Research And Application Of Face Mask Recognition Based On Deep Learning

The current face recognition method cannot accurately complete the face recognition task in the unconstrained environment in reality,and under the raging situation of the new crown virus,people need to wear a mask when going out,which causes the face to be covered by a large area,so it is effectively handled These problems to improve the recognition efficiency are still difficult points in the face recognition system.Nowadays,the network model based on deep learning has become a research hotspot in the detection of targets and occluded faces.Therefore,this paper mainly focuses on the problem of mask detection and occlusion face recognition based on deep learning to optimize the network model and analyze the optimal loss function to improve the accuracy of face and mask recognition.The main work and innovations are as follows:1.For the deep learning original SSD target detection model,there is a problem that the lower feature layer needs to learn local information and high-level information at the same time.The SSD mask detection model with residual structure is designed;in order to obtain local features of different sizes,and get more abundant Feature representation,add improved spatial pyramid pooling to the residual-based SSD mask detection network,and use feature fusion to optimize the mask detection network;use the CIo U loss function to avoid the divergence problem of the SSD network,and form the final RS?SSD mask detection The network improves the efficiency of mask detection while speeding up network training.Experiments have proved that the RS?SSD mask detection network can perform mask detection well.2.According to the problem of poor recognition efficiency when multiple parts of the face are occluded,this paper proposes a recognition method of Re?Face occlusion of the face.The preprocessed face target is detected through the RS?SSD network,and then the face image is segmented to remove the sample redundant information and reduce the calculation of the entire network;then use the improved IVGGNet to extract more refined features in the face segmentation;In order to reduce the interference of occluded feature blocks on recognition performance,a parallel convolution pooling processing module,namely In?KPCANet,is added to the KPCANet network as an occlusion discrimination network to discriminate occluded faces;finally,the feature extraction and occlusion discrimination results are combined and sent Enter the SVM classification group for training to achieve efficient face recognition with occlusion.Experiments have proved that the Re?Face occluded face recognition method has a high recognition effect for different proportions of occluded faces.3.The above improved model is applied to the field of real-time mask occlusion face recognition to ensure the health of people wearing masks when entering and exiting places that require face recognition during the epidemic,so as to design a real-time recognition mask occlusion system.First,use python to crawl the static pictures of many stars wearing masks and not wearing masks.After manual processing,the face data set covered by the mask is constructed,and the preprocessed data set is sent to the RS?SSD mask detection network model to complete the mask detection task.Input to the Re?Face occlusion face recognition model for training,obtain the trained mask detection model and face recognition model respectively,and use the training model to build a mask occlusion face recognition system.