting.

 Noise Reduction: Filters such as Gaussian Blur are applied to reduce noisein images.

 Label Encoding: Categorical labels are encoded into numerical values compatible with
the model.

4.5 Model Training and Optimization:

 Training: The dataset is split into training (80%) and validation (20%) sets. The
training set is used to fit the model parameters, and the validation set monitors
performance.

 Hyperparameter Tuning: Parameters like learning rate, batch size, and number of
epochs are optimized to improve the model’s efficiency.

 Loss Function: Categorical Cross entropy is employed to calculate the prediction error
for multiclass classification.

 Optimization Algorithm: Adam optimizer is used for faster convergence.

4.6 ML Algorithms and Techniques

 Convolutional Neural Networks (CNNs): Utilized for image feature
extraction and classification due to their high performance in visual data.

 Ensemble Techniques: Methods such as bagging and boosting areexplored to

enhance model robustness.

 Regularization: Techniques like dropout are applied to prevent overfitting.

20
4.7 Model Evaluation Metrics
To assess the performance of the machine learning models, the following metrics were used:
1. Accuracy: Proportion of correct predictions to total predictions.
2. Precision: Measure of how many positively predicted instances were actually correct.
3. Recall (Sensitivity): Ability of the model to identify all relevant instances.
4. F1-Score: Harmonic mean of precision and recall, balancing the two metrics.
5. Confusion Matrix: Visual representation of model performance across classes.
6. AUC-ROC Curve: Evaluated the trade-off between true positive rate and false positive rate

21
 CHAPTER 5

CODE IMPLEMENTATION

Description of the Implementation Process

Process of the plant disease detection model. It highlights the software tools andlibraries utilized,
and provides an overview of the steps involved in training and testing the machine learning model.

5.1 Software Tools, Libraries and Frameworks

 Programming Language: Python was used due to its simplicity andvast ecosystem of
libraries for machine learning.

 Development Tools: Google Collab: For writing andrunning Python scripts interactively.

 Key Libraries:

o NumPy: For numerical operations.

o Pandas: For handling and analyzing data.
o OpenCV: For image preprocessing and manipulation. Preprocessing tasks such as
image enhancement and filtering.

 Machine learning and Deep Learning:

o Scikit-learn: For implementing machine learning algorithms.
o TensorFlow/Keras or PyTorch: For training and deploying deeplearning models. For
building and training CNN architectures.

 Visualization:
o Matplotlib and Seaborn: For visualizing data distributions andresults.
o Plotly: For creating interactive visualizations.

22
5.2 Dataset Setup, Data Augmentation, and Preprocessing

5.3 Displaying Sample Images from the Training Set

23
5.4 Building the Convolutional Neural Network (CNN) Model
To develop an effective plant disease detection system, a Convolutional Neural Network (CNN)
was constructed. CNNs are a class of deep learning models that excel in analyzing visual data,
making them suitable for image-based tasks like disease identification in plants. The following
steps outline the development process of the CNN model.

5.5 Compile and Train the CNN Model

The training was conducted using the TensorFlow/Keras framework with GPU
acceleration. Loss functions like categorical cross-entropy and optimizers such as Adam were used
Regularization techniques like dropout and early stopping were implemented to prevent overfitting.
Key metrics like accuracy, precision, recall, F1-score, and confusion matrix were calculated to
measure the effectiveness of the model in classifying plant diseases.

24
 5.6 Plotting Training and Validation Accuracy and Loss
.

25
 CHAPTER 6

RESULT ANALYSIS

The result analysis section evaluates the performance of the developed plant disease detection model
and its implications for practical applications. The analysis involves assessing the model's accuracy,
reliability, and efficiency in detecting and classifying plant diseases based on visual data .

1.Model Performance Metrics: The results obtained after the experiments are analyzed using
various statistical parameters like Accuracy, Precision, Gain or Recall. These parameters are defined
by concepts of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN)
as follows:

TP+TN
Accuracy =
TP+TN+FP+FN
TP
Precision =
TP+FP
TP
Recall =
TP+FN

Precision . Recall
F1 Score = 2.
Precision+Recall

"True" and "negative" indicate the presence or absence of a fall, with "true positive" identifying a
fall and "true negative" confirming no fall. These metrics are crucial for assessing fall detection
system accuracy and performance.

