Face Mask Recognition Based on Improved YOLOv7-Tiny

Abstract

A face mask recognition algorithm based on the improved YOLOv7-Tiny is proposed to address the current problems such as time-consuming and poor real-time performance of manual checking whether a mask is worn. Firstly, the Backbone of YOLOv7-Tiny is replaced by the MobileNetV3 network as a whole, making the structure more lightweight. Secondly, the edge loss function uses EIOU to improve the localization accuracy of face mask edges. Finally, the CBAM attention mechanism is added to improve the model detection performance. Experiments were conducted on the AIZOO dataset, and the improved YOLOv7-Tiny increased the mAP from 93.9% to 94.2% compared to the original algorithm, with a 28.3% decrease in the number of parameters and a 37.2% decrease in inference time. The experimental results show that the improved model is not only able to reduce the model size, but also improve the accuracy and speed of face mask detection, showing good mask recognition results.