Research On Face Recognition Algorithms Based On Deep Learning

Face recognition technology is not only getting closer to our daily life,but also expanding its application fields.It has great potential in the fields of Internet of Things,Internet security,consumer electronics,automotive electronics and so on.However,when the illumination intensity becomes stronger or weaker,the recognition accuracy of the face recognition system will become very poor,which can not achieve the desired results.With the rapid development of artificial intelligence and machine learning,face recognition technology based on deep learning has become a hotspot of research and application development.Convolutional neural network avoids manual feature extraction or artificial rules,but automatically extracts features from raw data.Compared with traditional methods,convolutional neural networks can learn more efficient features and patterns.On the basis of reading and researching a large number of documents and materials related to in-depth learning and face recognition,the main work of this paper is as follows:(1)Firstly,the origin and development of deep learning are introduced in detail,the principle of deep learning is analyzed,and two common models of deep learning,i.e.,restricted Boltzmann machine and self-encoder,are discussed.Then,the structure of convolution neural network is analyzed,and the operation process and functions of convolution layer,pooling layer and full connection layer in convolution neural network are summarized,as well as the two network characteristics of local connection and value sharing.Finally,several commonly used convolution neural network models are illustrated by examples.(2)Aiming at the problem of image classification with small samples,this chapter proposes a method of migrating the Inception-v3 model trained on ImageNet to its own image classification data set.This method can directly use the trained neural network to extract image features,and then use the extracted feature vectors as input to train a new single-layer fully connected neural network to deal with new classification problems.Finally,94.5%recognition rate is achieved on the self-built sample data set.(3)Through the research and study of LeNet and ZFNet models,this paper designs an 8-layer convolution neural network model,which optimizes and adjusts the network continuously through data set enhancement,loss function improvement,Dropout technology and optimizer.Finally,the convolution God-level network algorithm in this paper is optimized and adjusted with traditional PCA,KPAC and two-dimensional PCA.Linear Discriminant Analysis 2DLDA and Kernel Direct Discriminant Analysis KDDA are compared in ORL database.The experimental results show that the recognition rate of this algorithm in ORL face database is better than that of traditional face recognition algorithm.(4)The shortcomings of in-depth learning are discussed in depth,and the future of in-depth learning and artificial intelligence is prospected.