2. Dataset Evaluation: The model was tested on a diverse dataset comprising multiple classes,

including healthy leaves and those affected by various diseases. Key observations include:

26
  Training and Validation Accuracy: The model achieved a high accuracy on both the training

and validation datasets, indicating effective learning and generalization.

 Class-wise Performance: Certain classes, such as those with visually distinct symptoms, were

classified with higher accuracy, while classes with subtle differences or overlapping features
showed lower performance

3. Overfitting and Generalization

 Training vs. Validation Performance: A comparison of training and validation losses revealed

no significant overfitting, as the model maintained consistent performance across both

datasets.

 Test Dataset Evaluation: When applied to a separate test dataset, the model achieved

commendable accuracy, demonstrating its robustness and ability to generalize to unseen data.

4. Results and Discussion:

The loss and accuracy are shown as:

 Training Accuracy: 99.41%: This metric indicates how well the model is performing on the
training dataset.It represents the percentage of correctly classified samples from the training set
during each epoch.

 Validation Accuracy:99.44%: This metric measures the model's performance on the validation
dataset, which is separate from the training data.

27
 Training Loss: 0.90%:This represents the error or difference between the predicted outputs and the
actual labels on the training dataset.It is calculated after each epoch to evaluate how well the model
is learning from the training data. Lower training loss indicates that the model is fitting the training
data better

 Validation Loss:1.41%: This quantifies the error on the validation dataset, which is a separate
subset not used for training.Validation loss helps monitor how well the model generalizes to
unseen data. If the validation loss is significantly higher than the training loss, it may indicate
overfitting.

28
C. USER INTERFACE

29
30
 CHAPTER 7

CONCLUSION

In this project, a machine learning-based system for plant disease detection was successfully
implemented. The model demonstrated the ability to accurately classify plant diseases usingimages of
plant leaves, offering a promising solution for early detection and management in agriculture. By
utilizing advanced deep learning techniques and a well-structured dataset, the system achieved
commendable performance in terms of accuracy and generalization. The integration of transfer
learning and data augmentation further enhanced the model's robustness, ensuring its applicability
in real-world scenarios

Scope for Future Research

1. Expanding the Dataset: Gathering a more diverse and larger dataset, including different plant
species and environmental conditions, can improvemodel performance.

2. Real-Time Detection: Developing a mobile or web-based application for real- time plant
disease diagnosis using live camera feeds could increase accessibilityand practicality.

3. Advanced Techniques: Implementing more sophisticated models like Generative Adversarial

Networks (GANs) for data generation or exploring other optimization algorithms could further
enhance classification accuracy.

4. Multi-Class Classification: Extending the model to handle multiple diseasessimultaneously,

potentially reducing the need for separate models for each disease.

31
 REFERENCE

[1]. PlantVillage Dataset - A dataset containing images of healthy and diseased

plantleaves: https://www.kaggle.com/datasets/emmarex/plantdisease.

[2]. Drive Link for Dataset - A dataset containing images of healthy and
diseased plant leaves: /content/drive/MyDrive/dataset

[3]. Plant Disease Detection Using Deep Learning - A research paper

discussing theapplication of deep learning techniques for plant disease detection:
https://arxiv.org/abs/2005.01437

[4]. Fruits and Vegetables Disease Classification Dataset - A dataset for

training models to classify plant diseases:
https://www.kaggle.com/datasets/rahul1010/plant-disease-dataset

[5]. Plant Disease Classification with TensorFlow - A tutorial on using

TensorFlow forplant disease classification:
https://www.tensorflow.org/tutorials/images/transfer_learning

[6]. S. D. Khirade and A. B. Patil, "Plant Disease Detection Using Image

Processing," 2015 International Conference on Computing Communication Control
and Automation, Pune, India, 2015, pp. 768-771, doi: 10.1109/ICCUBEA.2015.153.
keywords: {Diseases; Feature extraction; Image color analysis; Image segmentation;
Monitoring; Classification algorithms; Image acquisition; Segmentation; feature
extraction}

[7]. Kaur, Simranjeet, Geetanjali Babbar, Navneet Sandhu and Dr. Gagan Jindal.
“VARIOUS PLANT DISEASES DETECTION USING IMAGE PROCESSING
METHODS.” (2019).

[8]. Plant Disease Detection Using CNN

[9]. A Model Proposal for Enhancing Leaf Disease Detection Using Convolutional
Neural Networks (CNN): Case Study