Deep learning has significantly advanced automated waste management. Current solutions lack a generalized object detection model and compact mechanical sorting system, which has obstructed the implementation of automated waste sorting systems in domestic environments. To address this, this paper presents a compact, edge computing smart waste bin integrated with a novel dual-axis actuation mechanism. By curating a diverse dataset, optimizing the learning mechanism, and using a closed-loop actuation mechanism for precise movement, the waste bin can sort domestic waste into four main categories: biodegradable, plastic, metal, and hazardous. The real-time deep learning model is powered by a lightweight YOLOv11n neural network running on a Raspberry Pi 5, achieving a mean average precision at 0.5 threshold (mAP@0.5) of 90.17%. Furthermore, this system incorporates a strict confidence threshold to prevent cross-contamination and high-purity recycling. This research suggests a way to implement an automated waste-recycling system that supports sustainable urban infrastructure and the global circular economy.
Siyad et al. (Fri,) studied this question.