Application of YOLOv11 Classification for Efficient Waste Segmentation in Australia's Recycling Facilities

This paper investigates the use of YOLOvll, an advanced deep learning model, for waste classification in Aus-tralian recycling facilities, aiming to improve sorting accuracy and reduce contamination in line with Australia's National Waste Policy. Utilizing the RealWaste dataset, which contains over 10,000 annotated images of waste materials collected at the Whyte's Gully Waste and Resource Recovery facility in Wollongong, NSW, this study demonstrates YOLOvll's superior performance across various metrics. The YOLOvll variants achieved top-l classification accuracies ranging from 82.39% to 85.95% and top-5 accuracies from 99.37% to 99.79<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>. The highest-performing model, YOLOvllm-cls, achieved a top-l accuracy of 85.95% and an inference time of 0.5545 ms, balancing classification precision with computational efficiency. These results significantly improve upon prior models used in similar tasks, showcasing YOLOvll ‘s potential to automate waste classification, reduce human error, and enhance recycling operations.