Identity Recognition Based On SSVEP Signal

The security and reliability of individual identification has always been a hot issue.Moreover,with the development of information technology,biometrics has received an in-depth study due to its own security and portability.There are more and more applications for personal identification based on biometrics such as fingerprints,faces,gaits,sounds,irises,etc.,but these fixed features are not suitable for some specific occasions due to their shortcomings in anti-counterfeiting and living body collection.Faced with this situation,biometrics based on EEG or ECG with concealment and variability has received more and more attention and research.It is expected that these signals will become an important part of the biometrics field in the future.The relevant theories and methods for identification based on the steady-state visual evoked potential(SSVEP,a kind of EEGs)are studied in this paper.According to the previous literature,an experimental paradigm for the SSVEP data acquisition is designed,and the pre-processing by the Empirical Mode Decomposition(EMD),such as filtering and denoising,is carried on the collected SSVEP data.On the basis of that,two identification methods on SSVEP are proposed.(1)Identification based on frequency domain encoding for SSVEP.According to the characteristics of the SSVEP spectrum,the paper designs a new encoding method to represent individuals,and the eigenvectors indicating the individuals are obtained following the SSVEP spectrum encoding.Afterwards,the identification system can match and identify the detection instances based on the feature vectors of all users.Finally,experimental results show that each individual's SSVEP signal remains stable for a sustained period of time,but there is a large difference among all individual's SSVEP signals.And the recognition accuracy of the recognition system can reach 90%-92.95%.(2)Identification based on deep learning for SSVEP.The paper studies in-depth the current popular deep learning methods and uses the Convolutional Neural Network(CNN)to learn the individual's SSVEP features.First,one-dimensional SSVEP signals are transformed into a three-dimensional image that preserves SSVEP time-frequency information and spatial information.And the converted image is the input of CNN.After that,the SSVEP data of the subjects in different time periods are analyzed and identified by a multi-classification network and a recognition system composed of multiple two-class networks respectively.Finally,according to the output of the loss function in each network we found that the SSVEP signals of three subjects remained stable for a long time while the signals of the other four subjects changed.In addition,the recognition performance of an identification system composed of multiple two-class networks is better than that of a multi-classification network.When the user is continuously identified twice,the recognition accuracy of the recognition system can reach 93.57%.Finally,according to the above experimental results,an idea of identifying users by establishing SSVEP database was proposed.And the basic functions of the SSVEP database were introduced in detail.