Research On Individual Identification Approaches Based On EEG Signals

Biometric identification is a heated topic in recent years.Compared with traditional identification systems,the biometric identification system is much safer and it is hard to copy or be hijacked.With the development of EEG,the EEG-based identification system starts to draw lots of attention.In present work,studies in this field focus on resting-state of eye open(REO)/closed(REC)and event-related potential(ERP)of EEG signals and studies are limited to a single run experiment.But most of these studies ignored the time-robustness for features of EEG signals between experiments.It is important for an EEG-based identification system and is worth investigating the stability and time-robustness of EEG signals over time.To study these issues,in this work,we design three paradigms,which are restingstate of eye open(REO),resting-state of eye closed(REC)and Oddball,to obtain the EEG signals.Five feature extraction methods are proposed to obtain different features of EEG signals.For both REO and REC,the classification results of single intra-run and fusion intra-runs data can reach 90%,which are higher than the results of ERP.So we use SVM as a classifier to obtain results of inter-run data independently and to find the optimized frequency range which is robustness over time The optimized results of cross-spectrum amplitude,PSD series,and channels coherence can reach 80% for REO and REC,which are higher than optimized results of PSD mean and phase lags.The optimized result of PSD mean for REC is 70%,which is much higher than the result for REO.We get the optimized combined frequency range of REO is 13-30 Hz and the optimized combined frequency range of REC is 13-30 Hz.The time-domain feature are extracted for Oddball paradigm to analyze and classify.The classification results of ERP are around 70% for single intra-run and fusion intra-runs data.In conclusion,we test the different features for single intra-run and fusion intraruns data of resting-state and ERP,and further study on features for inter-run data of resting-state.We found that features have stability and time-robustness for inter-run data of REO and REC.This is important and provide theoretical support for the research of individual identification based on EEG signals.