Classification Of Motor Imagery EEG Signals Based On Sparse Representation

Motor Imagery(MI)is one of the hot topics in Brain Computer Interface(BCI)research field in recent years.Recently,the classification and recognition of EEG signals for motor imagery have need high stability and high accuracy.In this paper,motor imagery EEG signals were used as the research object,to effectively improve the classification accuracy of left or right hand motor imagery,and use the Sparse Representation(SR)method for multi-domain and multi-channel EEG signal feature extraction and fusion.And based on the individual differences of the subjects,time periods weighted linear discriminant analysis(LDA)classification was used for classification.The main work of this paper is as follows:(1)According to the EEG signals being multi-domain and the disadvantages of single-domain features,this paper proposed the multi-domain feature extraction method for motor imagery based on sparse representation.This method used sparse representation to fuse multi-domain features and remove the redundancy between features of different domains.And the method could well extracted features using the datasets of brain-computer interface competition in 2008.(2)According to the EEG signals being multi-channel signals,this paper proposed the multi-channel feature fusion method combined with the Joint Sparse Model(JSM).This method can consider its spatial position information when fusing energy features of multi-channel.And then time-weighted LDA classification was used for classification and recognition of left or right hand motor imagery.And the method could well improve classification accuracy using the datasets of brain-computer interface competition in 2008.(3)According to the theoretical research method proposed in the paper,EEG signals was collected using offline EEG signals collected by the equipment named Neuracle.And after offline analyzing,the correctness and feasibility of the proposed method were proved.