Mask wearing Detection Based on Improved YOLOv3

Mask-wearing detection in public places is a very important and challenging task under the situation of normalized epidemic prevention and control. There is high error rate of mask- wearing detection by use of traditional YOLOv3 obviously. In this paper, an improved YOLOv3 object detection method was proposed. In order to improve the method’s insufficient ability to detect small objects, the following improvements were made in the paper. The number of anchor boxes was increased and the activation function was replaced to improve the accuracy. The Focal Loss was used to improve the model convergence speed. In this paper, a dataset with a total of 2707 images was built by taking pictures from smartphones and acquiring relevant pictures on the Internet. The results of the comparative experiment show that the mAP (mean average precision) of the proposed is improved 1.33% than others.