Research On Face Recognition Method Based On Deep Learning

Face recognition has a wide range of application scenarios,and has become a very popular research content in the field of computer vision.Convolutional neural network as the representative of deep learning technology in the field of face recognition shows great advantages,its network structure is developing towards a deeper and wider direction,and gets better and better recognition effect.However,because of its complex structure,it is difficult for training to converge,occupies large memory and takes a long time to load,which is not conducive to the application in practical engineering.In order to solve this problem,this paper studies the application of deep learning in the field of face recognition.The research contents of this paper are as follows:Firstly,in order to solve the problem of long loading time due to the complex structure and large memory occupation of the original model,different improved models are designed by changing the structure and parameters of the deep residual network for the research of face recognition.The CASIA-Web Face data set is used as the training set.The different network structure is combined with cross entropy loss function and Triplet Loss function,and then supervised learning method is used to train.In the process of training,additional momentum method and batch normalization are used to optimize the training;At the end of training,the performance of different models is compared from memory consumption,loading time and accuracy rate.The experimental results show that the improved Res Net-30B model can simplify the network structure and reduce the load time and maintain a high accuracy.After that,in order to meet the needs of practical application,and it is difficult to label the category of face images in real environment,unsupervised learning method is used in the face matching stage,the trained models are combined with Chinese Whispers and K-means clustering algorithm to form multiple schemes for cluster matching.In this paper,the face images in the test set are randomly taken from LFW data set.After the cluster matching is completed,and the clustering effect is evaluated by using information entropy,F₁-Measure and other indicators.The experimental results show that the improved residual network combined with Chinese Whispers can complete the task of face recognition faster and better.Finally,the above scheme is applied in the intelligent catering platform based on face recognition in this paper,at the same time,the system framework,application mode,business process and middleware construction of the platform are described.The test results show that the system has good robustness,fast recognition speed and high accuracy.