The Design And Implementation Of Face Recognition System Based On Deep Learning

With the rapid development of computer vision technology and artificial intelligence technology,face recognition technology has also developed by leaps and bounds.Compared with traditional methods,the face recognition method based on deep learning has high precision,and its detection and recognition speed is fast.Affirmed by academia and industry,it is widely used in intelligent security,financial risk control,mobile "face-to-face" payment,and personal witness comparison in online business processing.This paper combines deep learning technology and TensorFlow deep learning framework to study and learn the current face recognition related technologies,and designs and implements a face recognition system based on deep learning.The system has functions such as face detection,face recognition,face key point positioning,and living body detection,which can accurately and quickly complete face recognition related services.The main contents of this article are as follows:1.The training of the algorithm model used by the system: using the TensorFlow deep learning framework as a platform,the training of face detection based on deep learning SSD,face recognition of FaceNet,and SEnet face key point positioning model is completed.It mainly includes preprocessing of the data set,modification of the network model,setting of training parameters,and testing of the trained model.Among them,the SSD-based face detection model uses SSD_ResNet50_v1_fpn as the backbone network,and uses the WIDER FACE data set to train and test the model;the FaceNet-based face recognition model uses the deep convolutional neural network of the Inception architecture and uses the loss of Triplet Loss Function,using the CASIA-facev5,CASIA-Webface,Celeba data set as the training set and the LFW data set as the test set,completed the training of multiple face recognition models through the comparison experiment of the training set,and selected the best through the test Model;based on SENet,the face key point localization model uses SE-ResNet as the network structure,and uses 300W-LP data set to train and test 68 face key point localization models.2.The overall design and implementation of the system: using the model weight files obtained after the aforementioned algorithm model training,the design and implementation of the server and the client are completed,and the system functions are tested.The design and implementation of the server side uses the TensorFlow deep learning framework as the platform and uses Flask as the web application framework to load the trained model,complete image upload,face detection,face registration,face login,live detection,etc.The design and implementation of the functional module interface;thedesign and implementation of the client,using the stable version of WeChat Developer Tools Stable Build as a tool,using HTML markup language,CSS cascading style sheets,JavaScript scripting language,complete the system main page,face detection,Face registration,face login,live detection and other pages design and implementation;system function test,through the use of WeChat developer tools of the real machine debugging function,completed the test of all system functions.