Plant Diseases Prediction Using Machine Learning

Submitted in partial fulfillment of the requirements of the degree

BACHELOR OF ENGINEERING IN COMPUTER

ENGINEERING

By

Malhar Sanjay Kulkarni

Sahil Vinayak Shinde

Viraj Dattatraya Patil

Vedant Vilas Shinde

Soham Viraj Wani

Sr. No. Content

1. Introduction

2. Objective

3. System Architecture

4. Tools and Technologies

5. Dataset Description

6. Result and Disccusion

7. Conclusion
 CHAPTER 1: Introduction
Agriculture is still the base of the economy of the world, supporting billions of people by
providing food and jobs for a large portion of the population. Agricultural productivity is
frequently threatened by diseases of plants which can, at worst, result in total yield loss, a
decreased quality of crops, and significant economic harm for farmers. The traditional
approaches to detecting diseases include field inspections or analyzing the plants in a
laboratory. These approaches often take significant time to identify a range of plant diseases,
they depend on skilled labor knowledge, and are not scalable solutions when operating at
huge numbers for farming operations.

Due to the availability of new technologies, therefore, Artificial Intelligence (AI) and
Computer Vision are powerful tools utilized in precision agriculture and and can help
automate highly important steps such as identifying plant disease. AI and Computer Vision
can provide a valuable tool that is not only efficient but less expensive and reliable. Along
with AI, Deep Learning, especially Convolutional Neural Networks (CNNs), perform very
well on image-based classification tasks including plant disease detection tasks.
 CHAPTER 2: Objective
Automate Plant Disease Detection

Develop a deep learning-based model capable of accurately classifying bell pepper leaf
images into healthy or diseased categories.

Integrate Real-Time Image Monitoring

Establish a continuous monitoring setup using cameras to capture plant images at scheduled
intervals for real-time analysis.

Build a User-Friendly Web Application

Create a web interface (using Streamlit or similar frameworks) where users can upload plant
images and receive instant prediction results along with suggested treatments.

Ensure Reliable Data Transfer and Storage

Implement a secure and robust mechanism for transferring captured images from the
Raspberry Pi (or edge device) to the main server for processing and logging.

Provide Actionable Insights to Farmers

Deliver clear, easy-to-understand prediction outcomes that help farmers take preventive or
corrective actions to protect their crops.

Enable Error Handling and System Reliability

Design the system architecture to include error handling, retries, and fallback options to
ensure reliable operation under real-world conditions.

Promote Cost-Effective and Scalable Deployment

Ensure the solution is practical for small and medium-sized farms by using affordable
hardware and open-source tools, with scope for scaling up.
 CHAPTER 3: System Architecture

1. Image Acquisition

- Start image capture using OpenCV from cameras.

- If image capture fails, retry capturing.

- Ensures reliable acquisition under varying conditions.

2. Data Transfer and Storage

- Store captured images on Raspberry Pi for initial buffering.

- If storage fails, retry storing until successful.

- Transfer images to central computer/server for processing.

- If transmission fails, retry data transmission.

- Server receives the image for processing.

- Run the trained deep learning model (.h5) for disease prediction.

- If prediction fails, retry the prediction process.

- Ensures robust and reliable prediction before result delivery.

4. Result Delivery

- Display prediction results to the user via a web application (Streamlit).

- Suggest possible treatment or actions based on disease detected.

- Complete the process and optionally log data for history tracking.
 CHAPTER 4: Tools and Technologies
 Hardware Tools

The hardware components form the foundation of the image acquisition and monitoring
setup.

1. Raspberry Pi

Used as an edge device to interface with cameras, capture images periodically, and handle
initial data buffering before transfer to the main server.

2. IP Camera / USB Camera

Employed for continuous image capture of bell pepper plants in greenhouse or open-field
environments. These cameras are controlled and managed through the Raspberry Pi.

3. Computer Server / Local Machine

Serves as the processing unit where the trained deep learning model is deployed. It handles
image preprocessing, prediction, and hosts the web application.

 Software Tools and Frameworks

Software frameworks play a critical role in the model development, backend processing, and
frontend deployment.

1. Python

The core programming language used for developing the AI model, integrating system
components, and building the web application.

2. TensorFlow / Keras

Deep learning frameworks utilized for designing, training, and saving the Convolutional
Neural Network (CNN) model used for disease detection.

3. OpenCV

Open-source computer vision library used for image capture from cameras and preprocessing
tasks like resizing and normalization.

