Research On Eeg Feature Detection Method Of Rapid Serial Visual Presentation

Rapid serial visual presentation(RSVP)is a typical experimental paradigm of Brian computer interface(BCI),which presents a large number of image stimulation at 10 Hz rapidly.Electroencephalogram(EEG)was collected and processed in real time when subjects were watching image serials.EEG features related to the target image can be extracted and used for classification,so as to recognize specific target images.In this paper,by analyzing the RSVP-EEG signal in time domain,frequency domain,spatial domain and synchronous coupling analysis,it turned out that besides the P300 and N200 component responsed in delta and theta band,the energy characteristics of specific spatiotemporal-frequency distribution and the significant coupling difference between specific channels of gamma rhythm RSVP-EEG was discovered in RSVP paradigm.Therefore,the feature extraction method based on filter bank spatio-temporal component discriminant analysis and the feature extraction based on brain network parameters of multi-channel coupling were proposed to achieve the reliable classification of the target and non-target RSVP-EEG signals.Firstly,in terms of the spatio-temporal-frequency variability of individual EEG responses,a feature extraction method of filter bank spatio-temporal component discrimination analysis(FBSCDA)was proposed.Considering the characteristics of the uneven distribution of energy in different rhythms of RSVP-EEG signals,the gamma EEG signals were decomposed into several time and frequency sub-components.The multi-channel signals of each sub-component were filtered by the common spatial pattern algorithm.The linear discriminant analysis algorithm was used to integrate the sub-components respectively,so that the spatio-temporal-frequency characteristics of the gamma rhythm of RSVP-EEG signals was obtained.Secondly,in view of the coupling oscillation of RSVP-EEG in gamma rhythm,the coupling relationship of channels of RSVP-EEG signals on target and non-target stimulation was studied by combining nonlinear analysis and brain network theory.A feature extraction method based on brain network parameters of multi-channel coupling(BNPMC)was proposed.The method constructed brain network for each subject by selecting the optimal coupling feature,and extracted the parameters of brain network as the classification feature.The results show that compared with non-target tasks,the EEG signal induced by target stimulation does not increase in the gamma rhythm of brain network connection,instead,suppresses the synchronization of EEG response in some brain regions.Finally,an image retrieval system based on RSVP-EEG features was designed,and 10 subjects were recruited to test this system and verify the two methods proposed above.To avoid the limitation of classification because of single feature,DS decision fusion theory were introduced to fuse the FBSCDA classification and BNPMC classification on decision layer and make full use of the spatio-temporal-frequency domain and coupling feature information in gamma rhythm.This study compared proposed methods i.e.FBSCDA and DS decision fusion methods,with two art-of-state methods i.e.HDPCA and DCPM by 10-fold cross validation on EEG data collected in the system.The results show that the classification performance of FBSCDA and DS decision fusion is better than the other two methods,especially,the DS decision fusion method which combines coupling information achieves the best classification performance in all subjects.