CAPSTONE – PROJECT (CSE339)

REPORT ON

Plant Disease Detection Using Deep Learning and

Classification.
Submitted by

S. No Name Registration No
1 Pusuluri Mohan Sai 12104818
Swaroop
2 Bandla Ramakoteswara 12107105
Rao
3 Surya Prabhas Sirasapalli 12111700
4 Komati Neni Karthik 12101090

Under the Guidance of

Mr. Khalid Hafiz Mir - 30571

Associate Professor

School of Computer Science and Engineering

Lovely Professional University. Phagwara
(September- November - 2024)

1
1.1 ABSTRACT:

By 2050, the world population is likely to increase to 10 billion. This growth will
drive competition in developing agriculture on a more efficient basis to better the
situation with food security. Crop yield and quality losses due to plant diseases demand
developing highly sophisticated detection systems. This research explores applying an
already trained CNN named EfficientNetV2S in plant disease detection and optimizes
hyperparameters to achieve higher performance. The model successfully copes with
low-resolution images, elaborate textures, mixed backgrounds, and luminance
complications. For enhancing robustness, a variant of the Plant Diseases Dataset was
used with 38 classes, contaminated as well as diverse samples of infected and healthy
crop leaves. By training the model on noisy data intentionally, the adaptability of the
network and its generalization capabilities are enhanced for strong performance in real
conditions. In this regard, dataset engineering in combination with strong training
facilitates better plant disease surveillance, providing a sustainable avenue for
augmenting agricultural productivity and food security.

1.2 DESCRIPTION OF THE INVENTION:

In this project, our team members are going to predict the different diseases that
occurs in Plant disease detection by using deep learning approaches. This present
invention is based on a leading-edge system for detection and classification of plant
diseases using deep learning techniques. The system involves identification of critical
needs in the area of efficient and accurate plant disease detection that majorly affects
agricultural productivity and food security.

A vital component of the design is a pre-trained Convolutional Neural Network (CNN),

such as the EfficientNetV2S that has been optimized for the task of plant disease
classification. It has been designed so as to function in real-world agricultural conditions
by solving issues caused by low-resolution images, distracting and complex
backgrounds, varying leaf textures, shadows, and lighting fluctuations.

The system is trained using an improved version of the Plant Diseases Dataset, which
accommodates several classes of healthy and diseased plants. To make the model more
robust and adaptable, noisy as well as diverse samples are added to the dataset,
simulating practical conditions, so that the model performs reasonably well on a broad
range of inputs and, more importantly, in adverse scenarios.

1.2.0 INTRODUCTION OF THE PROJECT:

Ensuring food security will be one of the major challenges by 2050 due to the rise in
the population estimated to be about 10 billion. Agricultural productivity improvement
is, therefore, required to enhance the country's food supply levels. Among the major
barriers to crop yield and quality is plant diseases. Detecting these diseases early and

2
accurately is essential for effective crop management and for sustainable agriculture.
Methods for disease identification traditionally have been time-consuming, subjective,
and prone to error, particularly for large-scale farming operations.

The rising power behind deep learning techniques, notably Convolutional Neural
Networks (CNNs), have been capable of automatic plant disease detection. These
models, however, fail to consider factors such as low resolutions, cluttered
backgrounds, and environmental variations- which are the realities in practical
application.

This study addresses these limitations by using a pre-trained model, EfficientNetV2S,

which is optimized for plant disease detection. By altering the Plant Diseases Dataset
to contain noisy and diverse samples, the model is trained to generalize better under
challenging conditions. Improved performance in the handling of low-quality images
and shadows and varying textures is achieved through hyperparameter tuning.

This proposed system captures robust dataset engineering blended with sophisticated
classification models, scalable, reliable, and efficient solution towards plant disease
detection. It is, therefore, part of the giant steps forward in deep learning's leverage for
agricultural productivity and food security in real-world settings.

1.2.1 OBJECTIVE OF THE INVENTION:

1. By this project, the CNN models are going to predict different diseases of Plant.
2. This Project will be going to help farmers to detect the diseases.
3. The algorithm that we will be going to use gives good accuracy of disease
detection, great performance and takes less processing time to
predict the diseases.

1.2.2 SCOPE OF THE PROJECT:

The scope of this project is wide and significant, as it deals directly with the critical
challenges facing agriculture by employing advanced plant disease detection and
classification. The system would be highly scalable, efficient, and reliable to address a
variety of cases in divergent agricultural settings, ensuring better crop management and
productivity. Significant areas within the scope of this project are:
Disease Detection and Classification:
The system is developed to identify and classify numerous plant diseases in a large
number of crops and hence can be easily adjusted for different areas of agriculture and
farming practices.
Robustness in Real World Conditions:
The system is trained on noisy datasets and heavily diversified by the project so that the
performance in real-world conditions like low-resolution images, complex background,
varying textures, shadows, and luminance changes will be robust.
The system can be scaled across different crops and diseases.
The developed pre-trained CNN model can be applied to detect diseases in a variety of
crop types. Such flexibility can easily adapt to other crops and agricultural settings.

