A mask-wearing detection algorithm based on improved YOLOv7
Zil Ding, Jianxin Guo, Jiashan Liu, Hangjie Zhu · 2023
Aiming at the problem that the accuracy of mask-wearing detection is poor and it is difficult to deploy in low-cost equipment, this paper has proposed a mask-wearing detection algorithm based on improved YOLOv7. By adding an attention mechanism to the backbone network and introducing the partial convolution method, the algorithm enhances the perception ability of the network to small targets, improves the accuracy of mask-wearing detection, reduces the parameters and calculation of the network, and improves the efficiency of feature extraction of the network. By integrating Distribution Shift Convolution in the head network, the required memory and equipment requirements are reduced while the operation speed is also improved. The experimental results show that, compared to the original model, the mean average precision of the improved YOLOv7 network improved by 0.5%, the parameters are reduced by 20%, the computation only accounts for 51% while the model size is lower to 63%and the reasoning speed remains basically unchanged. These improvements can be applied in the current mask-wearing monitoring task and provide a useful reference for future object detection tasks.