randyver/trash-classification-cnn

Model Overview

This model classifies images from the TrashNet dataset into one of six categories: Cardboard, Glass, Metal, Paper, Plastic, and Trash. It uses a convolutional neural network (CNN) architecture for image classification tasks, specifically aimed at waste management and recycling systems.

• Dataset: TrashNet (6 waste categories)
• Architecture: CNN with 3 convolutional layers, max pooling, and fully connected layers.
• Evaluation Metrics: Accuracy, Precision, Recall, F1-Score
• Repository: https://github.com/randyver/trash-classification.git

Use Cases

• Direct Use: Classify waste images for recycling or waste management systems.
• Downstream Use: Can be integrated into smart recycling and waste sorting ecosystems.
• Limitations: Not suitable for fine-grained classification or tasks outside of waste classification (e.g., medical, security).

Training Details

• Data: Preprocessed TrashNet dataset (images resized and normalized).
• Hyperparameters: Learning rate: 0.001, Batch size: 32, Epochs: 10.
• Model Architecture: Standard CNN with convolutional layers followed by max pooling and fully connected layers.

Recommendations

• Retraining: Retrain the model if expanding to new waste categories or environments.
• Image Quality: Ensure high-quality images for optimal performance.