Tomato Leaf Diseases Prediction

A PROJECT REPORT

Submitted by

S Dheyshnan[RA2211003011795]
K M Surya [RA2211003011763]
Under the Guidance of

Dr R Thilagavathy
Associate Professor
Department of Computing Technologies

in partial fulfillment of the requirements for the degree of

BACHELOR OF TECHNOLOGY
in
COMPUTER SCIENCE ENGINEERING

DEPARTMENT OF COMPUTING TECHNOLOGIES

COLLEGE OF ENGINEERING AND TECHNOLOGY
SRM INSTITUTE OF SCIENCE AND TECHNOLOGY
KATTANKULATHUR- 603 203

APRIL 2025
 Department of Computational Intelligence
SRM Institute of Science & Technology
Own Work* Declaration Form

This sheet must be filled in (each box ticked to show that the condition has been met). It must be
signed and dated along with your student registration number and included with all assignments you
submit – work will not be marked unless this is done.
To be completed by the student for all assessments

Degree/ Course: Bachelor of Technology CSE

Student Name : K M Surya, S Dheyshnan

Registration Number: RA2211003011763, RA2211003011795

Title of Work: Tomato Leaf Diseases Prediction

I / We hereby certify that this assessment compiles with the University’s Rules and Regulations
relating to Academic misconduct and plagiarism**, as listed in the University Website, Regulations,
and the Education Committee guidelines.

I / We confirm that all the work contained in this assessment is my / our own except where indicated,
and that I / We have met the following conditions:

● Clearly referenced / listed all sources as appropriate

● Referenced and put in inverted commas all quoted text (from books, web, etc)
● Given the sources of all pictures, data etc. that are not my own
● Not made any use of the report(s) or essay(s) of any other student(s) either past or present
● Acknowledged in appropriate places any help that I have received from others (e.g. fellow
students, technicians, statisticians, external sources)
● Compiled with any other plagiarism criteria specified in the Course handbook / University
website

I understand that any false claim for this work will be penalized in accordance with the University
policies and regulations.

DECLARATION:
I am aware of and understand the University’s policy on Academic misconduct and plagiarism and I certify that
this assessment is my / our own work, except where indicated by referring, and that I have followed the good
academic practices noted above.

If you are working in a group, please write your registration numbers and sign with the date for every
student in your group.

i
 SRM INSTITUTE OF SCIENCE AND
TECHNOLOGY KATTANKULATHUR – 603 203

BONAFIDE CERTIFICATE

Certified that 21CSP302L - Project report titled “Tomato Leaf Diseases

Prediction” is the Bonafide work of “K M Surya [RA2211003011763], S
Dheyshnan [RA2211003011795]” who carried out the project work under my
supervision. Certified further that to the best of my knowledge the work reported
herein does not form any other project report or dissertation on the basis of which
a degree or award was conferred on an earlier occasion or on this or any other
candidate.

SIGNATURE SIGNATURE

DR. R. Thilagavathy DR. G Niranjana

SUPERVISOR HEAD OF DEPARTMENT

ASSOCIATE PROFESSOR
DEPARTMENT OF
DEPARTMENT OF
COMPUTING TECHNOLOGIES
COMPUTING TECHNOLOGIES

EXAMINER 1 EXAMINER 2

ii
 ACKNOWLEDGEMENTS

We express our humble gratitude to Dr. C. Muthamizhchelvan, Vice-Chancellor, SRM Institute of

Science and Technology, for the facilities extended for the project work and his continued support.

We extend our sincere thanks to Dr. Leenus Jesu Martin M, Dean-CET, SRM Institute of Science
and Technology, for his invaluable support.

We wish to thank Dr. Revathi Venkataraman, Professor and Chairperson, School of Computing,
SRM Institute of Science and Technology, for her support throughout the project work.

We encompass our sincere thanks to, Dr. M. Pushpalatha, Professor and Associate Chairperson -
CS, School of Computing and Dr. Lakshmi, Professor and Associate Chairperson -AI, School of
Computing, SRM Institute of Science and Technology, for their invaluable support.

We are incredibly grateful to our Head of the Department, DR. G. Niranjana, Professor and Head
of Department of Computing Technologies, SRM Institute of Science and Technology, for her
suggestions and encouragement at all the stages of the project work.

We want to convey our thanks to our Project Coordinators, Panel Head, and Panel Members
Department of Computational Intelligence, SRM Institute of Science and Technology, for their inputs
during the project reviews and support.

We register our immeasurable thanks to our Faculty Advisor, DR. R. Thilagavathy, Department of
Computing Technologies, SRM Institute of Science and Technology, for leading and helping us to
complete our course.

Our inexpressible respect and thanks to our guide, DR. R. Thilagavathy , Department of Computing
Technologies, SRM Institute of Science and Technology, for providing us with an opportunity to
pursue our project under his / her mentorship. He / She provided us with the freedom and support to
explore the research topics of our interest. His / Her passion for solving problems and making a
difference in the world has always been inspiring.

We sincerely thank all the staff members of Department of Computing Technologies, School of
Computing, S.R.M Institute of Science and Technology, for their help during our project. Finally, we
would like to thank our parents, family members, and friends for their unconditional love, constant
support and encouragement

Authors
iii
 ABSTRACT

Tomato cultivation is a vital component of global agriculture, yet it is highly

susceptible to various leaf diseases that can significantly reduce yield and quality. Early
and accurate detection of these diseases is crucial for effective management and
prevention of crop loss. This project presents a machine learning-based approach for
the automated detection and classification of tomato leaf diseases using image
processing and deep learning techniques. A convolutional neural network (CNN)
model is trained on a publicly available dataset containing images of tomato leaves
affected by common diseases such as early blight, late blight, and leaf mold, as well as
healthy leaves. The results demonstrate the effectiveness of the model in achieving
reliable disease classification, offering a promising tool for precision agriculture and
sustainable farming [Link] methods of disease identification are often
manual, time-consuming, and require expert knowledge. This project presents a deep
learning-based approach for the automated detection and classification of tomato leaf
diseases using image data. The system utilizes Convolutional Neural Networks (CNNs)
and transfer learning techniques to train models on a publicly available dataset of
diseased and healthy tomato leaf images.

iv
 TABLE OF CONTENTS

ABSTRACT v

TABLE OF CONTENTS vi

LIST OF FIGURES vii

LIST OF TABLES viii

ABBREVIATIONS ix

CHAPTER TITLE PAGE

NO. NO.

