Research On Rhythm Analysis Method Of Motor Imagery EEG

Motor imagery electroencephalogram(EEG) is the EEG which is evoked during theimagination of limb movement but without any actual physical action. EEG based onmotor imagery has the characteristics of the event-related synchronization and theevented-related desynchronization. These characteristics could be analysed to judge theintention of the imagination, so as to control the external equipment and realize specificpurpose without any limb movement. The brain-computer interface (BCI) based on motorimagery EEG is developing rapidly a hot research field in nowadays, which brings greatapplication value in medical treatment, entertainment, brain cognitive etc.This thesis is a research of the analysis method in the motor imagery rhythms. Thefeature extraction and classification algorithms of motor imagery EEG are analyzed anddiscussed, meanwhile the Hibert-Huang Transform (HHT) and complexity analysis arefocused on in the application of feature extraction. This paper introduces the basic theoryand realization method of the HHT and complexity by which the BCI competition datafeature is extracted. At first, the movement imagery EEG data after preprocessed isdecomposed with empirical mode decomposition (EMD) into a series of intrinsic modefunctions (IMFs). Then the low frequency IMFs are removed, and the rest of IMFs areconducted by Hilbert transform to get Hilbert marginal spectrum. The marginal spectrumdifference between the channel C3 and channel C4 were selected as the original featureswhich will be then decreased the dimensions by the principal components analysis (PCA)so as to be jointed with EEG complexity to construct the feature vector. Finally, the BPneural network and support vector machine are designed to classify the BCI competitiondata of lefthand and righthand. After the classifier parameters optimization and adjustment,the classification accuracy is up to 87% and 86% respectively, which are comparablydesired results.HHT was proposed for analysis of the non-stationary signals, and the Hilbertspectrum got from it which contains time domain and frequency domain information issuitable for analysis of EEG with nonlinear and non-stationarity. The results proved the feasibility and validity of the feature extraction method on EEG analysis in this paper.