Analysis And Application Of EEG Signal Based On Blind Source Separation

Electroencephalogram(EEG)is the main tool to record and reflect brain function.At present,EEG signals are measured usually using the non-invasive method from the scalp,and the gotten scalp EEG observations are the mixture of EEG source signals,physiological artifacts and power frequency interference,etc.In order to extract the expected source components effectively,we used several blind source separation(BSS)algorithms on the scalp EEG observations.The employed BSS algorithms included independent component analysis(ICA)based on improved quantum particle swarm optimization(QPSO),constrained ICA(cICA),sparse component analysis(SCA)and convolutive blind source separation(CBSS).The research objects included event related potential(ERP)extraction,EEG rhythm analysis and artifact removal.In ERP extraction research,we took P300 extraction as the application background and carried out the researches in two parts.In the first part,aiming at the multichannel EEG signals,we put forward an improved-QPSO-based ICA algorithm for the extraction of P300.The proposed algorithm effectively achieves and speeds up the global convergence of ICA through the introduction of more adequate behaviors of learning and competition between particles based on QPSO,and then gets the more independent P300 component.In the second part of ERP extraction research,aiming at the single channel EEG signal,we put forward a method using cICA algorithm for the extraction of P300.We constructed the reference signal according to the P300 average waveform.P300 extraction was effectively accomplished only using single channel EEG signal.The proposed method has loose requirements for EEG lead position,and has the advantage of simple operation.It can achieve the desired recognition accuracy and information transfer rate when applied to the brain-computer interface(BCI).In the study of the EEG rhythm analysis,we took recognition of fear at different levels as the research background and proposed a feature extraction method based on sparse component analysis.In underdetermined case,sparse sources more than scalp EEG leads were extracted,effectively ensuring the EEG information that represented different brain activities could be fully separated.On this basis,the sources related to fear were selected using prior knowledge,and feature vector was constructed for recognition of fear at different levels using EEG rhythm powers.The proposed method can avoid the error caused by analysis on directly the scalp EEG observations to some extent,and lays a good data base for the following research.In the study of artifact removal,we took the ocular artifact removal as the application background and put forward an artifact removal method based on CBSS algorithm.Compared with instantaneous mixture model used by other methods,convolutive mixing model is more consistent with the formation mechanism of the scalp EEG observations.We effectively realized the artifact removal using CBSS method,providing a reference for other applications of EEG and the blind separation of other physiological signals.In this paper,the proposed methods aiming at the corresponding EEG applications were all tested on simulation data,public dataset and our measured data,and were all compared with the same kind of methods.The experimental results show that these methods can achieve the ideal effects.