Study On Feature Extraction And Recognition Of EEG Based On Motor Imagery Of Left And Right Hand

Brain-computer interface technology based on electroencephalogram is a new interdisciplinary research field,which is successfully linked to many other powerful fields,such as brain science research,neuromedicine and artificial intelligence.The exploration of motor imagery signals has always been the focus of the rapid development of BCI.Opportunities and problems coexist.Given that some disabled people with damaged spinal or peripheral nerves but intact central nervous systems can use their own motion imagination to communicate directly with the outside world through BCI system,the study of EEG is of important practical significance and social value.Therefore,the study of MI EEG is pushed by the general trend of events.The characteristics of nonlinear non-stationary,easily disturbed and strong randomness of EEG signal are always the factors to be considered in signal analysis and processing in the study of BCI motion imagination.From the view of overcoming the low signal-to-noise ratio(SNR)and characterizing and preserving the nonlinear characteristics,the research of MI EEG is improved and optimized in two aspects,respectively.The specific contents of this subject are as follows:(1)In order to improve the denoising performance of nonlinear and non-stationary signals,EEMD threshold denoising and translation invariant algorithm were combined for the first time.TI-EEMD was proposed as an EEG denoising method.EEMD adaptive threshold overcomes the deficiency of wavelet threshold denoising for wavelet base selection in the second segment.The effective combination of translation-invariant algorithms further solves the pseudo-Gibbs phenomenon that occurs after the threshold function is processed.The proposed method is compared with other algorithms by denoising the simulation signal,and the signal-to-noise ratio,root-mean-square-error,Pearson correlation coefficient and maximum peak error are introduced as the quantization indexes of denoising effect,and the feasibility of the improved method is effectively confirmed.Finally,the application of the new theory to the processing of BCI 2003 motion imaginary EEG signal is emphatically discussed,and the applicability and practical value of the core theory are demonstrated again.(2)Taking the phenomenon of ERD/ERS of motor imagery EEG signal and the typical nonlinear and non-stationary characteristics of EEG signal into account,an approach of feature extraction based on BE-PeEn(Band Energy-Permutation Entropy)is proposed from the perspective of energy and complexity,aiming to recognizing EEG signals.Firstly,the validity and feasibility of permutation entropy feature in characterizing the ERD/ERS phenomenon of left and right hand motion imaginary EEG signal are analyzedin advance.And then the BCI 2003 MI EEG signal preprocessed by the improved algorithm is extracted by BE-PeEn.Support vector machine is used to complete the classification.