This project is a part of my Deputy Coordinator tenure from November, 2023 to May, 2024 in Sahaay Social Innovation Club at IIT Madras.
The objective is to design an image classification system using pre-trained MobileNet architecture which is capable of automatically sorting waste items into six distinct classes based on their material type: glass, metal, paper, wood, plastic and mixed. The 'mixed' class also holds other types of waste. The ultimate goal is to integrate this into a physical waste bin system in an effort to improve recycling efficiency and waste management.
https://github.com/RishiNandha/AWS_Dataset
Used a dataset of 5327 images with a train/val/test split of 70/15/15
Train Accuracy: 0.9447572827339172
Validation Accuracy: 0.9884687662124634
Validation F1 Score (macro): 0.9838134895955094
Test Accuracy: 0.84
Test F1 Score (macro): 0.8339493856287788
A good balance between high accuracy and high F1 score was obtained.