| Affected by COVID-19,public health security has become a high priority for the public.A series of anti-epidemic policies have been formulated,and wearing masks has proven to be the best method for cutting off the transmission route of the epidemic in public places.However,manual methods are inadequate for managing the public,not only due to the high density of people but also due to the high consumption of human resources.Therefore,an intelligent mask-wearing detection system has been developed to remind people to wear masks in public places.The relevant works are as follows:1.A dataset containing 13,000 images was created by combining RMFD from Wuhan University and other datasets found on the internet.Each image has its own XML file containing information on the target location and classification(Mask & No Mask).2.The detection accuracy of the neural network model has been improved by introducing modules such as SPP,PANet,and SENet to address the issues of missed and repeated detections in the original algorithm.Additionally,the convolutional layers were modified based on the idea of depthwise separable convolution to reduce the number of model parameters.3.The study focuses on the detection of the mask-wearing rate in crowds using the improved YOLOv4-Tiny algorithm.Firstly,the original YOLOv4-Tiny network was trained on a self-built dataset using pre-trained weights,and the results showed an average precision(m AP@0.5)of79.79%,recall rate of 77.78%,target missed detection rate of 15.3%,and target duplicate detection rate of 4%.Secondly,the same process was applied to the improved YOLOv4-Tiny network,and the experimental results showed an average precision(m AP@0.5)of 83.65%,recall rate of79.39%,missed detection rate of 4.6%,and duplicate detection rate of 1.1%.The evaluation indicators were respectively increased by 3.86%,1.61%,and decreased by 10.7% and 2.9% for missed detection and duplicate detection rate,which improved the detection performance of the YOLOv4-Tiny network.4.To verify the practical detection effect of the improved algorithm,a mask wearing detection software based on Py Qt5 is implemented,which is used for debugging equipment and displaying detection results. |
