Multi Object Detection and Classification in Solid Waste Management using Region Proposal Network and YOLO model

Rapid urbanization has given us many benefits in terms of giving us a better standard of life, but it sure has brought a lot of problems with it. Solid waste management is a major issue that has been in the forefront of the issues caused by urbanization. It has brought a lot of domains together namely social, environmental, and climate together. Technology has been used occasionally but has not made significant advances in this domain. This domain is relatively new in terms of deep learning. This is due to some of the issues like lack of proper dataset, effective architecture to classify multiple objects and so on. The goal is to build multi-class dataset in this domain and perform detection and classification using both single stage and two stage object detection networks. The single stage network that is to be implemented is YOLOv5 and the two stage network that is to be implemented is Faster Region based CNN using Resnet50. The single object dataset used is TrashNet dataset and the multi-object dataset used is Waste-mart self-built dataset. The result obtained shows mAP around 0.84 for the two-stage network and mAP around 0.98 with IoU threshold placed at 0.5 for both the systems.