Research And Implementation Of Face Recognition Based On Convolutional Neural Network

With the continuous development of deep learning technology,the convolutional neural network model is continuously enriched,and the face recognition technology has been studied by many scholars.Face recognition has penetrated into people's daily lives,such as access control systems,camera surveillance systems and smartphones.The implementation of traditional face recognition techrnology requires complex operational steps,including artificial feature extraction,feature selection,and classifier selection.And in the case of strong illumination,obstruction,attitude change,etc.,the robustness of the algorithm is relatively poor,and the recognition accuracy is not high.With the increase of data volume and the improvement of GPU computing performance,convolutional neural networks have made major breakthroughs in face recognition.However,under the premise of limited hardware configuration,it is difficult to train the convolutional neural network to play a better role in face recognition.To this end,this paper researches and improves the AlexNet shallow layer network and the VGGNet deep layer network respectively,aiming at improving the accuracy and speed of model recognition and enhancing the robustness of the model while reducing the demanding hardware requirements of the original network.The main contents of the full text include:1.Outlines the current status and related knowledge of face recognition.The structure of each level of convolutional neural network is introduced,and the related methods involved are theoretically deduced.Introduced the TensorFlow framework and the official face database used in this article.2.Face recognition based on improved AlexNet shallow layer network.First,the improvement method for AlexNet is to replace the original large convolution kernel with multiple small convolution kernels,and reduce the number of layers of FC to adjust its network structure,and integrate the idea of improved Triplet Loss into the network training.in.Then,the improved network structure is tested on the relevant data set.The improved analysis of the improved AlexNet network further improves the accuracy of the network,thus demonstrating the feasibility of the model adjustment and improved Triplet Loss method.3.Face recognition based on the improved VGGNet deep layer network.The way to improve VGGNet is:First,draw on the effective improvement method in AlexNet network,and then introduce more convolution kernels to adjust the structure of the network;then,introduce the ResNet residual block on the initial improved model.The face recognition experiment on the relevant dataset proves that the network model can avoid the gradient disappearing and achieve better recognition accuracy with the increase of the number of network model layers.Finally,in the stage of face matching,The speed of matching is improved by introducing a KD tree.4.Implementation of face recognition in the natural environment.First of all,self-built face database in the natural environment,and then through the two network models that have been trained and improved to experiment on the face data set in the natural environment,verify the feasibility of the above two improved models.