1 INTRODUCTION 1
1.1 Introduction to Project 2
1.2 Problem Statement 3
1.3 Motivation 4
1.4 Sustainable Development Goal of the Project 5
2 LITERATURE SURVEY 6
2.1 Overview of the Research Area 7
2.2 Existing Models and Frameworks 8
2.3 Limitations Identified from Literature Survey (Research Gaps) 9
2.4 Research Objectives 10
2.5 Product Backlog (Key user stories with Desired outcomes) 12
2.5 Plan of Action (Project Road Map) 13

3 SPRINT PLANNING AND EXECTION METHODOLOGY 14

3.1 SPRINT I 14
3.1.1 Objectives with user stories of Sprint I 15
3.1.2 Functional Document 17
3.1.3 Architecture Document 19
3.1.4 Outcome of objectives/ Result Analysis 21
3.1.5 Sprint Retrospective 22
3.2 SPRINT II 23
3.2.1 Objectives with user stories of Sprint II 23
3.2.2 Functional Document 25
3.2.3 Architecture Document 26
3.2.4 Outcome of objectives/ Result Analysis 27
3.2.5 Sprint Retrospective 28

v
4 RESULTS AND DISCUSSIONS 30
6.1 Project Outcomes (Performance Evaluation, Comparisons, Testing Results )

5 CONCLUSION AND FUTURE ENHANCEMENT 34

REFERENCES 36
APPENDIX 37
A CODING 37
B CONFERENCE PUBLICATION 40
C PLAGIARISM REPORT 41

vi
 LIST OF FIGURES

CHAPTER PAGE
NO. TITLE NO.

2.2 (i) Sample curl virus 8

2.2 (ii) Sample mosaic virus 8

3.1 System architecture 21

4.1 Training vs validation Accuracy 31

4.2 Training vs Validation Loss 31

4.3 Testing Results 33

vii
 LIST OF TABLES

FIGURE TITLE PAGE

NO. NO.

2.6 Limitations and Descriptions 10

2.7 Product Backlog 12

2.8 Plan of action 13

3.1 Objectives of user story 16

3.1 Functional Document 18

viii
 ABBREVIATIONS

AI - Artificial Intelligence

ML - Machine Learning

CNN - Convolutional Neural Network

UI - User Interface

API - Application Programming Interface

CRUD - Create, Read, Update, Delete

JSON - JavaScript Object Notation

IoT - Internet of Things

SDG - Sustainable Development Goals

RF - Random Forest

SVM - Support Vector Machine

TFLite - TensorFlow Lite

UAT - User Acceptance Testing

GLCM - Gray Level Co-occurrence Matrix

EC2 - Elastic Compute Cloud

ix
 CHAPTER 1

INTRODUCTION

Agriculture forms the pillar of the majority of economies in nations, and tomato
(Solanum Lycopersicon) ranks among the most grown and utilized vegetable crops worldwide.
They form the pillar in the agricultural sector on the basis of their nutritional content, market
demand, and economic value. Tomato cultivation is, however, faced with a chain of plant
diseases, specifically leaf infections. Early blight, late blight, leaf mold, bacterial spot, and
Septoria leaf spot are some diseases that can potentially reduce yield significantly, impair crop
quality, and result in severe economic losses to farmers.
Historically, tomato leaf disease has been determined by manual scouting by farmers or
agronomists. Although efficient when done by experienced practitioners, this process is not
without challenges. Manual determination is subjective, time-consuming, and not scalable for
large estates. The phenotypic presentation of various diseases may also mimic one another and
hence result in misdiagnosis or delayed diagnosis. This would result in extensive infection,
wastage of pesticides, and subsequently, lower harvest yields.
The growing need for precision agriculture and sustainable agriculture has spurred the
demand for new, innovative technologies that will help farmers with precise, real-time
information. Advances in Artificial Intelligence (AI), particularly Machine Learning (ML) and
Computer Vision, have provided new possibilities for the diagnosis of plant diseases. Of all
such technologies, Convolutional Neural Networks (CNNs) have performed remarkably well
in image classification and, therefore, have proven to be extremely effective in identifying
patterns of diseases on the leaves of plants. This project intends to utilize the power of deep
learning methods, in this case, CNNs, to create an intelligent system for detecting tomato leaf
diseases. The system will be trained on a large pool of images of tomato leaves—both healthy
and unhealthy—and learned to classify images into several categories of diseases.
The model will be made to learn to provide high accuracy using preprocessing methods,
data augmentation, and model optimization. The ultimate goal of this project is to create an
easy-to-use and functional tool which can be installed on mobile phones or embedded in web-
based applications. The tool will enable farmers, farmhands, and agronomists to diagnose
tomato leaf diseases remotely and accurately in real time, which will lead to faster intervention,
improved crop management, and reduced crop loss.

1
1.1 Introduction to Project:

Tomato (Solanum Lycopersicon) is one of the most widely cultivated and economically
valuable vegetable crops around the world. It serves as a staple in both subsistence and
commercial agriculture due to its nutritional benefits and versatility in culinary applications.
However, tomato cultivation is frequently threatened by a variety of diseases, particularly those
that affect the leaves. These diseases—such as early blight, late blight, leaf mold, bacterial
spot, and Septoria leaf spot—not only reduce the yield and quality of tomatoes but also cause
significant economic losses for farmers each year. The early detection and diagnosis of these
diseases is critical to implementing effective treatment and preventing large-scale crop
damage.

Traditionally, disease detection has relied on manual visual inspection by farmers or

agricultural experts. While this approach can be effective for experienced individuals, it has
several limitations. Manual inspection is time-consuming, labor-intensive, and often
subjective. The accuracy of diagnosis may vary depending on the individual’s knowledge,
experience, and environmental conditions. Furthermore, early-stage symptoms of many
tomato leaf diseases are visually similar, making precise identification difficult without
laboratory tests—which are not always accessible or affordable, especially in developing
regions.

In recent years, advancements in artificial intelligence (AI), machine learning (ML), and
computer vision have provided promising alternatives to traditional methods. These
technologies enable the automatic detection and classification of plant diseases using digital
images, offering a faster, more consistent, and scalable solution. In particular, deep learning
techniques—especially Convolutional Neural Networks (CNNs)—have shown remarkable
performance in image recognition tasks, including agricultural applications.

This project aims to develop an intelligent tomato leaf disease detection system using
CNN-based deep learning models. The system is trained on a comprehensive dataset
containing images of both healthy and diseased tomato leaves. It leverages image
preprocessing techniques, data augmentation, and model optimization strategies to enhance
prediction accuracy and generalization. The ultimate goal is to build a user-friendly, real-time
detection system that can be integrated into mobile or web-based platforms.

2
1.2 Problem Statement:

Tomato plants are highly susceptible to a range of leaf diseases such as early blight, late
blight, leaf mold, bacterial spot, and more. These diseases significantly reduce crop yield and
quality, leading to considerable economic losses for farmers worldwide. Early and accurate
identification of these diseases is essential for timely intervention and effective disease
management.

However, traditional methods of disease detection—based on manual visual

inspection—are often unreliable, time-consuming, and require expert knowledge. In many
rural or resource-constrained areas, access to trained agricultural professionals and diagnostic
tools is limited, resulting in delayed or incorrect diagnoses. Additionally, the visual symptoms
of different diseases can appear very similar, making it difficult for even experienced farmers
to distinguish them accurately.

