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

The information system can authenticate user identities based on shared keys,public key encryption algorithms,and biological characteristics,and grant corresponding access rights.With the development and popularization of information technology,the functions of information systems tend to become more complex and comprehensive,and the requirements for identity authentication capabilities are getting higher and higher.The widespread popularity of smart phones has caused users’ work and life and other types of private information to gradually shift from the PC side to the mobile device side,and the security requirements for mobile identity authentication technology continue to grow.The current identity authentication methods used on mobile terminals mainly include password authentication,fingerprint and face recognition,and behavior analysis.In terms of identity authentication results,these authentication methods can verify whether the token is correct to determine whether the request comes from a legitimate user;in addition,face recognition and behavior analysis can also identify which user the request originated from.Behavior analysis uses the biological behavior captured by the embedded sensor in the smart phone to perform continuous implicit identity authentication.This paper proposes an identity authentication method based on deep learning in the scenario of detecting abnormal behavior and identifying behavior identity on the mobile terminal.The specific research work of this paper is as follows:(1)For the scenario of abnormal behavior detection,this paper designs and implements a mobile identity authentication system based on the ConvLSTM neural network model,which can detect attackers who have stolen the access password and tried to access the smartphone,achieving an accuracy of 99.4%.And in the application program,the additional cost of behavior analysis technology to the system was tested,and an average 8.04%extra CPU loss and 9.45%time delay were achieved.(2)For the scene of identifying the current user’s identity,this paper proposes a method of learning behavioral characteristics through the SE residual network model of deep connection attention and uses the angle-enhanced Softmax-loss layer to strengthen the classification ability of the network.In order to make the experimental data more suitable for users’ daily habits,this article uses the public data set BrainRun collected in an uncontrolled environment and adds a comparative experiment on the H-MOG data set.Our method is evaluated in both the graphic unlocking data set and the BrainRun data set,and the accuracy of identifying the correct identity is 96.94%and 87.11%,respectively.A 93.68%accuracy rate was achieved on the H-MOG data set.