Research And Implementation Of Identity Authentication Technology Based On User Behavior Analysis

Now represented by smartphone mobile devices are becoming access various services such as social networking,the way such as financial transactions,to store a lot of privacy or sensitive information,so the identity authentication of the security demand of the smart phones are also growing,in order to more convenient,safer access to data.Implicit identity authentication based on gait behavior has attracted more and more researchers’ attention due to its convenience and privacy.It utilizes each person’s unique gait pattern to authenticate users through the gait data obtained from inertial sensors in smart phones,and almost requires no interaction with users.Identity authentication is defined as the process or action of authenticating the identity of users and then authorizing different users with different permissions based on the authentication results.Implicit identity authentication based on gait behavior can be divided into identity recognition and authentication,which correspond to the purpose of identifying the specific user identity and verifying the legal user identity respectively.The traditional identification method based on gait behavior has some problems in accuracy and efficiency,which can not meet the needs of practical use.With the continuous improvement of computing capability of mobile terminal devices,gait identity authentication using deep learning has become a new development trend.In view of the shortcomings of the existing research,this paper designs two different methods for identity identification and authentication respectively.The specific research work of this paper is as follows:(1)For the purpose of user gait identification,this paper proposes a gait identification technology based on multimodal image fusion(GMR).By converting the gait time series data into images that integrate the static and dynamic features of the gait data,and then using lightweight neural network combined with the attention module and the unbalanced loss function of multi-classification samples to extract the features of multimodal fusion images for classification to achieve user identification.The attention module is used to focus on the key feature information in the image and simplify the complexity of the model,and the unbalanced loss function of multi-classification samples is used to solve the problem of unbalanced sample number caused by the different acquisition time of different users in the gait data collection stage.In this paper,experiments are carried out on two public gait datasets,including exploring the effects of different image encoding methods,different attention modules and different loss functions on user identity recognition performance,and the effectiveness of the method is proved.(2)For the purpose of user gait authentication,a gait identification method based on auto-encoder neural network is proposed to verify user identity.Different from gait recognition in model training,the training data only contains samples of legitimate users.In particular,this method first loads the gait identification have been trained in lightweight neural network used in legal class user gait feature extraction,and then extracted to have good identity the characteristics of the degree of differentiation of the neural network as the encoder input for training,training a good model for legitimate users samples can very good reconstruction,However,the user sample of exception class cannot be reconstructed well.By judging whether the sample to be verified can be reconstructed well,the identity of the user to be detected can be verified.Experiments are carried out on public data sets and self-collected data sets to prove the effectiveness of the proposed method.