There is a growing need for an automated, reliable, and scalable solution that can assist
in the early detection and classification of tomato leaf diseases. Such a system should be
accessible, easy to use, and capable of functioning in real-time to support farmers in making
informed decisions.

This project addresses this problem by developing a deep learning-based model—

specifically using Convolutional Neural Networks (CNNs)to automatically detect and classify
tomato leaf diseases from digital images. The solution aims to enhance disease diagnosis
accuracy, reduce dependency on human expertise, and contribute to more efficient and
sustainable agricultural practices.

1.3 Motivation:
A vital component of global agriculture, tomato farming greatly boosts both the
economy and food security. In addition to being a staple food in many nations, tomatoes are a
significant crop in terms of production and trade. However, growing tomatoes presents many
difficulties, especially because leaf diseases are so common. These illnesses, which include
bacterial spot, leaf mold, and early and late blight, can significantly lower crop yield and
quality.

These diseases early onset can harm tomato plants irreparably, drastically reducing farm

3
productivity and causing large financial losses. Farmers are constantly under pressure to
increase yield and reduce loss from pests, diseases, and unfavorable weather, especially in
developing nations. Regretfully, the process of manually identifying tomato leaf diseases is
still difficult and unreliable. The visual indicators of various diseases frequently overlap, which
can result in misdiagnosis, even though skilled agricultural specialists may be able to recognize
specific disease symptoms. Since early-stage symptoms are subtle and may mimic those
brought on by dietary deficiencies or environmental stress rather than infections, they can be
particularly difficult to identify. The use of labor-intensive and subjective inspection
techniques exacerbates this issue even more. Large fields must usually be inspected by hand
by farmers, and disease detection is frequently delayed in areas with a lack of or in access to
qualified agricultural specialists. A disease may have already spread throughout the crop by
the time it becomes clearly visible, causing extensive damage that could have been avoided
with prompt action. Traditional disease control methods, such as excessive pesticide use,
consequently become the norm, endangering human health and the environment needlessly.
Given these difficulties, a more effective, dependable, and scalable method of disease detection
in tomato crops is obviously required. The goal of closing the technology gap in agriculture is
another factor driving this project. Smallholder farmers in many developing nations face major
obstacles because they lack access to modern agricultural equipment and knowledge. This
project intends to make state-of-the-art technology available to farmers in underserved and
rural areas by creating an inexpensive and user-friendly disease detection system. This system
can be made available to farmers on their smartphones through mobile apps and web-based
platforms, giving them immediate access to disease diagnosis and treatment recommendations.
Additionally, by ensuring that farmers can use AI to enhance their incomes regardless of their
location or level of technical expertise, this project advances the broader objective of digital
inclusion in agriculture. Agriculture is in a unique position to gain from AI's ongoing
transformation of many industries, particularly in terms of enhancing food production, quality,
and sustainability. To sum up, the Tomato Leaf Diseases Detection project is driven by the
desire to use cutting-edge technology to address the urgent problem of crop disease
management.

1.4 Sustainable Development Goal of the Project:

A number of the Sustainable Development Goals (SDGs) of the UN are directly

supported by the Tomato Leaf Diseases Detection project, which encourages responsible
innovation in the farming industry, improves food security, and promotes sustainable

4
agriculture. The following SDGs are the most pertinent to this project:

SDG 2: No Hunger

which lowers crop losses and boosts food production. The system plays a vital role in lowering
hunger and guaranteeing food availability by assisting farmers in identifying diseases early on,
which leads to healthier crops and higher yields. Which aims to end hunger, achieve food security,
and promote sustainable agriculture. By leveraging image processing and deep learning
technologies, the system enables early and accurate identification of diseases in tomato plants,
significantly reducing crop loss and increasing agricultural productivity.

Justification:

Tomato leaf disease prediction contributes directly to SDG 2: Zero Hunger, which aims to end
hunger, achieve food security, and promote sustainable agriculture. Specifically:

A) Increased Agricultural Productivity (Target 2.3):

Early detection and prediction of tomato leaf diseases allow farmers to take timely action,
reducing crop losses and increasing yield. This is particularly beneficial for smallholder farmers,
who rely heavily on healthy crops for their livelihoods.

B) Sustainable and Resilient Farming (Target 2.4):

Disease prediction systems support sustainable farming by minimizing the overuse of

pesticides. Instead of applying chemicals preventively, farmers can use them only when a disease
is likely, reducing environmental impact and promoting resilience against climate variability.

SDG 12: Responsible Consumption and Production:

Justification:

A) Minimizing Waste and Crop Loss:

By detecting diseases early, farmers can take preventive actions to reduce crop damage,
minimizing food waste during production and post-harvest stages.

C) Promoting Sustainable Agricultural Inputs:

Better disease management reduces the reliance on unsustainable practices and encourages
more balanced input-output cycles in farming.

5
 CHAPTER 2
LITERATURE SURVEY

Tomato crops are vulnerable to a wide variety of diseases that negatively impact yield,
quality, and economic returns. Detecting these diseases accurately and early is crucial to avoid
significant losses. Over the years, researchers have explored several traditional and modern
approaches to address this problem. The following literature review highlights the evolution
of tomato leaf disease detection methods, focusing on image processing, machine learning,
and deep learning techniques.

2.1 Traditional Methods:

Earlier approaches to plant disease detection were largely manual. Farmers relied on their
experience or agricultural experts for diagnosis, which often led to late or inaccurate
identification due to the subtle differences in visual symptoms among various diseases.
Traditional laboratory-based techniques such as polymerase chain reaction (PCR), ELISA, and
culturing pathogens, although accurate, are time-consuming, expensive, and not feasible for
on-field use.

2.2 Image Processing Techniques:

In the early 2000s, researchers began exploring image processing techniques for
automatic disease detection. Methods involved segmentation, color analysis, texture
extraction, and feature engineering using tools like: K-means clustering for lesion
segmentation, Gabor filters for texture features, and Histogram analysis for color features.

2.3. Machine Learning Approaches:

CNN are particularly effective in image-based classification tasks due to their ability to
automatically extract and learn spatial hierarchies of features such as leaf color, texture, shap and
the presence of disease-specific lesions. Common CNN architectures like VGG16, ResNet,
InceptionNet, and MobileNet have been successfully applied to classify various tomato leaf
diseases, including early blight, late blight, and leaf mold. In many cases, these models are
enhanced through transfer learning, which involves using pr-trained networks originally trained on
large datasets such as ImageNet and fine-tuning them with tomato leaf disease datasets. This not

6
 only reduces the computational cost and training time but also significantly boosts the accuracy,
even when working with relatively small, domain-specific datasets. In addition to deep

2.4 Overview of the Research Area:

The growing demand for food security, sustainable farming, and effective crop
management worldwide has made the nexus between agriculture and artificial intelligence (AI)
a crucial area of study in recent years. The early identification and diagnosis of plant diseases,
especially in economically important crops like tomatoes, has become a crucial area of focus
among the many challenges facing agriculture. Early blight, late blight, bacterial spot, and leaf
mold are just a few of the leaf diseases that can seriously affect crop quality and yield and
affect tomato plants (Jha et al., 2019).

