A Method of Mask-Wearing Detection Based on Improved YOLOv4

Abstract

In this paper, a lightweight network based on improved YOLOv4 is proposed to solve the problems of complex model structure and unsatisfactory performance of detection speed in fast-moving situations. Firstly, the inverted residual blocks (IRB) based on Mobilenetv2 are adopted into the backbone feature extraction network to reduce the complexity of the model structure. Then, the feature fusion network based on the depth-wise separable convolution is applied to minimize model calculations and parameters. Experimental results show that the proposed model has the advantages of high accuracy, fast detection speed, and lightweight, which has satisfied the requirements of real-time detection of mask-wearing in fast-moving situations.