Research And System Implementation Of Real-time Detection Algorithm For Mask Wearing Based On Deep Learning

Since December 2019,the new crown pneumonia epidemic has ravaged the world.Up to now,the epidemic situation in many countries in the world except China has not been effectively controlled,which seriously threatens the lives,health and safety of people around the world.Respiratory droplets are one of the main vectors for the spread of the new coronavirus.Wearing masks in public places is the most simple and effective protective measure to prevent virus infection.Places with high traffic such as shopping malls and stations are important positions for epidemic prevention and control.In order to achieve efficient and intelligent non-contact detection of wearing masks,this topic has carried out research on the detection method of mask wearing based on deep learning technology.It mainly includes four parts,specifically:(1)Using the open source face detection data set,plus the pictures taken by individuals and collected from the Internet,the mask wearing detection data set in this paper is constructed,which contains a total of8,700 pictures with rich scenes that have been marked in detail.The clustering algorithm performs cluster analysis on the data set to obtain appropriate a priori detection boxes.(2)Aiming at the detection problems of dense crowds,far and small targets,and occluded targets in the mask wearing detection task,a mask wearing detection method for complex application scenarios is proposed.First,by introducing a lightweight channel attention mechanism into the backbone feature extraction network of YOLOv4,the model can adaptively learn the importance of different channel features,the relationship between different channel features and the current detection task.model the importance of,thereby focusing on key feature information while suppressing the interference of irrelevant features.Secondly,a new feature enhancement network is constructed by reusing a spatial pyramid pooling module,which further improves the model’s ability to extract and fuse spatial features in different receptive fields.By fusing global features and local features,the features are greatly enhanced.The expressiveness of graphs.It is verified through experiments that the average accuracy of the improved algorithm for detecting face targets and face targets wearing masks has increased from 89.05% to 94.81%,and the detection speed on the RTX3090 graphics card can still reach about 39 FPS,achieving high Accurate real-time detection.(3)In order to meet the needs of real-time mask wearing detection on relatively low-end hardware devices,a lightweight real-time detection algorithm for face wearing masks is proposed.First,the CBAM module integrating the channel and space dual attention mechanism is embedded into the backbone feature extraction network of YOLOv4-tiny,so that the model can learn the distribution of feature information in space and channel dimensions,and improve the model’s ability to express key details..Secondly,the spatial pyramid pooling module is introduced to perform multi-scale receptive field pooling on the deep features of the backbone network,and stack and fuse the acquired features with the original features to improve the fusion and utilization efficiency of the deep features of the model.Finally,combined with the design idea of the path aggregation network,a feature aggregation path from low-level and deep layers is added on the basis of the original feature pyramid network,which further improves the fusion ability of the model with different levels of features and enhances the efficiency of feature utilization.The experimental results show that the mask wearing detection accuracy of the improved algorithm is 7.47% higher than that of YOLOv4-tiny,and the real-time detection speed of about 90 frames per second is achieved on the RTX2060 graphics card.(4)Pycharm,Python,PyQt and other tools are used to complete the system design of mask wearing detection,which realizes functions such as static image detection,local video detection,and calling the camera to collect real-time video streams for mask wearing detection.After algorithm testing and robustness experiments,the detection performance of the two improved algorithms studied in this topic has been greatly improved,and they can complete the mask wearing detection task in various scenarios with high detection accuracy and real-time performance.It achieves high-precision mask wearing detection in complex scenarios,and the lightweight algorithm can meet the requirements of real-time detection on ordinary hardware devices,further reducing the hardware cost of algorithm deployment,and has good application value.