AI Waste Sorting

How It Works

Waste mismanagement is a global crisis. This project aims to simplify recycling by using Artificial Intelligence to instantly classify waste items via a smartphone camera.

The system captures an image, processes it through a custom-trained neural network, and returns a segmented overlay indicating whether the object should go in the recyclying bin or the trash bin.

System Architecture

1. Model Training: The brain of this application is a YOLOv11m-Segmentation model by Ultralytics. It was trained on the TACO (Trash Annotations in Context) Dataset using a NVIDIA A100 GPU via Google Colab. The training process involved:

• Data Cleaning & Augmentation
• 300 Epochs of rigorous training
• Inclusion of "Negative Samples" (empty backgrounds) to reduce false positives.

2. The Backend: The trained model is hosted on Hugging Face Spaces using a custom Docker container running Python and FastAPI. This allows for high-performance inference without draining the user's phone battery.

3. The Frontend: The website you are seeing is hosted on Cloudflare Pages and built using HTML. It communicates securely with the backend API to send images and receive segmentation masks in real-time.