Mask Recognition Based on Improved YOLOv5 Target Detection Algorithm

Since the outbreak of the new crown epidemic, in order to curb the spread of the epidemic, the World Health Organization and others have carried out a series of work studies. At present, wearing a mask correctly when traveling in any public place is one of the effective means of protection. However, for example, in places with high traffic such as high-speed railway stations and airports, the efficiency of manual detection and supervision of mask wearing is low. Therefore, it is very necessary to use the automatic detection mask device to supervise the wearing of masks in real time. With the deepening of research on deep learning network models, in the real-time detection of masks Deep Learning-based network models, it is difficult to achieve satisfactory results in terms of precise and real-time in all performance. In order to improve the detection accuracy and other issues, based on the YOLOv5 object detecting algorithm, we first created a dataset MaskData for mask recognition by using the open source dataset downloaded from the Internet and adding various types of face mask datasets. Secondly, in the detection network, the DIOU_nms method is designed to be a replacement for the IOU in the NMS, and under the same parameters, the detection accuracy of occluded and overlapping targets is improved. Finally, replacing the GIOU loss function with the a-CIoU loss function can obtain higher-quality localized image regions faster and more accurately, generate bounding boxes and predict categories. The results of the experiment have demonstrated that the refined network can more accurately identify whether the face is wearing a mask, and to some extent, the detection accuracy was greatly improved. And using the designed GUI interface, the trained and improved YOLOv5 model can be directly called to perform real-time mask-wearing detection for videos and pictures.