Plant diseases have traditionally been identified by manual inspection, which is frequently
arbitrary, inconsistent, and constrained by the availability of expert knowledge, particularly in
isolated or impoverished areas. Combining aspects of computer vision, machine learning, deep
learning, image processing, and mobile computing, automated plant disease detection has
consequently emerged as a top research priority (Ferentinos, 2018).

The application of Convolutional Neural Networks (CNNs), a subset of deep learning

models created especially for image analysis, is among the most revolutionary advancements
in this field. CNNs are especially well-suited for classifying plant diseases based on images of
leaves because of their remarkable performance in recognizing visual patterns. This method
removes the need for manually created feature engineering by enabling automatic feature
extraction and classification.

To train and assess models that can identify a variety of plant diseases, researchers have
developed and utilized a variety of image datasets, such as the Plant Village dataset. Thousands
of annotated photos of both healthy and diseased leaves in controlled settings are commonly
found in these datasets. In order to perform well in real-world settings, where variables like
lighting, background clutter, and leaf orientation vary greatly, efforts have more recently
concentrated on improving the model's robustness.

The application of transfer learning, which makes use of pre-trained CNN architectures
(such as VGG16, Res Net, and Mobile Net) and modifies them for plant disease detection
tasks, is another expanding area within this research field. Effective models can be deployed
more easily, even in environments with limited resources, thanks to this method's significant
reduction in training time and data requirements.
7
 Another important area of current research is the integration of these AI-driven models
with IoT-based agricultural systems and mobile applications. By enabling farmers to take
pictures of their leaves with smartphones and get prompt disease diagnosis and treatment
recommendations, these apps democratize access to agricultural knowledge and lessen reliance
on physical infrastructure.

2.5 Existing Models and Frameworks:

1. Conventional Models for Machine Learning:

After features like color histograms, edge descriptors, and GLCM (Gray Level Co-
occurrence Matrix) are extracted, Support Vector Machines (SVM) are commonly used for
classification. Random Forest (RF): An ensemble model that enhances classification
performance on structured feature sets by utilizing decision trees Fig 2.2 (curl virus).

Fig 2.2 a Sample curl virus Fig 2.2 b sample mosaic virus

2. CNN-Based Deep Learning Models:

Convolutional Neural Networks (CNNs), in particular, are the state-of-the-art in deep
learning because they can automatically extract features from unprocessed images. A lot of
research develops unique architectures that consist of three to five convolutional layers,
followed by dense layers. These perform well with small datasets fig b(mosaic virus) . If not
trained on a variety of datasets, it might not generalize well. Transfer learning using pre-trained
models is a common solution. These models are refined for tomato disease classification after
being trained on sizable datasets (such as ImageNet) (Mohanty et al., 2016).
VGG16 and VGG19 are deep architectures renowned for their robust performance and
ease of use. efficient in transfer learning environments. ResNet50 and ResNet101 are well-
known for their residual connections, which enable deeper network training.

8
frequently demonstrates high accuracy in tasks involving disease classification.
MobileNet and MobileNetV2 are lightweight models that are perfect for embedded and mobile
deployment. gives up some accuracy in favor of efficiency and speed. renowned for its high
image classification task accuracy and multi-scale convolutions. An effective architecture that
enhances feature reuse by feed-forwardly connecting each layer to every other layer.

2.6 Limitations Identified from Literature Survey:

1. Limited Diversity of Datasets:

The majority of models are trained using publicly accessible datasets, such as Plant
Village, which are frequently gathered in controlled settings with uniform lighting and
backgrounds. Absence of noise from the real world (such as different lighting, overlapping
leaves, and camera angles).little information about mixed infections or rare diseases.

2. Difficulties with Generalization:

Frequently, models that are trained on high-quality lab images do not translate well to
actual farm conditions. When deployed in fields with the following, performance drastically
declines:
 changing lighting circumstances.
 background noise.
 overlapping leaves or partial occlusions.

3. Disease Likeness:
It can be challenging for models to differentiate between various diseases because
their visual symptoms can be so similar:
For instance, in their early stages, early blight and septoria leaf spot may look alike.
causes disease categories to be misclassified or confused.

9
4. Limitations in Computation:
CNNs and other deep learning models might need:
GPUs with high performance.
large memory and storage, which might not be accessible for mobile apps or on-field
deployment.
5. Differences Absence of Standardization in:
 methods for preprocessing images.
 techniques for augmentation.
 Metrics for evaluating the model.
 Make it challenging to compare the performance of the model in various studies.

TABLE: 2.6 Limitations and description

Limitations Description
Poor Field Models perform poorly under varied lighting,
Adaptability backgrounds and angels

Domain Shift Models trained on clean data do not generalize well to

field conditions

2.7 Product Backlog:

User story 1: Farmer
● As a Farmer.
● I want to upload a picture of a tomato leaf so that I can detect if the
plant is diseased
● so that I can understand what’s affecting my crop.
● Desired Outcome: The system accepts and processes the image, returning a disease or
"healthy" result.

User Story 2: Agronomists/Experts:

● As an Agronomists/Experts
● I want to verify disease predictions and provide expert advice to farmers.
● So that so that I can track the progression of diseases and the effectiveness of
interventions.
● Desired outcome: A backend portal allows agronomists to review and annotate .
10
User story 3: System Admins:
● As a system admin
● I want to manage user accounts and permissions so that only authorized users can access
certain features.
● so that the system remains accurate and up to date with the latest disease information.
● Desired outcome: Admin panel for creating roles, managing access rights.

User story 4: General Users:

● As a general user.
● I want to understand the symptoms of each disease so that I can visually check my
plants.
● So that I can visually inspect and identify issues in my plants more effectively.
Desired outcome: A disease info page with symptom descriptions.

