Overview This project develops a deep learning model to classify rice grain images into five varieties using transfer learning with the ResNet-18 architecture. Implemented in Python and MATLAB, the model achieves 100% accuracy on a 1,100-image subset and 99.8% accuracy on the full 75,000-image dataset, demonstrating robust performance and efficient training. Technologies Used
Programming Languages: Python, MATLAB Libraries: PyTorch, MATLAB Deep Learning Toolbox Tools: Kaggle (dataset hosting), Git, Jupyter Notebook, MATLAB Editor
Methodology
Data Preprocessing: Normalized images and applied data augmentation (e.g., rotation, flipping) to improve model generalization. Model Architecture: Employed ResNet-18 with transfer learning, fine-tuning the fully connected layers for rice grain classification. Training: Conducted two-stage training: Stage 1: Trained on a 1,100-image subset with a batch size of 32 and Adam optimizer. Stage 2: Scaled to the full 75,000-image dataset for enhanced performance.
Evaluation: Measured classification accuracy and training time, optimizing for efficiency.
Results
Accuracy (Subset): 100% on 1,100 images Accuracy (Full Dataset): 99.8% on 75,000 images Training Time: 384.33 seconds for the subset, scalable to larger datasets Key Achievement: High accuracy with optimized computational efficiency
Installation
Clone the repository: git clone https://github.com/TNZRalf/Rice-Grain-Classification-Using-Deep-Learning.git
Install Python 3.8 or later and required libraries: pip install torch torchvision matplotlib numpy
(Optional) Install MATLAB R2023a or later for MATLAB-based scripts.
Download the rice grain dataset from Kaggle or Google Drive and extract it to the data/ folder.
Usage Python Implementation
Navigate to the project directory: cd Rice-Grain-Classification
Open rice_classification.py or rice_classification.ipynb in your preferred editor (e.g., VS Code, Jupyter).
Update the dataset path in the script to point to data/.
Run the script: python rice_classification.py
Outputs include accuracy metrics, confusion matrix, and sample classification visualizations.
MATLAB Implementation (Optional)
Open MATLAB and set the working directory to the project folder.
Open rice_classification.m.
Ensure the dataset is in the data/ folder.
Run the script: run rice_classification.m
Outputs include accuracy metrics and visualization plots.
Project Structure
rice_classification.py: Main Python script for training and evaluating the model. rice_classification.ipynb: Jupyter Notebook with a detailed workflow. rice_classification.m: MATLAB script for alternative implementation. data/: Placeholder directory for the rice grain dataset (not included in the repository). README.md: Project documentation.
Dataset The dataset contains 75,000 rice grain images across five varieties. Due to its size, it is not included in the repository. Download it from:
Kaggle (recommended) Google Drive (alternative, if uploaded)
Place the dataset in the data/ folder and update the script paths accordingly. Future Improvements
Explore alternative architectures like EfficientNet or Vision Transformers for improved accuracy. Develop a web application for real-time rice grain classification. Optimize training for lower-resource environments using model pruning or quantization.
Contact For questions or collaboration, contact Zakaria Tanani at zakaria.tanani12@gmail.com. License This project is licensed under the MIT License. See the LICENSE file for details.