%0 Journal Article %A ZHANG Luda %A DENG Chao %T Multi-scale Fusion of YOLOv3 Crowd Mask Wearing Detection Method %D 2021 %R 10.3778/j.issn.1002-8331.2103-0505 %J Computer Engineering and Applications %P 283-290 %V 57 %N 16 %X

Under the requirements of the COVID-19 epidemic prevention and control, wearing masks under crowd environment in public places such as shopping malls and stations has become a necessary condition for people to travel. Because there are often crowded personnel in the crowd environment, easy to block each other, and the target size is small, it is easy to misdiagnose and miss the inspection of mask wearing. To solve these problems, based on the YOLOv3 algorithm, a crowd mask wearing detection algorithm based on improved YOLOv3 is proposed. First, add a shallow feature map, increase the shallow detection scale to the original 3-scale detection structure to form a 4-scale detection structure, and improve the detection accuracy. Then, the top-down and bottom-up multi-scale fusion structures are introduced to further use the feature information to achieve feature enhancement. Finally, the CIoU loss function is used for frame regression to improve the positioning accuracy. Experimental results show that the average accuracy of the improved YOLOV3 algorithm reaches 93.66%, which is 5.61 percentage points higher than the original YOLOV3 algorithm. Compared with other mainstream algorithms, this algorithm has higher detection accuracy in the mask wearing detection task and has good practicability.

%U http://cea.ceaj.org/EN/10.3778/j.issn.1002-8331.2103-0505