11
Table: 2.7 Product Backlog

ID
SS 1 2 3 4 5

Image
Preprocessing TrainML Predicti Result
Title Upload
Pipeline Model on API Dashboard
Module

Model
Data Data Model User
Epic Integrat
Collection Processing Development Interface
ion

(Moscow)
Must Must Must Should Could
Priority

Status To Do In Progress To Do To Do To Do

API
User can Accuracy ≥ User sees
Acceptance Images resized returns
upload leaf 90% on test prediction
Criteria & normalized disease
images data result
class

Upload Image REST Result

Function Model Training
Interface Processing API Display

Non- Clean,
Mobile- Efficient Response <
Functional Reproducibility readable
friendly UI memory use 1s
Requirement output

Original
4h 5h 8h 6h 3h
Estimate

Actual
4h 2.5h 5h 5h 1h
Effort

12
Table 2.8 Plan of action

[Link]. Task Description

Problem Definition & Literature Define objectives, review research papers

1
Review on tomato leaf disease detection
Collect tomato leaf images (e.g., Plant
2 Dataset Collection
Village dataset)
Resize, normalize, label images, augment
3 Data Preprocessing
data
Model Selection & Architecture Choose model(s) like CNN, ResNet;
4
Design define architecture
Train model, tune hyperparameters,
5 Model Training & Validation
validate with split data
Test on unseen data, evaluate accuracy,
6 Model Testing
precision, recall
Analyze results, visualize performance
7 Result Analysis & Visualization
using charts
Write the project report including
8 Report Writing
methods, results, conclusion
Presentation & Demo Create PowerPoint, prepare for final
9
Preparation presentation

Despite being one of the most extensively grown crops in the world, tomatoes are
extremely vulnerable to a number of leaf diseases that can drastically lower crop quality and
productivity. For sustainable agriculture and efficient pest management, early and precise
disease diagnosis is essential. Conventional illness identification techniques frequently depend
on manual inspection, which can be laborious, prone to mistakes, and necessitate specialised
knowledge. Convolutional Neural Networks (CNNs), a deep learning technology.

13
 CHAPTER 3

SPRINT PLANNING AND EXECUTION METHODOLOGY

We used the Agile methodology for this project, placing a lot of emphasis on the
Scrum framework. This method made evolutionary delivery, adaptive planning, and
incremental development easier. The project was broken up into several sprints, each
with distinct objectives and outputs. Agile encourages accountability, collaboration, and
iterative advancement toward a clear objective. One of the most widely used Agile
approaches, Scrum offers an organized framework that facilitates all stages of
development, from planning to delivery.

The entire team was split up into smaller groups that were in charge of front-end,
back-end, testing, and documentation, among other project-related tasks. To preserve
workflow discipline, scrum rituals like sprint planning, daily stand-ups, sprint reviews,
and sprint retrospectives were closely adhered to. Task tracking and communication
were facilitated using tools such as Trello, GitHub, and Slack.

We were able to enhance teamwork, react swiftly to changes, and gradually

produce a high-caliber product thanks to this methodology. Every sprint's planning and
execution tactics are thoroughly documented. The thorough documentation of Sprint I,
including its planning, execution, and outcome analysis, is provided below.

3.1 SPRINT I:
Sprint I, which was the project's start phase, was mostly concerned with establishing
the framework. The goal of this first sprint, which lasted two weeks, was to set up all the
necessary elements needed for the next sprints. These included completing the
development stack, configuring the tools, designing the backend schema, starting the
authentication module, and producing basic UI wireframes.
Onboarding the team to the tools and techniques used for this project was a major priority
during Sprint I. This covered responsibilities, task distribution, and setup instructions. To

14
confirm expectations, time was also devoted to stakeholder meetings and requirement.

Basic connectivity was tested, and the frontend and backend development environments
were set up. Together, the database schema was drafted, and authentication methods were
determined. The user interface's wireframes were made and approved by both internal and
external reviewers. The system architecture was thoroughly documented and connected to
the project objectives.

The foundation for a well-organized and maintainable application had been effectively
established by the conclusion of Sprint I. In addition to providing clarity on execution, it
raised team morale by demonstrating progress.

3.1.1 Objectives with User Stories of Sprint I

Objectives:
● Provide a consistent environment for all contributors to develop in.
● Create scalable and safe user authorization and authentication systems.
● Establish and put into practice the fundamental database schema.
● Create and review user interface wireframes for key screens.
● Start documentation procedures and match them to sprint tasks.
● Every goal was linked to tasks and user stories that were recorded in the project
management application. Mid-sprint evaluations and daily stand-up meetings were
guided by objectives.

User Stories:
User Story 1: I want to make a safe account as a user so that I can log in and utilize the
system according to my own preferences.

User Story 2: I want a structured database schema as a developer to facilitate quick

queries and future feature integrations.

User Story 3: In order to uphold data integrity and enforce system rules, I want to manage
15
user roles and permissions as an administrator.

Table 3.1 Objectives of user stories

Sprint Planned Activities Status Comments

Dataset acquisition Used PlantVillage,

Sprint 1 ✅ Completed
and cleaning cleaned noisy data

Image Resized,
Sprint 2 preprocessing and ✅ Completed normalized, applied
augmentation augmentation

Model design and CNN model built

Sprint 3 ✅ Completed
training using TensorFlow

Evaluated with
Model evaluation
Sprint 4 ✅ Completed accuracy, loss,
and optimization
confusion matrix

API development Flask API for image

Sprint 5 ✅ Completed
for prediction prediction

UI development for Upload feature with

Sprint 6 ✅ Completed
user interaction result display

Integration of model Backend integrated

Sprint 7 ✅ Completed
with UI with frontend

Display disease info

Added info cards
Sprint 8 (symptoms, ✅ Completed for each disease
remedies, etc.)

Final testing, Verified accuracy

Sprint 9 polishing, and bug ✅ Completed and UI
fixes responsiveness

User Story 4: For the frontend team to comprehend and develop the design, I, as a UI/UX
designer, require clear wireframes.

User Story 5: In my capacity as project manager, I want thorough documentation to

facilitate knowledge transfer and make it simple for new developers to onboard.

16
These narratives were divided into assignments with due dates. The team assigned tasks
according to the availability and expertise of each member. To guarantee logical
progression, dependencies between tasks were identified.

3.1.2 Functional Document

The functional document lists the modules created during Sprint I along with their
primary features. Inputs, outputs, user interactions, and interdependencies are all included
in each module. Developers, testers, and other stakeholders can use these documents as a
guide to comprehend the system's functional behavior.

Module 1: Verification of Users

● New users can sign up for the system using their email address and password.
● Bcrypt is used to encrypt the password before it is saved.
● A JWT token, which is used for session management, is issued upon login.
● To distinguish between user and admin privileges, role-based access is enforced during
login.
● Users can troubleshoot common errors with the help of detailed login failure messages.

Module 2: Interface for the Dashboard

● The user is taken to a customized dashboard upon logging in.

● Users can access available features, view system alerts, and manage their profiles.
● Additional controls, such as logs and user management, are accessible to
administrators.
● The dashboard adapts to various screen sizes and is responsive.

