Research On Face Recognition Based On Deep Learning Algorithm

With the rapid development of Convolutional Neural Networks(CNN)and computer hardware,deep learning has gradually become the core technology of image recognition.Thanks to the continuous evolution of deep learning algorithms in recent years,it provides a good research condition for high-precision recognition of face images.This paper studies the face recognition based on deep learning related algorithms.The face recognition system inputs video image data through the camera,and after detecting the face,the target detection algorithm is used to analyze the three characteristics of the gender,age and expression of the face,and finally the recognized result is output.Aiming at the environmental factors in the process of face recognition,the small amount of face data collected by humans,and the low accuracy of single recognition neural network,the corresponding solutions are given in the paper,and finally verified by experiments.Mainly conducted research on the following aspects:(1)Build a combined database to improve the accuracy of the experimental results.Avoid directly using data sets published by the network,and form a combined database through human collection and web spider.(2)Use data enhancement to solve the problem of insufficient data volume and objective factors.A variety of different ways of data augmentation for combined databases,increasing environmental conditions for multiple lighting and angles,and increasing the number and richness of combined databases.(3)Build and train a deep learning network.Based on the research of convolutional neural networks,a small neural network is designed and cascaded to solve the problem of low accuracy of single identification network,and an improved deep learning cascade network is realized.Prove the superiority of cascaded networks by designing controlled experiments.(4)Achieve successful porting of the face recognition system from PC to iOS.The system can accept the real-time video data collected by the camera and complete the detection and identification,and design the user interface to display the indicators and results of the system to identify the author's own face.In the face recognition system built in this paper,both the training phase and the test phase use a self-constructed combined database to complete the data annotation work on the image.For some types of face data that are lacking,data augmentation technology is used to expand the amount of data and improve the complexity and diversity of the combined database.The face image data is trained based on the existing deep learning network framework.In order to achieve the convergence of the loss function as much as possible,the hyperparameters are repeatedly adjusted until the gradient falls close to the local optimal value,and the optimal network model is selected.The test data is used to test the network model.The experimental results show that the average accuracy rate is over 90%,which indicates that the face recognition system studied in this paper has good feasibility and practicability.