Research On Face Recognition In Specific Scene Based On Convolutional Neural Network

As an important part of biometric identification,face recognition technology has been a hot topic in recent years,whether in the field of computer vision,artifi cial intelligence,or other scientific fields.In practical applications,the face data samples that can be obtained due to various factors are limited,and in some cases,only a single face data sample can be obtained.In view of this situation,this paper studies how to train the face recognition model with insufficient sample data,and then appl'ies the trained face recognition model to Win dows and Android platforms respectively.This paper starts from the two aspects of face data samples and convolutional neural network structure,and solves how to realize face recognition in the absence of training samples.In terms of face data expansion,the ED small sample face data set is first constructed.There are 90 individual face data im ages in this sample.Secondly,the data enhancement technique,the symmetric face expansion meth od,the bit plane method and the confrontation generation networ k are used to augment the small sample face data set.Finally,the extended ED face dataset is compared with the ORL face dataset.The experimental results show that the extended ED face dataset is comparable to the ORL face dataset during network training.In terms of network structure,in order to construct a convolutional neural network model suitable for small sample data sets,this paper first analyzes the VGG network model,improves the VGG n etwork model from the aspects of the model's parameter quantity and calculation amount,and then improves the VGG.The SVGG network model is constructed by combining the network model with the twin neural network.Finally,the SVGG network model is train ed using the extended ED face data sett,and the trained network model is analyzed.The experimental results show that the recognition rate of the SVGG network structure constructed in this paper is 92.6%on the small sample dataset,which is significantly higher than that of the twin neural network on the small sample dataset.In the face recognition application,this paper uses the trained SVGG network model to implement the face recognition system on Windows and Android platforms respectively.Due to the limitations of the Android platform's computing power and memory mechanism,it is more difficult to implement face recognition using convolutional neural networks on the Android platform.Therefore,this paper puts the training process of the model in the upper computer,and then transplants the trained model to the Android platform,which not only satisfies the hardware requirements of the Android platform,but also ensures the efficiency of face recognition.