17
Table :3.1 Functional Document

Module Name Functionality Inputs Outputs

Allows user to upload

Uploaded image ready
Image Upload tomato leaf images for Leaf image (JPG/PNG)
for processing
analysis

Applies image
Processed image for
Preprocessing processing (resize, Raw leaf image
model input
normalize, augment)

Extracts relevant
Feature Extraction features from the image Preprocessed image Feature vector
using CNN layers

Predicts the disease

Feature vector / Disease label (e.g.,
Disease Classification class using a trained
Processed image Early Blight, Healthy)
deep learning model

Shows prediction result Disease name &

Result Display Model output
with confidence score confidence (%)

Provides info about

predicted disease, Disease details
Disease Info Retrieval Disease label
symptoms, and (text/images)
remedies

Web or app interface

Navigation, uploads,
User Interface (UI) for interacting with the User actions
results view
system

18
Module 3: Administration of Data

● A normalized MySQL database contains user data.

● Every essential data entity has CRUD operations implemented.
● Register, /login, /profile, and /logout are examples of backend endpoints.
● Data consistency is achieved using database triggers and constraints.

Module 4: Alerts

● Based on user activity, in-app notifications are produced.

● Admin messages, system updates, and login alerts are examples of notifications.
● The dashboard user interface incorporates a notification center.
● Before each module was declared finished for the sprint, it underwent internal review
and testing.

3.1.3 Architecture Document

The system's architecture is set up to guarantee maintainability, scalability, and
performance. A layered architecture was chosen, with each layer handling a distinct
application concern.

1. Frontend Client Layer

The goal of this section of the program is to facilitate user engagement.

In this undertaking:
Gradio , a Python package that makes user interface development easier, was used in its
construction.
• Users can upload pictures of leaves to receive forecasts.
• Features such as an image upload box make up the simple and user-friendly display.

19
2. Layer of Applications:
Includes Python methods that:
 Preprocess images (resize, normalize, etc.)
 Put the machine learning model into use.
 Make predictions using the picture that was uploaded.

3. The Backend Layer

● Developed using Flask APIs with RESTful endpoints.
● Cloud messaging and authentication can be done with Firebase.
● Secure communication is ensured by input validation, sanitization, and error handling.
● APIs adhere to RESTful conventions and support the JSON format.

4. Configuring Deployment
● Containerized development and testing are done with Docker.
● Using Nginx as a reverse proxy, the initial deployment was set up on an AWS
EC2 instance.
● GitHub Actions is used to plan CI/CD pipelines.
● The configuration and deployment (Fig 3.1) phase focused on setting up the
environment for model training and integrating the trained model into a user-
accessible application. The model was developed using Python with libraries such
as TensorFlow, Keras, NumPy, and OpenCV. For efficient execution and training,
Google Colab was used, which provided access to GPU resources and helped
reduce training time. Once the model achieved satisfactory accuracy, it was saved
in a serialized format (e.g., .h5 file) for easy integration. The application was tested
locally and can be hosted on cloud platforms like Heroku or Streamlit Community
Cloud for broader accessibility. This deployment demonstrates how AI-powered
solutions can be transformed into practical tools to support farmers and
agricultural professionals in disease diagnosis and decision-making (Choudhary et
al., 2021).

20
 Fig 3.1 SYSTEM ARCHITECTURE

3.1.4 Outcome of Objectives / Result Analysis

In Sprint I, every goal was tracked for completion status, obstacles, and output quality.
Deliverables, stakeholder feedback, and sprint goals are used to evaluate the results.

Findings:
● All developers were able to access and successfully set up the development
environment.
● Modules for session management and authentication were examined and verified.
● After receiving favorable comments, the UI wireframes were given to the frontend
team.
● Dummy data was used to test and implement the database schema.
● Architecture and functional documentation drafts were sent in for evaluation.

Evaluation:
● According to preliminary findings, foundational modules are scalable and reliable.
● Trello, GitHub, and Slack were among the tools that the team used to collaborate
efficiently.
21
● Proactive risk management was used, and problems were fixed during daily stand-ups.
● Task time estimates were largely accurate; documentation showed a few minor delays.
● A clean and consistent codebase was maintained with the aid of peer code reviews.

Important Takeaways:

● Early wireframe validation is crucial for avoiding rework.

● Tracking was enhanced by standardized commit messages and branch naming
conventions.
● In subsequent sprints, continuous integration techniques ought to be implemented
earlier.
● Time-boxing tasks ensured focus but need better planning in overlapping
modules.

3.1.5 Sprint Retrospective

All team members participated in a sprint retrospective to discuss what
worked, what didn't, and what could be done better. The retrospective was
organized as a free-flowing conversation with practical recommendations.

What Worked Well:

● The majority of the team's tasks were finished on schedule.

● Effective cross-functional cooperation between the frontend and backend teams
was achieved.
● Frequent stand-up meetings aided in the prompt resolution of problems.
● The final deliveries had fewer bugs thanks to internal testing.

What Could Be Made Better:

● The review of the documentation was prone to rework and took longer than
22
anticipated.
● When frontend and backend tasks overlapped, there were communication
breakdowns.
● During the initial setup phase, deployment guides were not completely updated.

Items of Action:

● Use designated reviewers to implement version-controlled documentation.

● To match frontend and backend expectations, schedule mid-sprint demos.
● Use Postman collections to automate repetitive tasks like testing APIs.
● When planning, include buffer tasks to account for unforeseen obstacles.

In order to improve performance in Sprint II, the team used this retrospective to
identify and fix inefficiencies.

3.2 SPRINT II
The main features of the Tomato Leaf Disease Detection system were the
focus of Sprint II. Creating and integrating a machine learning model that could
recognize common tomato plant diseases from leaf photos was the main objective.
The integration of AI features into the Flutter mobile app, model deployment
techniques, and user interface enhancements were also covered in this sprint.

Sprint II lasted for two weeks. ML engineers, frontend developers, and cloud
engineers collaborated across teams during this sprint. To guarantee that all
modules are integrated seamlessly, the team separated tasks based on
specializations while staying in close contact.

Testing the model on a variety of datasets and fine-tuning it for mobile device
performance were given particular attention.

23
Following an analysis of the stakeholder input from Sprint I, several enhancements
were added to the Sprint II backlog. To lessen server dependency, the emphasis was
moved to improving usability and integrating edge-based AI features. Between the
MVP foundation and an AI-integrated, production-ready application, Sprint II
served as a link.
3.2.1 Objectives with user stories of Sprint II
1) Goals:

● Gather and prepare a labeled dataset of pictures of tomato leaves.

● To classify diseases, train a Convolutional Neural Network (CNN) model.
● For mobile use, optimize and convert the trained model to TensorFlow Lite
(TFLite) format.
● Connect the model to the camera module and incorporate it into the Flutter
application.
● Improve the user interface to show results, including recommendations and
diseases found.

2) User Narratives:
● User Story 1: As a farmer, I would like to be able to snap a photo of a tomato leaf and
get an immediate diagnosis of the disease.

● User Story 2: In order to train a high-accuracy CNN model, I need a clean, labeled
dataset as a data scientist.

