Lighter and Faster Face Mask Detection Method Based on YOLOv5

At present, the face mask detection model based on deep learning is more complex and difficult to deploy on edge devices with limited computing resources. In this paper, a lighter and faster face mask recognition method based on YOLOv5 is proposed. ShuffleNet and GhostNet modules are introduced respectively in the backbone and neck sections to reduce parameters. An attention mechanism is added after each feature fusion part to make it pay more attention to the important information on the feature map. Experimental results shows that the proposed algorithm achieves a higher 1.6% mAP compared to YOLOv5 while the model size is reduced by 89.4%, which is easier to deploy.