Research On Face Recognition Based On Convolutional Neural Network

With the rapid development of computer technology and deep learning,face recognition technology based on convolutional neural network has become a hot research topic in the field of computer vision.The application scenarios of face recognition are also widely used.Face recognition technology plays an important role in social security,human-computer interaction,intelligent transportation,Internet,financial and other fields.Therefore,higher accuracy requirements are put forward in the field of the face recognition algorithm.It is difficult to meet these two conditions that the network model is light and the recognition accuracy is high.It is the current status of most face recognition algorithms based on convolutional neural networks.The deep learning network structure generally has the defects of large parameter and complex model;the existing deep learning loss function is also difficult to achieve the ideal feature classification standard.It is difficult to achieve the minimum inter-class distance greater than the maximum intra-class feature distance.In this thesis,a face recognition algorithm based on angular distance loss function and dense convolutional convolutional neural network is designed in the open-set protocol.The feature is more discriminative,meets the ideal classification criteria of features,and improves the accuracy rate of face recognition task.At the same time,the network structure of this thesis adopts the most advanced dense connection module,which greatly reduces the parameter quantity of the traditional network structure,making the network model more concise and light.The face recognition algorithm based on convolutional neural network designed in this thesis adopts a dense connection network with a depth of 122 layers and a width of 32 layers.It uses the angular distance loss function of the superparameter ? designed in this paper to take 0.75 to supervise the network structure.After a lot of analysis and experiment,the accuracy of face recognition on the LFW dataset reached 99.45%,and the face recognition accuracy of 72.534% and 85.348% was obtained in MegaFace's face confirmation task and face verification task respectively.The superiority of the algorithm presented in the face recognition domain is confirmed.Finally,this thesis implements a face recognition system based on the face recognition algorithm,which can identify the face in the video collected by the remote IP camera in real time and obtains high recognition accuracy in the real test scenario.It is verified that the algorithm is not only rigorous and feasible in the theoretically but also efficient in practical engineering applications.