Research On Face Recognition Based On Deep Learning

Face recognition technology is an important branch in the field of biometrics and has a wide range of application scenarios.It uses the biological characteristics or behavioral characteristics of the human body to identify the identity,has good advantages,and has high application value in the fields of national public security,social security,financial security and human-computer interaction.Because face recognition technology is interfered by various environmental factors,it can not extract features well,and meet the real-time requirements of field applications.Deep learning has better generalization performance and expression ability,as an emerging field of machine learning.With the attention of researchers,more and more researchers apply deep learning to the field of face recognition.Therefore,applying deep learning to the field of face recognition is of great significance in the study of face recognition.This paper mainly extracts facial features from deep networks for exploration and research,and designs two improved models based on LeNet-5 network.The specific research contents are as follows:(1)In order to solve the problem that the LeNet-5 network can not extract the face features better,by adding the cascaded original LeNet-5 convolutional neural network layer in the network design,and decomposing the 5×5 convolution kernel in the network.Two low-dimensional convolutions of 3×3 can further increase the depth of the network,extract deeper feature information,and improve network performance and recognition.(2)Based on the above improved neural network model,a model with two layers of convolutional layer and 14 convolutional layers is designed to compare the effects of different convolution depths on the performance of the model.In order to reduce the training time of the model,make the network converge quickly,and improve the network performance,compare the performance of the three activation functions of ReLU,Tanh and Sigmoid on the model based on two different depth network models,because the function of the activation function is in the neural network.The introduction of nonlinear elements enables the neural network to perform nonlinear mapping.If the activation function is not used,no matter how many layers of the neural network,the output is only a linear combination of inputs,so the activation function has an important influence on the nonlinearity of the network.(3)In order to solve the over-fitting problem in neural network training,from the network model of the above training,select the model with the best performance,introduce Dropout technology into the network,and make the network sparse,so as to improve the generalization of the network model.The ability to further make the model more abstract and make the extracted features more distinguishable.For this reason,the effects of three different parameters on the network performance when Dropout are 0.3,0.5 and 0.8 respectively are compared.