Research Of Smartphone User Identification Model Based On Sensor Data

With the rapid development of the Internet era,smartphones,which are currently widely used in social,e-commerce,business and entertainment,have become an indispensable part of people's lives.Unlike traditional communication tools that only support communication functions such as text messaging,the applications installed on smartphones offer them more functions,and the involvement of the Internet makes smartphones truly feature-rich smart terminals.At the same time,various social applications installed on smartphones contain users' private information,meanwhile,Alipay,WeChat and other payable applications also involve user property information,so the data on smartphones is very sensitive and the user security of smartphones is especially important.In order to solve the above problem,this thesis proposes to identify the user identity based on the data returned by sensors when smartphone moves.The specific research contents are as follows:(1)Acquire smartphone sensor data of model research in two scenarios: self-collecting and searching for public dataset.Program the Android application to collect the data to construct dataset generated by users shaking the smartphone;select the public smartphone sensor dataset that conforms to the smartphone identification.(2)Processing of raw perceptual data.Since the identification of the user requires high accuracy,the extraneous data and noise caused by the difference of the sensor hardware device are processed.Extracting from the frequency domain to increase the discrimination between the users,and other pre-processing operations before the input are performed.(3)Aiming at the defects of the existing identification methods,starting from the sensor data generated when the smartphone is moving,corresponding to the two datasets,a user identification model combining convolutional neural network and recurrent neural network is proposed.Among them,the two-stream convolutional neural network is used to extract the time domain and frequency domain features of sensor data,and the subsequent fusion convolution is used to fuse the time-frequency characteristics of different types of sensors;clockwork recurrent neural network(CW-RNN)further analysis of the hidden information in the feature sequence obtained after convolution,and mining the context of the time series.(4)The proposed model is compared with the common convolutional neural network and traditional machine learning algorithm.The experimental results show that the proposed model outperforms other methods.Compared with other similar work,the proposed model also has an advantage in performance.It can effectively identify smartphone users in various situations with good universality.