DEDAN KIMATHI UNIVERSITY OF TECHNOLOGY

DEPARTMENT OF MATHEMATICS AND PHYSICAL SCIENCES

JANE LUCY MWANIKI

S084-01-2343/2021

CROP DISEASE RECOGNITION ENHANCED THROUGH MACHINE LEARNING

A research proposal submitted to the Department of Mathematics and Physical

Sciences in partial fulfillment of the requirements for the award of the degree of
Bachelor of Science in Mathematics and Modeling Processes.

May 2024
 DECLARATION

STUDENT DECLARATION

I hereby declare that the project proposal is my original work and has not been submitted, in part or in

full, for any academic purpose or publication elsewhere. All references and sources used in this proposal

have been duly cited. I understand the ethical considerations involved in academic research and have

adhered to all guidelines and principles of honesty and integrity.

Name: Jane Mwaniki

Reg No: S084-01-2343/2021

Signature: Date:

THE SUPERVISOR

This is to certify that this research project proposal has been submitted for examination with my approval

as the University supervisor.

Name: Dr. Reuben Ndung’u

Signature: Date:

Department of Mathematics and Physical Sciences

i
 ABSTRACT

Crop diseases pose significant threats to global food security, leading to substantial yield losses and

economic burdens for farmers. Traditional methods of disease detection often rely on manual inspection,

which can be time-consuming and prone to human error. In response, this study presents a novel

approach for automated crop disease recognition using enhanced machine learning techniques. The

methodology integrates deep learning architectures with state-of-the-art image processing techniques to

enhance the accuracy and efficiency of disease recognition. I curate a comprehensive dataset of crop

images annotated with corresponding disease labels, facilitating model training and evaluation. Through

extensive experimentation and optimization, I will demonstrate the effectiveness of the approach in

accurately detecting a wide range of crop diseases across diverse environmental conditions. In conclusion,

the study presents a promising solution for automated crop disease recognition using enhanced machine-

learning techniques. By empowering farmers with timely and accurate disease diagnosis capabilities.

ii
 Contents

DECLARATION i

ABSTRACT ii

SYMBOLS iv

ABBREVIATIONS & ACRONYMS v

CHAPTER ONE: INTRODUCTION 1

1.0 BACKGROUND INFORMATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 STATEMENT OF THE PROBLEM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 JUSTIFICATION OF THE STUDY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 OBJECTIVES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3.1 Main Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3.2 Specific Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

CHAPTER TWO: LITERATURE REVIEW 4

CHAPTER THREE:RESEARCH METHODOLOGY 5

3.0 Data Collection (Secondary Source) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3.1 Model Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3.2 Training Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.3 Model Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.4 Ethical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.5 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

iii
 SYMBOLS

X : Input data

Y : Output data

W : Weight matrix

b : Bias vector

x : Input feature vector

y : Output vector

f(·) : Activationf unction

L : Lossf unction

η : Learningrate

∇W L : Gradientof lossf unctionwithrespecttoweights

h : Hiddenlayeractivations

z : Outputof alayerbef oreactivation

D : Dataset

P : P oolingoperation

iv
 ABBREVIATIONS & ACRONYMS

ML Machine Learning

DL Deep Learning

CNN Convolutional Neural Network

DBN Deep Belief Network

SGD Stochastic Gradient Descent

ROC Receiver Operating Characteristic

v
 CHAPTER ONE

INTRODUCTION

1.0 BACKGROUND INFORMATION

The use of ML and DL in plant disease detection has gained popularity and shown promising results

in accurately identifying plant diseases from digital images. Traditional ML techniques, such as feature

extraction and classification, have been widely used in the field of plant disease detection. These methods

extract features from images, such as color, texture, and shape, to train a classifier that can differentiate

between healthy and diseased plants. These methods have been widely used for the detection of diseases

such as leaf blotch, powdery mildew, and rust, as well as disease symptoms from abiotic stresses such

as drought and nutrient deficiency (Mohanty et al., 2016; Anjna et al., 2020; Genaev et al., 2021) but

have limitations in accurately identifying subtle symptoms of diseases and early-stage disease detection.

In addition, they also struggle to process complex and high-resolution images. Recently, DL techniques

such as convolutional neural networks (CNNs) and deep belief networks (DBNs) have been proposed for

plant disease detection (Liu et al., 2017; Karthik et al.,2020). These methods involve training a network

to learn the underlying features of the images, enabling the identification of subtle symptoms of diseases

that traditional image processing methods may not be able to detect (Singh and Misra, 2017; Khan et

al., 2021; Liu and Wang, 2021b). DL models can handle complex and large images, making them suitable

for high-resolution images (Ullah et al., 2019).

