Mask Wearing Detection Algorithm Based on Improved YOLOv4 for Complex Environments

In a complex environment with dense crowds, there are a large number of objects with small scale and blocked by background information, so the mask wearing detection algorithm has poor effect. In order to improve the detection effect of the model in complex environment, an improved mask wearing detection algorithm of YOLOv4-Mask was proposed. Firstly, SE attention mechanism was introduced in YOLOv4 backbone network to suppress background interference and improve feature extraction ability of small-scale targets. Secondly, the shallow features and deep features of the backbone network are fused to realize the complementary advantages between different levels of features and enrich the details of deep features. Then an improved Feature Enhancement Module (FEM) is proposed to enhance the feature expression capability of shallow features. Finally, a four-scale head structure is designed to make full use of the shallow features and reduce the missing rate of small-scale targets. Experimental results show that mAP of the YOLOv4-Mask algorithm increases by 8.15% on the self-built data set, and mAP increases by 5.81% and 7.38% on the AIZOO and MAFA test sets, respectively, which can meet the requirement of accuracy of mask wearing detection task in complex environment.