Research On Implicit User Authentication Methods Of Smart Terminals

Smart terminals usually use passwords,physiological features such as fingerprints and other means to achieve user authentication.Traditional authentication methods work when the user unlocks the phone,but cannot continuously authenticate the user’s legal identity.So the one-time authentication achieved by traditional authentication methods cannot meet the security requirements.In order to achieve continuous and non-interference user authentication of smart terminals,the implicit authentication technology based on user behavior features is proposed.However,existing implicit user authentication methods have many problems such as poor authentication accuracy,weak robustness and long authentication time.Therefore,this paper studies implicit user authentication methods of smart terminals.In the no-operation authentication scene,incomplete representation of gait features and changes in user activity scenes lead to poor authentication accuracy and weak robustness of methods,respectively.This paper proposes an implicit authentication method for fine-grained gait representation based on scene awareness.In the proposed method,the multi-sensor signals and their different frequency component signals are fused as the input and input signals is segmented according to the gait cycle to construct a gait dataset.The proposed method constructs a feature extraction network by fusing convolutional neural network and long-short-term memory network for mining the hidden features of input signals and realizing a more comprehensive gait feature representation.To reduce the impact of changes in the user’s activity scene,a staged cascaded authentication framework is constructed that first recognizes the activity scene and then authenticates the user’s identity,and considers the impact of changes in the terminal’s carrying position.Experimental results show that when the input is a signal of a gait cycle length,the authentication F1 value of the proposed method can reach 99.64% and the performance can be kept stable when the user’s activity scene and the terminal’s carrying position change.In the operation authentication scene,aiming at the problem of unbalance between authentication accuracy and authentication frequency caused by long authentication time,this paper proposes an implicit authentication method that integrates tapping,swiping and keying features.In the proposed method,a time expansion model of tapping is first constructed,which enriches the feature representation of tapping.To improve the authentication accuracy,the proposed method designs a neural network-based feature extraction model that integrates touchscreen data and motion sensor data,and then constructs a dual-channel fusion feature space.The proposed method also considers changes in the user’s activity scenes and differences in the terminal’s holding postures.Experimental results show that the proposed method can reduce the authentication time while ensuring the authentication accuracy.The authentication F1 value can reach 99.40% by combining the identity judgment results of the user’s three operations.To verify the implicit authentication method proposed in the two authentication scenarios,this paper collects the gait data and operation data of volunteers of different ages and backgrounds.The gait dataset is composed of motion sensor data for walking,upstairs,and downstairs and the operation dataset is composed of touchscreen data and motion sensor data for tapping,swiping and keying.To further verify the practical usability of the authentication method,the paper divides the gait dataset and the operation dataset into a sub-dataset for training the model and a sub-dataset for validating the model transferability,and demonstrates the generalization ability of the proposed authentication model on a dataset of few samples.This paper verifies the performance of the proposed user authentication methods on the above datasets,and also provides a reference dataset for subsequent research.