● User Story 3: In order for the model to function well on mobile devices, I, as a
developer, wish to convert it to TFLite.

● User Story 4: As a user, I want an intuitive interface that clearly shows results and
offers instructions for what to do next.

● User Story 5: In my capacity as a tester, I wish to confirm the model's accuracy using
both test and unobserved real-world data.

During daily standups, tasks were monitored and evaluated. Feedback loops aided in the
early detection of problems, particularly with app performance and model accuracy.

24
3.2.2 Functional Document
The disease detection feature was at the heart of Sprint II's functional
requirements. The app interface, result visualization, and machine learning pipeline
all had specific functionalities defined.

Feature 1: Obtaining Images

● Users can use the camera on their mobile device to take a picture of a tomato leaf.
● There was also the option to upload from the gallery.
● Prior to being fed into the model, the image is cropped and preprocessed.

Feature 2: Integration of Models

● Several tomato diseases, such as Septoria, Early Blight, and Leaf Mold, are
recognized by the CNN model.
● Each class's probabilities are provided by the model, and the best prediction is
shown.
● During prediction, image quality is improved by edge detection and enhancement
techniques.

Feature 3: Display of Results and Suggestion

● A confidence percentage for the disease result is shown.

● A brief explanation of the illness and its treatments is included with each
result.
● Suggested pesticides, irrigation practices, or isolation strategies are
examples of recommendations.

Feature 4: Tracking Performance

● The prediction time is computed and recorded.

● The user has the option to report inaccurate predictions.
● Logs are kept for analysis of model improvement.
25
3.2.3 Architecture Document

The integration of a machine learning model with the mobile application was
the main goal of the Sprint II architecture. The system used a hybrid strategy that
combined cloud-based feedback logging with local inference.

1. The Data Layer

● The dataset was manually labeled and sourced from publicly available plant
pathology datasets.
● Rotation, flipping, and noise were used as augmentation techniques to enhance
generalization.
● For training, the dataset was locally stored and transformed into TFRecords.

2. Layer of Models
● The CNN architecture was selected due to its capacity to process features in
images.
● TensorFlow was used to train the model, and an 80/20 split was used for
validation.
● On test data, accuracy exceeded 92%.
● Quantized and converted to TFLite to boost speed and decrease size.

3. Layer of Integration

● A platform channel is used by the Flutter application to connect to the TFLite

interpreter.
● A lightweight plugin called tflite_flutter is used to process user input (images).
● The prediction is parsed into output that is readable by humans and returned in
JSON format.

26
4. Layer of Interface

● For visual clarity, the user interface was created using Flutter widgets and
animations.
● Users can report inaccurate predictions using this feedback form.
● Color codes and alerts are examples of visual cues that show the severity of a
disease.

Strategy for Deployment:

● The app comes with a TFLite model for offline use.
● Cloud services for improving datasets and recording feedback.
● For simpler testing and updates, modular architecture was used.

3.2.4 Outcome of Objectives / Result Analysis

Key performance indicators like model accuracy, prediction speed, and
user experience scores were used to keep a close eye on the goals for Sprint
II.

Findings:

● On the validation set, the model trained successfully with 92% accuracy.
● The conversion to TFLite has been finished and verified on several devices.
● During inference, Flutter integration is seamless with little lag.
● Users expressed satisfaction with the UI improvements.
● Dataset preparation is recorded for sharing and reuse in the future.

The findings from this project reinforce the effectiveness of deep learning
techniques, particularly Convolutional Neural Networks (CNNs), in image-based
disease classification tasks.

27
Evaluation:

● The CNN-based method worked well for classifying leaf diseases.

● The file size was decreased by the quantized TFLite model without sacrificing
accuracy.
● Preprocessing steps for images increased the overall consistency of predictions.
● Offline use was made possible by the edge inference technique, which is perfect
for rural settings.

Important Takeaways:

● Dataset diversity has a major impact on model generalization.

● Real-world deployment requires ongoing feedback loops and validation.
● Adoption by non-technical users is greatly influenced by UI design.
● Careful memory and lifecycle management are necessary for Flutter integration.

3.2.5 Sprint Retrospective

The team was able to consider their AI development process and integration
results thanks to the Sprint II retrospective. Overall, the sprint was productive, with
several key accomplishments. The project planning was effective, allowing for a
smooth transition between tasks such as data collection, preprocessing, model
training, and evaluation. The dataset was prepared earlier than expected, which
provided sufficient time for experimenting with different models. The
Convolutional Neural Network (CNN) and transfer learning models such as ResNet
performed well, achieving satisfactory accuracy levels. Team coordination and
communication were consistent, which contributed significantly to the project's
steady progress.

28
What Worked Well:

● Training and optimizing the ML model was finished in the allotted time.
● smooth interaction between the Flutter frontend and machine learning.
● Implementing UI feedback raised app user engagement.
● During the stakeholder review, the offline model capability was valued.

What Could Be Made Better:

● Due to its imbalance, the original dataset needed to be manually enhanced.

● Limitations were revealed through testing under various lighting conditions.
● More clarity is required in the documentation of the model lifecycle and
retraining procedures.
● Small problems with image upload speed were encountered during real-world
testing.

Items of Action:

● Add more real-world photos from farmers to the dataset.

● For scalability, automate data augmentation pipelines.
● Introduce a pipeline for retraining the model based on user feedback.
● Boost the responsiveness of the user interface on older mobile devices.

The project was successfully transitioned from foundational development to

a working AI-based application that was prepared for end-user testing during Sprint
II. In addition to laying the groundwork for user onboarding in the upcoming sprint,
it demonstrated that running AI locally on mobile devices is feasible.

29
 CHAPTER 4

RESULTS AND DISCUSSIONS

The Tomato Leaf Disease Detection project is thoroughly evaluated in this
section by looking at performance indicators, model accuracy, testing procedures,
comparisons with other systems, and the usefulness of the application in actual
situations. Model efficiency, app usability, and feedback from early users were used
to evaluate Sprint II's results. To confirm the reliability of the deployed machine
learning model and its integration with the mobile application, the project
underwent extensive testing in both simulated and field settings.

4.1 Project Outcomes (Performance Evaluation, Comparisons, Testing

Results)

Performance Evaluation: Thousands of labeled images from a carefully selected

dataset were used to train the CNN model created for tomato leaf disease detection.
On the validation dataset, the final trained model's accuracy was roughly 92%. To
guarantee effective deployment on mobile devices, the TensorFlow-based model
was subsequently transformed into TFLite format. Through optimizations like
quantization, this transformation was able to reduce the model size to less than 10
MB while preserving a high level of accuracy and cutting down on inference time.

