Deep Learning Based Waste Detection: Evaluating the Evolution of YOLO Architectures for Smart Waste Management

With the increasing global emphasis on sustainable waste management practice, deep learning-based object detection methods have become pivotal in automating waste classification processes. This study presents a comparative analysis of the YOLOv4, YOLOv5, YOLOv7 and YOLOv8 models for waste classification, focusing on key performance indicators such as precision, recall, specificity, and mAP (Mean Average Precision). The result indicates that YOLOv8 shows superior performance with an accuracy of 99.0 %, followed closely by YOLOv7 in speed and accuracy. The study also explores the incremental learning capability of YOLO models, allowing the classifier to incorporate new waste categories-such as e-waste-without catastrophic forgetting of previously learned classes. This highlights the potential of YOLO architectures for the object detection task in the multiclass scenario. The findings support the integration of advanced object detection models into smart waste sorting and recycling systems, contributing to more efficient environmental sustainability practices.