Detection and Classification of Waste Materials Using Deep Learning Techniques

This paper deals with the detection and classification of waste materials by performing an analysis of the deep learning-based computer vision techniques. Object detection and classification algorithms use machine learning to produce meaningful results. Thus, in real-life following this approach we can get benefits of enhanced performance in waste identification and increase in recycled materials. We review existing research and solutions for waste object detection,focusing on deep-learning models and frameworks employed in similar object detection problems. Subsequently,we describe the creation and composition of our data sets,which include fourteen waste categories encompassing various items commonly found in waste streams such as glass waste,chip packets and plastic bags. We then discuss the object detection models that were utilized to classify and detect waste materials within our data sets. Specifically,we employ SSD MobileNet V2 FPN Lite, EfficientDet-D0, YOLOv7,and YOLOv8 models to achieve accurate waste identification and classification. Finally, we present and analyze the results obtained by these models. Our evaluation indicates that YOLO v8 achieves the highest accuracy,with a confidence level of 0.5 and a Mean Average Precision (mAP) of 76.6%.