Ironside35/Rice-Classifier-CustomCNN

🍚 Custom CNN Rice Classification System

This is the 5th application in my deep learning portfolio. It features a from-scratch Convolutional Neural Network (CNN) designed specifically to differentiate between 5 premium rice varieties.

🛠️ Technical Specifications

• —Architecture: 5-Layer Custom CNN (Conv2D -> MaxPooling -> Dropout).
• —Dataset: 75,000 images (Murat Koklu Rice Image Dataset).
• —Optimization: Adam optimizer with Categorical Crossentropy loss.
• —Input Size: Optimized at 100x100 pixels for fast processing.

📈 Performance Results

• —Training Accuracy: %98.86.
• —Validation Accuracy: %99.00.
• —Epochs: 5 (Early convergence achieved due to massive data scale).

🔬 Observation

By building the feature extraction layers from scratch, the model learned specific geometric patterns of rice grains, resulting in near-perfect classification even without pre-trained weights.