Coffee Disease Classifier

Table of Contents

• Coffee Disease Classifier
  • Table of Contents
  • Overview
  • Features
  • Technologies Used
  • Dataset
    • Dataset Structure
  • Model Architecture
    • Model Performance
  • Installation
    • Prerequisites
    • Steps
  • Usage
  • Future Work
  • License
  • Contact

Overview

The Coffee Disease Classifier is a mobile application designed to identify and classify diseases in coffee plants using deep learning. The app leverages a Convolutional Neural Network (CNN) model, trained on a dataset of plant images, to detect and diagnose various coffee diseases from images taken by the user.

Features

• Disease Detection: Identifies and classifies common coffee diseases.
• Real-Time Analysis: Users can take a photo or upload an image of a coffee plant for immediate analysis.
• User-Friendly Interface: Built with Flutter for a smooth and intuitive user experience.
• Offline Capability: The model can perform inference without an internet connection, making it suitable for use in the field.

Technologies Used

• TensorFlow: For building and training the CNN model.
• Keras (tf.keras): A high-level API for TensorFlow, used to simplify model development.
• Flutter: For creating the cross-platform mobile application.
• Dart: The programming language used with Flutter.

Dataset

The model is trained and tested using the Plant Disease Classification - Merged Dataset available on Kaggle. The dataset contains images of various plant diseases, including those affecting coffee plants.

Dataset Structure

• Number of Classes: 39 (includes various plant diseases and healthy plants). But in this project, we only use 4 classes: Coffee__cercospora_leaf_spot, Coffee__healthy, Coffee__red_spider_mite, Coffee__rust.
• Number of Images: Over 87,000 images. But we only use 1103 files belonging to the 4 classes.
• Image Size: 256x256 pixels

Model Architecture

The CNN model was developed using the following architecture:

• Input Layer: Accepts 256x256x3 images.
• Convolutional Layers: Multiple layers with ReLU activation and max-pooling.
• Fully Connected Layers: Dense layers leading to the output classification.
• Output Layer: Softmax activation for multi-class classification.

Model Performance

• Accuracy: Achieved an accuracy of approximately $75.98$ on the test set.
• Loss: Final loss value was $0.6019$ on the test set.

Installation

Prerequisites

• Flutter SDK
• Dart
• TensorFlow and Keras
• Python 3.x

Steps

1. Clone the repository:

   git clone https://github.com/TheMetaSetter/coffee-disease-classifier.git
   cd coffee-disease-classifier

2. Install the required Python packages:

   pip install -r requirements.txt

3. Install Flutter dependencies:

   flutter pub get

4. Run the application:

   flutter run

Usage

1. Open the app on your mobile device.
2. Capture or upload an image of a coffee plant leaf.
3. The app will analyze the image and display the predicted disease and confidence score.

Future Work

• Model Optimization: Improve the model's accuracy and reduce its size for faster inference on mobile devices.
• Additional Disease Classes: Expand the dataset to include more coffee-specific diseases.
• Multi-Language Support: Add support for multiple languages to increase accessibility.
• Cloud Integration: Enable cloud-based model updates and data collection for continuous learning.

License

This project is licensed under the Apache-2.0 license - see the LICENSE file for details.

Contact

For any questions or suggestions, please contact:

• Khoi Nguyen - nguyenanhkhoi0608@gmail.com
• GitHub: TheMetaSetter