4. Streamlit

Python-based web framework that enables rapid development of the web interface for users
to upload images and view prediction results.
5. NumPy and Pandas

Libraries for numerical computations and data management used during preprocessing and
analysis phases.

 Model Development and Training Tools

1. Jupyter Notebook

Interactive environment for model experimentation, visualization of training metrics, and

2. LabelImg / CVAT (Computer Vision Annotation Tool)

Image annotation tools used for labeling the dataset with appropriate disease categories
required for supervised model training.

3. Matplotlib / Seaborn

Data visualization libraries used to plot training metrics like loss and accuracy during model
evaluation.
 CHAPTER 5: Dataset Description
arjuntejaswi/plant-village

Currently we have a limited variety of data it consist of 2475 of total images, which are
divided into two different classes

1. Healthy

2. Bacterial_spot

[Link]
 CHAPTER 6: Results & Discussion
The Plant Disease Monitoring and Prediction System was designed to automate the detection
of diseases in bell pepper plants using a deep learning-based classification model integrated
with a real-time monitoring setup. Upon completing the development and training phases, the
system was evaluated for both its prediction accuracy and operational reliability in a
simulated field environment.

The model, trained on a well-annotated dataset of bell pepper leaf images categorized as
healthy or bacterial spot-infected, showed promising results during the training and validation
stages. After multiple iterations of model tuning, the final trained model achieved a training
accuracy of approximately 98%, with validation accuracy stabilizing around 95%. The testing
phase, which involved images unseen by the model, yielded a test accuracy of around 94%,
demonstrating the model's ability to generalize its predictions on new data. Data
augmentation techniques, such as image rotation, flipping, and zooming, played a crucial role
in preventing overfitting and enhancing the model's robustness against minor variations in
image input.

An analysis of the confusion matrix revealed that the model consistently identified healthy
leaves and accurately classified diseased leaves showing distinct symptoms of bacterial spots.
However, a few misclassifications were observed, primarily in cases where the disease
symptoms were visually subtle or when the images had inconsistent lighting conditions.
These observations highlight the importance of dataset diversity and the potential benefit of
including a wider range of environmental scenarios in future data collection efforts.

The integration of the trained model with a Streamlit-based web application provided a
seamless user experience. Users could easily upload images of bell pepper leaves and receive
prediction results instantly, accompanied by a confidence score indicating the model's
certainty. The simplicity of the web interface ensured accessibility for users without technical
backgrounds, while the backend reliably handled image preprocessing and model inference.
This integration confirmed the system's usability in practical applications, particularly in
agricultural monitoring scenarios where timely disease detection is critical.

From a system architecture perspective, reliability and error handling mechanisms were
tested thoroughly. The Raspberry Pi, acting as an edge device, successfully captured images
and stored them locally before transferring them to the processing server. In cases of
transmission failures or storage issues, retry mechanisms ensured that data was not lost,
maintaining the system’s reliability. Similarly, the prediction module was designed to handle
unexpected errors gracefully, preventing system crashes and ensuring continuous operation.
These features are vital for deploying such a system in real-world farming environments
where hardware and network stability cannot always be guaranteed.

The overall deployment and testing were conducted on a local server setup, with the web
application accessible via a browser interface. The system was tested with a variety of images
collected under different lighting and environmental conditions to evaluate its real-world
applicability. The results consistently aligned with the model's training outcomes, reaffirming
the system's capability to assist in precision agriculture by providing accurate disease
predictions and early alerts.

In conclusion, the system met its core objectives by successfully integrating AI-based disease
prediction with real-time monitoring and user-friendly result delivery. While the current
version focuses on two disease categories, it sets a strong foundation for further expansion,
including the detection of multiple diseases and integration with centralized databases for
historical analysis. The project validates the practical potential of deep learning applications
in agriculture, specifically in enhancing crop health monitoring and supporting data-driven
decision-making for farmers.
 CHAPTER 7: Conclusion
The Plant Disease Monitoring and Prediction System showcased how artificial intelligence
and deep learning can be utilized for agricultural applications in terms of early disease
detection. By combining a trained CNN model with real time monitored image capture and
an easy to use web application, the system was able to provide accurate predictions and
actionable insights for farmers. The automated image capture, data reliability, and AI-based
analysis demonstrated the existing practical potential for this system in precision farming.

The project successfully accomplished its key objectives of automating disease detection, and
providing time-stamped results. The project also demonstrated a pathway for future
developments and enhancements, including expanding disease categories, increasing
prediction accuracy, and distributing through cloud platforms to engage and add
additional users.