| Identity authentication is the verification and authentication of individual identity,which constitutes the basis of trust in social activities and interpersonal communication,and has a wide range of needs in the fields of national defense security and commercial economy.Modern brain functional imaging studies have found that the human brain not only has differences in the shape and structure of gene representation,but also has differences in functional connectivity caused by thinking patterns.The subjective initiative of the brain protects the user ’s subjective authentication intention,and also has the function of regular replacement of internal authentication content.Because of its high time resolution and portability,EEG has become a new biological feature and has attracted much attention.However,at present,there is still a gap between EEG identity authentication technology and practical application,and it still needs to be improved in terms of accuracy,stability and security.Especially in terms of security,the existing authentication paradigms generally use supraliminal presentation to induce brain responses containing individual characteristics,so it is difficult to resist the attack of non-blind intruders.Non-blind intruders deceive the authentication system by implementing a conscious psychological suggestion strategy to disguise the brain response of legitimate users.Subliminal stimulation can inhibit the conscious attack of the intruder from the source by inducing the brain instinct response of the individual in the unconscious state.On the other hand,the brain is inherently sensitive to self-information,and the unique brain response induced by subliminal self-information can effectively distinguish between legitimate users and intruders.Therefore,studying the subliminal self-information processing mechanism and its application in EEG identity authentication will provide an important reference for the internal identity authentication method with higher security requirements,and promote the practical application of this cutting-edge technology in the field of information security.This paper focuses on the problem of ’how subliminal stimulation-induced EEG signals represent individual identity,and conducts research from two aspects: self-information processing mechanism and subliminal EEG identity authentication algorithm.In the subliminal processing mechanism of self-information,a variety of subliminal stimulus paradigms were explored.From the two levels of scalp EEG signal and cortical source response,the information transmission and integration mode between brains during subliminal processing of self-information was explored,which provided theoretical support for subliminal EEG identity feature extraction and authentication algorithm design.In the subliminal EEG identity authentication algorithm,the identity feature design scheme based on artificial experience and deep learning is explored.The subliminal single trial EEG feature extraction,P300 EEG data enhancement algorithm and feature fusion EEG identity authentication model are studied,which improves the accuracy,stability and practicability of the EEG identity authentication model.The main research work of this paper is as follows:1.EEG identity authentication paradigm of subliminal visual stimulation.How to design a subliminal paradigm to induce individual differences in the unconscious state of the brain is the basis for achieving safe and reliable identity authentication.Aiming at this problem,this paper explores the subliminal presentation effect of various self-attribute stimulus materials,and on this basis,constructs an EEG identity database for different application scenarios to verify the feasibility of the subliminal paradigm.The experimental results show that on the data sets of 219 subjects in three different application scenarios,20 ms of name stimulation combined with masking map can induce subliminal semantic priming,and the brain response induced by subliminal self-name and other people ’s name is significantly different,especially in the P300 component of the parietal lobe and the N250 component of the prefrontal lobe.This study verifies the feasibility of subliminal name stimulation for EEG identity authentication,and provides important support for subliminal self-information perception mechanism research and EEG identity authentication model construction.2.Brain response mechanism of subliminal self-information processing.The self-information cognition and processing process of subliminal state is complex,and the related brain response mechanism is still unclear.Aiming at this problem,this paper analyzes the original scalp EEG signal and the traceable cortical response,locates the important brain functional areas of selfinformation processing,applies the directed time-varying brain network to study the dynamic information flow of self-cognitive process,and summarizes the unique processing mode of subliminal self-information.The results show that the brain has a unique processing mode for selfinformation,and self-information awakens high-intensity and large-scale brain activation.The results showed that self-information awakened the brain response of the superior frontal gyrus and the limbic system,and the cortical activation intensity and brain information interaction efficiency were higher after 300 milliseconds of self-information presentation.This study explores the specific brain response of subliminal state self-information processing,and provides theoretical basis and support for later identity feature extraction and classification model improvement.3.Single-trial EEG identity features induced by subliminal stimulation.The weak subliminal stimulation intensity has a low degree of brain arousal and a small activation range,which makes it difficult to extract effective identity features,especially increases the difficulty of single-trial identity authentication.Aiming at this problem,this paper proposes a single-trial identity authentication method based on adaptive feature extraction.The cross-domain feature evaluation criteria are used to select high-quality features and are used to resist professional attacks from nonblind intruders.The results of feature visualization and classification show that the performance of identity feature clustering is improved by 12.73 %,and the average area under the curve of single-trial authentication reached 93.32 %.When the authentication system is subjected to blind intrusion and professional non-blind intrusion,the recognition rate of the identity authentication model for the two intruders is 97.3 % and 97.1 %,respectively.This study provides a single-trial EEG identity feature extraction method,which effectively improves the resistance of the identity authentication model to non-blind intruders.4.P300 EEG data enhancement based on generative adversarial networks.The scale of the EEG identity database is generally small,and the small sample library limits the performance of the identity authentication model.Aiming at this problem,this paper proposes a network training framework for generating task-state data guided by EEG resting-state data,which can mass produce high-quality artificial EEG data.A hybrid training set of artificial and real EEG is constructed,and artificial EEG signals are used to expand,balance and enhance identity data.The results on the self-built identity authentication database and two public databases show that the EEG data generation model improves the convergence and stability of the model.In the timeliness test,the training time of the network is shortened by 95 % and the demand of subjects is reduced by 73 %.Data augmentation can alleviate the problem of uneven positive and negative samples,and significantly improve the classification performance in the performance test of multiple databases.This study improves the performance of EEG classification model under the limitation of small samples and provides data support for deep network model training of identity authentication.5.EEG identity authentication model based on cross-domain multi-scale feature fusion.The existing deep networks do not make full use of the multi-domain identity feature information in EEG data,and the identity authentication features are weak in interpretability and limited in performance.Aiming at this problem,this paper fuses the identity features at two levels : data presentation form and feature extraction method.Using the three-dimensional EEG presentation form of time,frequency and spatial domain fusion,a three-dimensional multi-scale EEG crossdomain hybrid identity feature extraction model is constructed.The combination strategy of multiview convolution kernel is used to extract multi-scale identity features.The experimental results show that the accuracy of the cross-domain multi-scale convolution model is improved by 5.23 %on average,and the visualization results show that the rectangular convolution kernel with time domain preference has better performance.Through the reverse heuristic parameter optimization of the model results,the results show that the number of multi-branches should match the number of EEG components,and the shallow multi-branch structure has a high application cost performance.This study constructs multi-angle identity features of mixed feature domains,which improves the classification performance of subliminal EEG identity authentication. |
PDF Full Text Request