MobileNet-YOLOv2 Mask Wearing Detection Method Fused With Attention Mechanism

In view of the problem of slow detection speed and low accuracy in manual detection and existing mask wearing detection algorithms in special public places, this paper proposes a MobileNet-YOLOv2 mask wearing detection method fused with attention mechanism. We have adjusted the backbone network structure of the original YOLOv2 algorithm, replacing the backbone network Darknet-19 with MobileNet, and using deep separable convolution for feature extraction, greatly reducing the number of model parameters and computational complexity, and improving detection speed. Introducing the convolutional attention mechanism module CBAM enables the feature extraction network to focus more on key image regions and improve detection accuracy. Collect and establish 850 facial mask wearing images to train and test the proposed algorithm model. The experimental result shows that the average detection accuracy of wearing a mask reaches $90 \%$, while the detection accuracy of not wearing a mask reaches $80 \%$. Compared to the DarknetYOLOv2 and SSD algorithms, the detection speed has been improved by 6.3 and 6.5 times, and the model size has been reduced by $77.5 \%, 89.1 \%$, and $89.2 \%$, respectively, compared to Darknet-YOLOv2, R-CNN, and Faster-RCNN.