| With the continuous expansion of the application boundary of face recognition,the research on face recognition is also deepening,the scene of face recognition application is changeable,there are style differences,this paper on the basis of face detection for face recognition,and complete the style change of face recognition system algorithm and software implementation,the main content is as follows:(1)The face detection algorithm based on YOLOv5 s is improved,and the structure of Ghost Bottleneck in Ghost Net network is used to replace Ghost＿SC the CSP structure in the YOLOv5 s backbone network and reduce the amount of parameters and calculations of the network.The CA attention mechanism is added to balance the accuracy loss caused by reducing the amount of calculation,and the new activation function Swish is used to improve the accuracy of face detection,and the results show that the accuracy of the proposed GSC＿YOLOv5s face detection algorithm reaches 96.2%.(2)For the face recognition algorithm,based on the feature extraction network design of FaceNet,the deep separable convolution with less computation is used to design the backbone network,combined with the residual structure to make full use of the effective features,and Re LU6 is used as the activation function in the feature extraction network,the model size is 9.1M,only one-tenth of the size of the original FaceNet model,and the recognition accuracy is 92.9%;Aiming at the difficulty of Triple Loss convergence,Cross-Entropy Loss is used for face classification to assist Triple Loss convergence,and when CET Loss is designed as the loss function of the MR＿FaceNet,the recognition accuracy is increased to 94.6%.In terms of face recognition optimization for style change,comprehensively considering the problem of style change in the specific scenario of face image recognition,PSC-GAN(adaptive face image migration processing)is proposed,and the PSC-GAN proposed by us can effectively retain the face information features of the source domain and migrate them to the target domain style.The experimental results show that the face data sets of different styles can effectively improve the average recognition accuracy of the face recognition model after the same style migration through PSC-GAN.(3)Based on PyQT5 and python,realize the design and implementation of the face recognition system on the desktop side.The improved detection algorithm and recognition algorithm in this paper were used to complete face detection and recognition,and the performance of the trained face detection algorithm and face recognition algorithm was verified on the software,and 100 people were divided into five groups for experiments,the average face recognition accuracy was 99.6%,and the single recognition time was 338 ms. |
