Research On The Recognition Of Face Individual And Gender Based On Convolutional Neural Network

Firstly,if humans want to identify everything in the world is to classify various species and then to distinguish each type of thing.Among them,the difficulty of distinguishing is the things between the same categories.The distinction and recognition between them are mainly based on the facial,body type and individual behavior.From the biological principle,the process of classification in humans is transmitted to the brain through the eyes,and then the brain makes feedback based on the images seen by the eyes,and finally classifies and names the things that are seen.This type of transmission of eye-to-brain classification recognition is achieved by a human neural network system consisting of a myriad of neurons.According to this biological principle,in the era of human beings entering the information age and the development of creation technology,people began to study machine learning instead of doing some simple classification problems.Face recognition is a representative classification application.Subsequently,the development of machine learning technology promoted the emergence of deep learning methods,especially in the field of deep learning,the principle method of imitating the human nervous system,representative of the Convolutional Neural Network(CNN),duo to the mature technology of the CNN framework method and classification technology,showing excellent classification recognition.Therefore,the CNN method has been sought after by scholars at home and abroad in recent years,and has been loved and applied by many scholars in the field of face recognition.Based on the investigation and research of CNN face recognition,the focus of this paper is to construct the model structure and classification method of CNN to improve the accuracy and efficiency of recognition.The main work arrangements are as follows:(1)In view of the problem that the size of the face data set used in this paper is too large and there is useless feature information,the integral part and the cascaded algorithm are used to extract the face part required for the research in the data set.Final this paper will extract the face of the required size from CASIA-WebFace,LFW and Adience data sets,and then rename the data set.(2)For many face recognition technologies based on CNN to improve face recognition rate,their network model input parameters are neglected,resulting in many model input parameters,long training time and inability to run on hardware with small memory.The problem is to propose a face recognition model based on the improved SqueezeNet.The improved SqueezeNet model uses the first and last pooling layers to introduce corresponding subsequent convolutional layers for feature fusion,and extracts subtle facial texture features to stabilize the model convergence.For the improvement of the classification function Softmax,the L2 norm constraint method is used to constrain the features of the last layer in a spherical plane,reducing the same feature spacing and improving the network convergence ability.and the effectiveness of the method was verified in the experiment.(3)In order to further improve the accuracy of face recognition of CNN,a new fusion deep convolutional neural network model(NFDCNN)is proposed based on the traditional fusion model.In the NFDCNN model structure,the convolution layer between every two sampling layers fuses the sub-sampling features of the previous level before the convolution feature extraction.This method can preserve the original feature information and fuse with the deep texture,which has a high degree.Reduce the degree of reduction and reduce network errors.The NFDCNN model classification function has been improved on the conventional Softmax,and the regional edge classification function AM-Softmax has been introduced.The classification function is divided by a region as a boundary.The function distribution has an extended inter-class distance and is reduced.Intraclass distance,and verify in the experiment.