Even on low-end Android devices, the average inference time was 1.2 seconds.
Evaluation metrics like precision, recall, and F1-score were used to compare the
model's predictions. The F1-score stayed above 0.88 for several disease categories,
such as Septoria, Early Blight, and Leaf Mold. The model's resilience in real-world
applications was demonstrated by the fact that its performance held steady even
when photos were taken in different lighting conditions or with slight blur.

Testing Methodology: A multi-tiered testing procedure was applied to the

application. Individual parts like the camera module, model loader, and prediction
display user interface were validated through unit testing Fig 4.1 . The smooth
30
connection between image capture, prediction logic, and result output was
guaranteed by

Fig 4.1 Training vs validation Accuracy

Fig 4.2 Training vs Validation Loss

In machine learning projects, testing and validation Fig 4.2 are crucial steps to
evaluate the performance, accuracy, and generalization ability of the model. For
this tomato leaf disease prediction system, a Convolutional Neural Network
(CNN) model was trained to classify various diseases based on leaf images. To
ensure the model is reliable and performs well on unseen data, proper testing and
validation techniques were implemented (Kumar & Patel, 2022). Integration
testing was conducted using numerous mobile devices for field testing in a
variety of lighting and environmental settings. By identifying edge cases, this
testing assisted in optimizing the preprocessing pipeline. Students, agricultural
researchers, and farmers were among the stakeholders involved in User
Acceptance Testing (UAT). Their input was crucial in determining the need for
more features and UI enhancements.

31
Comparisons: The performance of the suggested model was contrasted with that
of industry-standard models such as MobileNetV2 and ResNet-50 in order to assess
its efficacy. Despite achieving a marginally higher accuracy of 94%, ResNet-50 was
not suitable for deployment on mobile platforms due to its large model size (95 MB)
and high memory requirements. Compared to the custom CNN, MobileNetV2 had
a slower inference time but offered moderate performance.

The optimal balance between accuracy, speed, and size was provided by the
custom CNN. Because it was designed for edge inference scenarios, predictions
could be made without an internet connection—a critical feature for users in rural
areas. Our model was perfect for mobile-first deployment, as demonstrated by the
comparative table that was made to summarize model performance:

User Acceptance and Feedback: Users engaged with the app to identify diseases
from test images during a pilot test. Approximately thirty users took part, including
Agri-students and farmers. More than 85% of users thought the app was useful and
simple to use. Users valued the quick results, easy-to-understand interpretation of
the results, and simple treatment recommendations. Requests for features like voice
output, multilingual support, and history logs for examining previous predictions
were among the feedback gathered. To be implemented in future sprints, these items
were logged.

In conclusion, the disease detection model's successful implementation and

assessment attest to its preparedness for broader implementation. The application
strikes a balance between usability and technical complexity so that even non-
technical users can take advantage of state-of-the-art AI. In addition to achieving
its initial goals, the project creates opportunities for future iterations that could scale
and adapt for other crops.

32
Testing Results:

Fig 4.3 Testing Results

Any machine learning project must include testing to make sure the model
operates accurately and consistently on unknown data. The objective of tomato leaf
disease prediction is to accurately categorise photographs of tomato leaves into
distinct disease groups (e.g., Healthy, Late Blight, and Early Blight).

A tomato leaf disease prediction model's testing phase Fig 4.3 is essential for
assessing the model's functionality and suitability for practical use. Testing
evaluates the model's capacity for generalisation using unseen data, usually distinct
from the training dataset.

33
 CHAPTER 5
CONCLUSION AND FUTURE ENHANCEMENT:

5.1 Conclusion
An important development at the nexus of artificial intelligence and
agriculture is the Tomato Leaf Disease Detection system. This solution gives
farmers, agronomists, and other agricultural stakeholders quick, easy, and
trustworthy insights into crop health by using machine learning to analyze leaf
photos and identify plant diseases.

The system facilitates efficient crop management decision-making by

improving early disease detection through its user-friendly image-based interface,
expert-backed recommendations, and extensive disease knowledge base. This
results in lower crop loss and higher yields for farmers, better tracking of regional
disease trends for agronomists, and scalable tools for system administrators to
support and improve continuously.

In the end, the project helps to bridge the gap between traditional farming and
technology, fostering more data-driven and sustainable agricultural practices and
opening the door to more intelligent food production systems.

5.2 Future Enhancement:

1. Extend the model to identify diseases in additional crops, such as cucumbers,

peppers, and potatoes. Enables a larger agricultural audience to be served by the
system.

2. Video-Based Real-Time Disease Detection Use drones or mobile phone cameras

to conduct live video analysis. Helps keep an eye on bigger fields without having
to examine each plant by hand.

3. Edge AI in Offline ModeUse ONNX or TensorFlow Lite to incorporate

34
 the model into offline mobile applications. Farmers in isolated or poorly
connected areas will find it especially useful.

4. Environmental Diagnostics and Pest Detection Include the ability to identify pest
damage or stress brought on by nutrient shortages, poor soil, or drought. Provides
a diagnosis of plant health that is more comprehensive.

[Link] and Voice Support Provide voice assistance and local language
options for users who are illiterate or only partially literate. Improves user
engagement and accessibility.

35
REFERENCES

[1] Kumar, R., & Patel, A. (2022). Evaluation of deep learning techniques for plant
disease detection. International Journal of Computer Applications, 184(10), 25–30.

[2] Ferentinos, K. P. (2018). Deep learning models for plant disease detection and
diagnosis. Computers and Electronics in Agriculture, 145, 311–318.
[Link]

[3] Jha, K., Doshi, A., Patel, P., & Shah, M. (2019). A comprehensive review on
automation in agriculture using artificial intelligence. Artificial Intelligence in
Agriculture, 2, 1–12. [Link]

[4] Choudhary, G., Khandelwal, S., & Budhlakoti, R. (2021). Deployment of deep
learning models for plant disease detection in real-time environments. Procedia Computer
Science, 192, 2504–2513. [Link]

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

[6] Purushothaman, R., Ramesh, A., and Rangarajan, A. K. (2018). Classifying Tomato
Crop Diseases with a Pre-trained Deep Learning Algorithm.1040–1047 in Procedia
Computer Science, 133.

[7] Moussaoui, A., Boukhalfa, K., & Brahimi, M. (2017). Visualization of Symptoms and
Classification of Tomato Diseases Using Deep Learning.31(4), 299–315; Applied
Artificial Intelligence.

[8] Islam, M. M., and Hasan, S. M. (2020). Convolutional Neural Networks for the
Identification of Tomato Leaf Disease.9(3), 1005–1011, International Journal of
Engineering and Advanced Technology.
36
APPENDIX A
CODING

37
38
39
 APPENDIX B

PUBLICATION DETAILS

We submitted our research paper for publication at icici publication organized by

S.E.A College of Engineering and Technology
Bangalore, Karnataka, India
Proof of publication is attached in figure B The research

Figure B : The research

40
 APPENDIX C
PLAGIARISM REPORT

41