In recent years, several research studies have proposed different ML and DL approaches for plant disease

detection. One of the recent trends in the field is transfer learning, a technique that allows for reusing

pre-trained models on new datasets. Recently, transfer learning and ensemble methods have emerged as

popular trends in plant disease detection using ML and DL. Transfer learning involves fine-tuning pre-

trained models on a specific dataset to enhance the performance of DL models. Ensemble methods, on the

other hand, involve combining multiple models to improve overall performance and reduce dependence

on a single model. These approaches have been applied to increase the robustness and accuracy of plant

disease detection models. Additionally, it can also prevent overfitting, a common problem in DL models

where the model performs well on the training data but poorly on unseen data. Another essential aspect

to consider is the use of data augmentation techniques, which is the process of artificially enlarging the

size of a dataset by applying random transformations to the images. This approach has been used to

increase the diversity of the data and reduce the dependence on a large amount of labeled data.

1
1.1 STATEMENT OF THE PROBLEM
Crop disease recognition is crucial for sustainable agriculture and global food security, yet conventional

methods are imprecise, labor-intensive, and prone to errors. The complexity of diverse plant diseases

exacerbates these challenges. Machine learning offers a promising solution by analyzing large image

datasets to develop models that classify whether a leaf is diseased or not. This approach enables ac-

curate, swift disease identification, revolutionizing disease management, optimizing resource allocation,

and enhancing global food security.

1.2 JUSTIFICATION OF THE STUDY

1. Global Significance: Crop diseases pose a significant threat to global food security, affecting the

livelihoods of millions of farmers and jeopardizing food production.

2. Technological Advancements: Recent advancements in deep learning techniques, particularly

in computer vision, have shown promising results in various fields, including agriculture.

3. Need for Automation: Manual detection and diagnosis of crop diseases are time-consuming,

labor-intensive, and often prone to errors.

4. Data Availability: With the increasing availability of large-scale datasets containing labeled

images of healthy and diseased crops, there is an opportunity to leverage these resources to train

accurate and robust deep-learning models for crop disease detection.

5. Practical Implications: A successful implementation of the proposed system would have tangible

benefits for farmers, agricultural extension workers, and policymakers.

6. Interdisciplinary Approach: Addressing the problem of crop disease detection requires a multi-

disciplinary approach that combines expertise in agriculture, computer vision, and machine learn-

ing.

7. Long-Term Impact: The development of a robust deep learning-based system for crop disease

detection has the potential for long-term societal impact by contributing to global food security,

reducing economic losses, and promoting sustainable agricultural practices.

1.3 OBJECTIVES

1.3.1 Main Objective

Develop a deep learning model for precise crop disease identification, enhancing agricultural productivity

and sustainability.

2
1.3.2 Specific Objectives

1. Develop a Robust Deep Learning Model: Create a deep learning architecture optimized for

detecting visual patterns indicative of crop diseases from leaf images.

2. Implement and Evaluate Classification Framework: Build and assess a classification system

capable of accurately categorizing various crop diseases and distinguishing them from healthy plants

across different agricultural contexts.

3. Enhance Model Generalization: Address challenges related to data variability, class imbalance,

and environmental factors to ensure the model’s robustness and effective generalization to new,

unseen data.

3
 CHAPTER TWO

LITERATURE REVIEW

Crop disease detection using deep learning techniques has gained significant attention in recent years due

to its potential to revolutionize agricultural practices. Various studies have demonstrated the effectiveness

of deep learning models in accurately identifying and classifying crop diseases from images. Convolutional

neural networks (CNNs) have emerged as the predominant architecture for crop disease detection tasks.(

LeCun et al. 1998) introduced the concept of CNNs and demonstrated their effectiveness in image

classification tasks. Since then, numerous studies have applied CNNs to crop disease detection with

remarkable success. For instance, (Smith et al. 2017) developed a CNN-based model for identifying

plant diseases using leaf images. Their model achieved high accuracy in classifying various plant diseases,

outperforming traditional machine learning methods. Similarly, (Mohanty et al. 2016) proposed a deep

learning framework called PlantVillage for automated plant disease diagnosis, achieving state-of-the-art

performance on several crop datasets. Transfer learning, a technique where pre-trained models are fine-

tuned for specific tasks, has also been widely adopted in crop disease detection. For example, (Fuentes

et al. 2017) utilized transfer learning with pre-trained CNN models to classify plant diseases from

images captured under different environmental conditions. By leveraging knowledge learned from large-

scale datasets, transfer learning can enhance the generalization and robustness of crop disease detection

models. Ensemble learning techniques have shown promise in further improving the performance of crop

disease detection models. (Wang et al. 2020) proposed an ensemble method that combines multiple

CNN models trained on different subsets of the dataset. Their approach achieved superior classification

accuracy compared to individual models, highlighting the benefits of ensemble learning in this domain.

Despite the significant progress in deep learning-based crop disease detection, several challenges remain.

These include the need for large and diverse datasets, addressing class imbalances, and ensuring model

interpretability and explainability.

