Automated e-Waste classification: Real-time object detection using YOLO model

This study addresses the critical challenge of e-waste management by developing an automated classification system using YOLO-based object detection models. A custom-annotated dataset of 10,000 images across 20 e-waste categories was utilized to enhance model generalization and performance. The results demonstrated significant improvements in detection accuracy, with YOLOv11s achieving a precision of 93.1%, recall of 86.6%, mAP@0.5 of 93.2%, and mAP@50-95 of 86.4%. Lightweight models such as YOLOv11n have shown promise for real-time use, reaching an inference speed of 1.9 ms while retaining competitive accuracy. The system’s real-time detection capabilities were showcased through a user-friendly Streamlit application supporting image uploads, live webcam feeds, and model retraining via an integrated MLOps pipeline. Despite these achievements, challenges related to small and overlapping object detection, dataset preparation, and computational constraints were identified, highlighting areas for refinement. Recommendations for future work include dataset expansion, incorporating transformer-based architectures, instance segmentation, and IoT integration for real-world deployments. This research presents a scalable, efficient solution to automated e-waste sorting, reinforcing the role of deep learning in promoting sustainable waste management practices.