Research on YOLOV5 Electric Workers Wear Detection Model Based on Channel Pruning and Attention Fusion
Chengcheng Zhu, Miao Gong, Wang Luo, Xinsheng Chen, Xiaofa Zhou, Yuying Shao, Boyang Sun, Xiaolong Hao · 2022
At present, target detection based on deep learning has become a trend. The large model in target detection has high detection accuracy, but with the huge network depth and width, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, Firstly, we build a feature extraction network based on yolov5, and the CBAM (Convolutional Block Attention Module) attention structure are used to improve the detection accuracy. Finally, we force iterative channel-level pruning to the detection network to guide model Sparse training of BatchNormalization (BN) layers. The results show that the proposed method can improve the accuracy of target detection, While keeping a high average detection accuracy, the calculation amount of the YOLOv5m model is reduced by 39%, the inference speed on GPU 40% increase, which meets the requirements of real-time detection.