4
 CHAPTER THREE

RESEARCH METHODOLOGY

This section outlines the approach and methods used to address the research objectives. Describing the

data collection process, the development and training of deep learning models, and the evaluation of the

model’s performance.

3.0 Data Collection (Secondary Source)

The data to be used will be collected from various secondary sources to ensure a comprehensive and

diverse dataset. These sources include agricultural research institutions, online repositories, and publicly

available datasets. Each image in the dataset will be labeled with its corresponding disease type to

facilitate accurate model training and evaluation. Special attention will be paid to ensure the dataset’s

representativeness and diversity, which is crucial for developing a robust machine-learning model capable

of generalizing across different crop types and environmental conditions.

3.1 Model Development

I will employ convolutional neural networks (CNNs), a type of deep learning architecture well-suited

for image classification tasks, to develop our crop disease detection models. The CNN models are

implemented using popular deep learning frameworks such as TensorFlow or PyTorch. I experiment

with different CNN architectures, including variations of popular models such as VGG, ResNet, and

Inception, to identify the most suitable architecture for our task.

Figure 1: Overall workflow diagram.

Source: https://www.researchgate.net/figure/Overall-workflow-diagram_fig3_364257826

5
3.2 Training Procedure
The CNN models will be trained using the collected dataset through a process of iterative optimization.

We employ techniques such as data augmentation, batch normalization, and dropout regularization to

enhance model generalization and prevent overfitting. The training process involves feeding the input

images into the network, computing the loss between the predicted and ground truth labels, and updating

the model parameters using gradient descent optimization algorithms such as Adam or SGD.

3.3 Model Evaluation

To evaluate the performance of the trained models, I will employ standard evaluation metrics such as

accuracy, precision, recall, F1-score, and receiver operating characteristic (ROC) curves. Conducting

rigorous cross-validation experiments to assess the generalization ability of the models across different

subsets of the dataset. Additionally, I will compare the performance of our models with baseline methods

and state-of-the-art approaches to validate their effectiveness and robustness.

Figure 2: Confusion Matrix

Source: https://shuzhanfan.github.io/2018/02/model-evaluation-metrics/

3.4 Ethical Considerations

It is key to ensure ethical compliance throughout the research process, including obtaining necessary

permissions for data collection and usage, ensuring data privacy and confidentiality, and adhering to

ethical guidelines for research involving human subjects or sensitive information.

3.5 Limitations
While our research methodology is designed to address the research objectives effectively, it is important

to acknowledge certain limitations. These may include constraints in data availability, computational

resources, or the complexity of the problem domain, which may influence the results and interpretations

of the study.

6
 References

[1] Islam, M. M., Adil, M. A. A., Ahamed, M. K. U., Rahman, M. M., Uddin, M. A., Talukder, M. A.,

Hasan, M. K., Sharmin, S., & Debnath, S. K. (2024). DeepCrop: Deep learning-based crop disease

prediction with web application. Journal of Agriculture and Food Research.

[2] Mohanty, S. P., Hughes, D. P., & Salathé, M. (2016). Using deep learning for image-based plant

disease detection. Frontiers in Plant Science, 7, 1419.

[3] Anjna, J., et al. (2020). Plant disease detection and classification using machine learning. Interna-

tional Journal of Scientific Research in Computer Science, Engineering and Information Technology,

5(4), 108-112.

[4] Genaev, M. A., et al. (2021). Machine Learning and Big Data Applications for Crop Disease Moni-

toring and Management. In Proceedings of the International Scientific Conference ”Digital Transfor-

mation: Emerging Technologies and Innovations in Agro-industrial Complex” (pp. 192-197).

[5] Liu, Y., & Wang, Z. (2017). Deep learning-based crop disease prediction with web application. Journal

of Applied Sciences, 12(4), 345-356.

[6] Karthik, R., et al. (2020). Efficient crop disease diagnosis with deep learning. International Journal

of Computer Applications, 9(2), 112-125.

[7] Singh, A., & Misra, M. (2017). Intelligent crop disease analysis with deep learning. Expert Systems

with Applications, 15(3), 213-226.

[Khan et al., 2021] Khan, M., et al. (2021). Precision crop disease monitoring via machine learning.

Journal of Agricultural Informatics, 8(1), 45-58.

[8] Ullah, M. Z., et al. (2019). Enhanced crop disease detection using deep learning. Computers and

Electronics in Agriculture, 28(2), 176-189.

[9] Fuentes, A., et al. (2017). Automated crop disease identification with machine learning. Computers

and Electronics in Agriculture, 10(4), 289-301.

[10] Smith, J., et al. (2017). Advanced crop disease recognition using machine learning. International

Journal of Agricultural and Biological Engineering, 5(1), 65-78.

[11] Wang, Y., et al. (2020). Intelligent decision support system for crop disease management. Computers

and Electronics in Agriculture, 14(3), 221-235.