Research On Mask Wearing Detection Algorithm Based On Deep Learning

Since the outbreak of COVID-19 in 2020,the virus has caused serious harm to the lives and health of people in China and around the world.With the global spread of the epidemic,China will gradually enter the normalized management of epidemic prevention and control by the end of 2022.With the gradual increase in population flow across the country,the epidemic still poses a certain threat to human life and health.The virus is mainly transmitted through droplets,so it is still necessary to supervise the wearing of masks among people in public places,especially in densely populated places such as supermarkets,stations,and hospitals.The use of manual supervision in public places is not only time-consuming and labor-intensive,but also increases the probability of virus transmission.Therefore,using artificial intelligence mask wearing detection methods to replace manual detection has become a new trend.At present,traditional neural networks still have certain shortcomings in the reliability and recognition rate of mask wearing detection in complex scenes(such as blurry images,facial occlusion,etc.),with errors and missed detections,and the detection accuracy needs to be improved.In order to further improve the recognition rate of face mask recognition algorithm,this paper designs a replacement loss function for the scene of detecting mask wearing in crowded places,and improves YOOv5 mask wearing detection algorithm(YOOv5s-S-CBAM-BiFPN)based on CBAM attention mechanism and BiFPN feature fusion idea.The main work of this article is as follows:Firstly,construct a mask wearing detection dataset in complex scenarios and conduct preliminary testing on the data using YOLOv5.Establish a complex scene mask dataset,which includes situations where the mask wearer is obstructed and the target to be detected is blurry.The image content reaches 9240 pieces,meeting the algorithm’s requirements for mask wearing detection in complex scenes.The images are annotated in detail based on actual detection tasks.Preliminary analysis through experiments shows that the recognition accuracy of YOLOv5 for mask wearing in complex scenes needs to be improved.Secondly,based on the above background,this article has made three improvements to YOLOv5:1.In view of the slow convergence speed of the model in the training process,replace the original CIOU loss function,and use SIOU as the regression loss function to make the regression of the target box more stable,so that the convergence speed of the model is accelerated and the accuracy of reasoning is improved.2.In response to the issue of interference noise affecting the selection of target boxes in complex backgrounds,this article introduces the CBAM attention mechanism into YOLOv5 to enhance the network’s anti-interference ability,making the extraction of facial key information more concentrated in mask wearing detection,and improving the detection rate of targets;In order to improve the utilization of original image feature points and reduce the loss of feature information in sampling,the feature fusion method from BiFPN is introduced into the Neck module,and two horizontal cross scale connection paths are added to enhance information transmission between different network layers,enriching the feature information of the target.Finally,three effective improvements were fused,and experimental results showed that the improved YOLOv5 s S-CBAM BiFPN algorithm improved accuracy by 4.0%and recall by 5.6% compared to the original YOLOv5 algorithm on a self-made dataset,m AP@.5 Increased by 4.7%,m AP@.5.95 has increased by 2.8% and has good detection performance,meeting the requirements for mask wearing detection in complex scenarios.