Lightweight YOLOv4 Coal Gangue Detection Method Based on Embedded Platform

Abstract:
A lightweight YOLOv4 coal gangue real-time detection method based on embedded platform is proposed for the problem that the current traditional gangue separation method cannot balance detection efficiency, reliability and safety. YOLOv4 consists of a backbone network, neck structure, and YOLO head. Firstly, a new Ghost-SE module is proposed. The SE module does a good job of weighting the network channels for better features. The Ghost module that can generate more feature maps with fewer parameters to improve the network’s learning ability. The two structures are integrated into the backbone network to further improve the detection performance. Then, the backbone network is redesigned to reduce the width and depth of the backbone network. In addition, the channel pruning method is used to prune the model and further compress the model. Compared with the original YOLOv4 experimental results, the number of parameters and the amount of calculation were reduced by 91.25 % and 87.73 %. The mAP(IoU=50:95) reaches 60.4%.