Garbage Detection Algorithm Based on YOLO v3
Bohong Liu, Xinpeng Wang
Abstract
In recent years, garbage classification has become the focus of the world. YOLOv3 has good real-time performance and accurate detection accuracy, which can meet the requirements of garbage sorting equipment. In this paper, we introduce a lightweight cross-channel interaction Attention mechanism in the residual unit of YOLOv3, and propose a new target detection algorithm, namely Efficient Channel Attention YOLO (ECA-YOLO) detection algorithm. So the target detection algorithm can get more complete and more effective feature information. Experimental results show that the improved model performs better than YOLOv3 in the garbage image data set of “Huawei Garbage Classification Challenge Cup”. The improved model has a high accuracy and recall rate, and the detection mAP is improved by 1.07%, while the detection speed is kept unchanged and the number of parameters is introduced.