Analysis And Quantitative Evaluation On The Effect Of Artifacts For Motor Imagery Brain-Computer Interface

Brain-Computer Interface(BCI)technology can establish a transmission channel between the brain and external devices,so as to realize the exchange of information between the brain and the outside world,and to realize the control of external devices.This technology has broad application prospects in the fields of medical rehabilitation,psychological state monitoring and military,etc.BCI based on scalp electroencephalograph(EEG-BCI)is a non-invasive BCI implementation mode.Compared with implanted BCI,EEG-BCI has better applicability.However,the scalp EEG is extremely susceptible to various artifacts during the acquisition process,and the resulting low-quality EEG training samples will have varying degrees of negative effect on the performance of EEG-BCI.In the existing data,researchers have proposed many EEG artifact removal methods,but the effect of different types of artifact and its quantitative analysis on the EEG-BCI system are rarely reported.This thesis focuses on the realization of this specific application of the motor imagery BCI(MIBCI)system.It performs a more comprehensive analysis on the detection and classification methods of artifact and the effect of low-quality training samples on BCI system performance.The specific work are as follows:(1)Introduce the three-class motor imagery EEG(MI-EEG)and its analysis methods.According to the source and type,the artifacts present in the EEG signals were studied.The principles of Independent Component Analysis(ICA)and Common Spatial Pattern(CSP)spatial filtering methods are introduced,and the application of the two methods in artifact analysis is compared.(2)A method for recording the number of occurrences of artifacts based on independent component analysis is proposed.After ICA separation is performed on the single trial of the original EEG signal,the separated artifacts are identified and analyzed according to the characteristics of time domain,frequency domain or space domain.The occurrences of conventional artifacts such as electrooculogram(EOG),power-line burst,motion and electromyography artifact,and Electrode interference were recorded and statistically analyzed.The statistical results under the data of 600 single trials containing 8 subjects showed that the EOG artifacts,electrode interference and power-line burst occurred more frequently,with 596,218 and 243 occurrences respectively.The motion and EMG artifacts appeared 153 times.Among them,the occurrence of power-line burst and electrode interference is related to the equipment and the experimental environment,and is random;while EOG artifacts,motion and EMG artifacts occur less in the motor imagery data segment,and appear more in the non-motor imagery data segment frequently.(3)A method for analyzing and quantifying the effects of artifacts based on blind source separation and spatial filtering techniques is proposed.The anti-interference ability of filters designed based on ICA and CSP under different artifacts is also analyzed and compared.Quantitative results show that although EOG artifacts occur more often,about 77% of EOG artifacts have almost no effect on ICA-BCI,and 85% have almost no effect on CSP-BCI.The other artifacts have a slightly greater effect on the two systems.Among them,the effect of electrode interference is the most obvious.About two-thirds of the electrode interference has a certain effect on both two systems.In addition,the ICA-BCI system is slightly more affected by power-line burst and EMG artifacts than CSP-BCI.Therefore,in the design of BCI systems using different spatial filtering methods,according to the degree of influence of artifacts,the artifacts with greater effect on the system can be removed first,and the artifacts with less effect can be removed afterwards,thereby improving the efficiency of preprocessing work.The method of quantifying the effect of artifacts proposed in this thesis can be used to avoid artifacts during EEG signal acquisition,and provide a reference for artifact removal.It also has certain reference value for the research of EEG quality assessment.