Garbage Detection using Advanced Object Detection Techniques

Developing nations today face a major hurdle of excessive waste generation due to overpopulation and rapid urbanization. Also, the waste management systems in such countries are ineffective and limited. Considering this issue, an effective and efficient waste management system would be of great societal benefit. Artificial Intelligence and Deep Learning has found its way into many diverse areas in recent years. This research work proposes a Garbage Detection System using object detection models to automatically detect and locate garbage in real-world images as well as video. The work comprises of a detailed review of previous research and proposes new method with different algorithms to detect garbage. Five different models used in this paper are EfficientDet-D1, SSD ResNet-50 V1, Faster R-CNN ResNet-101 V1, CenterNet ResNet-101 V1 and YOLOv5M. After hyper-parameter tuning and evaluation, YOLOv5M achieved the best results for the proposed system by achieving a Mean Average Precision (mAP@0.5) value of 0.613. This system directly engages citizens to join a national movement to help the authorities to maintain a clean and green environment.