A Framework for Segmenting and Classification of Plastic Waste using Deep Networks

This paper discusses the development of an advanced deep learning framework for combating the problem of segregating plastic waste. The proposed approach aims to accurately detect and classify various plastic waste materials, overcoming challenges like occlusion, damage, cuts, etc… The main approach used for classification is based on an improved version of the YOLO algorithm alongside the Detectron2 package, using a dataset of 2,000 annotated images. The experimental results showed an excellent mean Average Precision (mAP) of 89.54%, outperforming traditional YOLO models. These results showed the power of combining deep networks with robust object detection for both segmentation and classification in addressing plastic waste management challenges and promoting sustainable practices.