Stacking Binary Neural Networks for Efficient Edge-Based Waste Classification

This project focuses on developing an efficient edge-based solution for waste classification using advanced deep learning techniques. The dataset includes various waste categories, including paper, plastic, glass, cans, cardboard, plastic bottles, and plastic bags. The objective is to create a system capable of accurately classifying and detecting these waste materials to support waste management efforts and recycling processes. To achieve high classification performance, a combination of YOLOv26, RT-DETR, Grad-CAM, and Soft-NMS is employed. YOLOv26 is used for real-time object detection, RT-DETR for robust and efficient object localization, while Grad-CAM provides visual explanations for model predictions, and Soft-NMS refines object detection results to minimize false positives. The model's performance is evaluated using precision, recall, and F1-score metrics. A user-friendly web application is developed to allow users to upload images of waste materials for real-time classification. This solution aids in automating waste sorting, contributing to more efficient recycling processes and promoting sustainable waste management practices.

Keywords: Waste Classification, YOLOv26, RT-DETR, Grad-CAM, Soft-NMS, Deep Learning, Edge-Based Systems, Object Detection, Recycling, Waste Management, Sustainable Practices.