Intelligent Monitoring Method of Helmet Wearing Identification in Power Plants
Xingke Li, Yikai Wang, Jiaxin Qu, Wei Wang, Xingming Xu, Shisong Jin, Yunhao Wu, Yingjian Li · 2023
To handle the problems of low recognition accuracy and slow detection speed of the current safety helmet wearing detection in power plant management, an improved safety helmet wearing detection algorithm based on YOLOv5 is proposed. Firstly, based on the idea of the weighted bidirectional feature pyramid network BiFPN, the weighted BiFPN network structure is developed. The FPN + PAN feature fusion structure in the original neck part of YOLOv5 is modified to the weighted BiFPN network structure, which reduces the feature information lost in convolution operation and improves the accuracy of safety helmet detection. Secondly, the CBAM attention mechanism is integrated into the YOLOv5 network to obtain relatively fine safety helmet feature data points. Finally, by using Kmeans + + clustering algorithm, the selected anchor box is closer to the optimal solution. Experiments show that the mAP of the improved algorithm is 90.44 %, which is 6.15 % higher than that of the original algorithm by using the frozen and non-frozen training method under the self-built experimental data in this paper.