3
Improving Food Security:
Early detection of plant disease leads to fewer crop losses and improves yield quality,
directly addressing global issues with food security.
Application to Precision Agriculture:
This will be easily integrated into existing precision agriculture tools such as drones or
smartphone applications to give real-time monitoring and disease detection in large
farms.
2. DETAILED DESCRIPTION OF THE PROJECT:

This project shall develop an advanced system for plant disease detection and
classification, using the latest deep learning techniques. This initiative shall help address
critical challenges in agriculture: early disease detection, precision monitoring, and
sustainable crop management, to mitigate the ever-increasing global food security
crisis.

Objective:

The aim of this project is to design a robust and efficient system that can accurately
detect and classify plant diseases in real-world conditions. This includes noisy low-
resolution images, different textures, shadows, and busy or distracting backgrounds.

Main Features:

Architecture with EfficientNetV2S

This system is derived from the pre-trained model EfficientNetV2S architecture of
Convolutional Neural Network (CNN). This model was in particular selected due to its
balance between efficiency and precision, making it perfect for real-time applications
in agriculture.

Dataset Engineering:

A modified version of the Plant Diseases Dataset is utilized, representing 38 classes of

both healthy and diseased crop leaves.
The dataset consists of augmented samples to simulate real-world conditions such as
low-quality images, noise, and varying lighting, to ensure the model's adaptability.
This approach ensures that the model generalizes efficiently, even when exposed to
challenging inputs.

Hyperparameter Optimization:

The key parameters, such as learning rate, batch size, dropout rates, and activation
functions, are tuned in the project.
The optimization improves the model's effectiveness in diagnosis of plant diseases with
high accuracy and reliability.

Robust Training Strategy:

4
The model is trained using a noisy and augmented dataset, intentionally exposing it to
imperfect data.
This strategy improves the model's robustness, enabling it to handle diverse and
complex field conditions effectively.

Performance Metrics:

The system is evaluated based on accuracy, precision, recall, and F1 score to ensure its
effectiveness across various plant species and disease types.
A focus is placed on minimizing false positives and false negatives to enhance practical
usability.

Applications:

 Early Disease Detection and Monitoring

The system issues early warnings about diseased plants, enabling timely correction by
the farmer.

 Precision Farming
Very easily integrates with high-technology precision farming tools, such as drones or
smartphone applications, to monitor large farms.

 Scalability Across Crops and Regions

Can be scaled up to detect diseases in various crops and can be customized to the
specific needs of agriculture globally.

 Low-Cost Farming
It reduces the dependence on manual inspection and expert judgment, thereby giving it
to the farmers at all levels.

Impact and Significance:

 Global Food Security: By saving crops from losses, the project enhances the
agricultural productivity, and hence, it directly intervenes in the problem of
feeding the ever-growing population.
 Sustainability: It promotes ecologically safe agriculture by saving the crops
from excessive usage of pesticides through targeted interventions.
 Advancement in AI It shows the deep learning ability to be powerful in solving
real-world problems, especially in resource-constrained environments.

5
 2.1 FLOW CHART OF THE MODEL TRAINING AND
DESIGNING:

DATA FLOW DIAGRAM:

6
USE CASE DIAGRAM:

ADVANTAGES OF THIS PROJECT:

This project provides a number of several significant advantages, that mark it as a

transforming solution to plant disease detection and classification. These benefits get
derived from its novel application of deep learning techniques, robust dataset
engineering, and practical applicability in real-world agricultural scenarios.
1. High Accuracy in Disease Detection
 Using the EfficientNetV2S architecture and optimizing the hyperparameters,
the system has achieved high precision to detect and classify plant diseases.
 Less errors; not false positives, and false negatives. It should give accurate
results.
2. Robustness in Challenging Conditions

 The system is developed to work correctly under real-world conditions of

agriculture, including:
 Low-resolution images.
 Complex and distracting backgrounds.
 Variations in lighting, shadows, and textures.
 Therefore, the robustness ensures this flexibility between different
environments and field situations.
3. Improved Generalization Capabilities
 The model achieves very good generalization to new and unseen data by training
on augmented and noisy Plant Diseases Dataset, thus improving usability in
diverse farming scenarios.
4. Scalability and Versatility

7
  The system is scalable and adaptable to a wide variety of plant species and
disease types.
 It supports classification across 38 classes of healthy and diseased plants, thus
it is applicable globally for different crops and agricultural practices.
5. Cost-Effective Solution
 It will reduce dependency on manual inspections and expert evaluation for the
detection of plant diseases.
 The use of this system is easy through affordable tools such as smartphones or
drones, which is common across small-scale and large-scale setups.
6. Impact on Food Security

 Detection of diseases through early and precise means prevent drastic losses in
crops, thereby enhancing yield and quality of food.
 This system directly contributes to world efforts toward food security for an
increasingly expanding population.
7. Pesticide Use Reduction

 The system identifies specific diseased plants, which helps the farmer target
treatments and avoid excessive pesticides on the farm.