🌾 Rice Classification with CNN and Transfer Learning
This project classifies different types of rice grains using two deep learning approaches:
rice_image.py: A custom CNN model built from scratch (no transfer learning)rice_transfer.py: A model using Transfer Learning with ResNet50
The application uses Streamlit for easy interactive deployment.
📦 Requirements
Install the required dependencies using:
pip install -r requirements.txtOr manually install:
pip install tensorflow numpy opencv-python scikit-learn matplotlib streamlit
1️⃣ rice_image.py - CNN from Scratch
This script builds a deep Convolutional Neural Network without using any pre-trained models.
✅ Features
- 5 convolutional layers and 5 pooling layers
- Dropout for regularization
- Data normalization and label encoding
- Metrics:
accuracy, precision, recall
- Saves model as
rice_model.h5
- Streamlit app for image upload and prediction
- Accuracy & loss training plots
▶️ Run the app
streamlit run rice_image.py
2️⃣ rice_transfer.py - Transfer Learning with ResNet50
This script uses ResNet50 with pre-trained ImageNet weights and a custom classification head.
✅ Features
- Pretrained ResNet50 model (
include_top=False)
- Custom classification head with Dense + Dropout layers
- Freezing base layers for feature extraction
- Data augmentation with
ImageDataGenerator
- Metrics:
accuracy, precision, recall
- Saves model as
rice_resnet_model.h5
- Streamlit app for image upload and prediction
- Accuracy & loss training plots
▶️ Run the app
streamlit run rice_transfer.py
🧪 Example Output
After training, both apps provide:
- Training accuracy & loss plots
- File uploader to classify new rice images
- Predicted class shown alongside the uploaded image
📚 Dataset Information
Classes:
- Arborio
- Basmati
- Ipsala
- Jasmine
- Karacadag
Each class contains 200 images, totaling 1000 labeled rice grain images.
💾 Model Files
| Model | File Name | Type |
|---|
| CNN from Scratch | rice_model.h5 | Keras model |
| ResNet50 Model | rice_resnet_model.h5 | Keras model |
| Label Encoder | label_encoder.pkl | Pickle file |
