Multiple Feature Extraction of Electroencephalograph Signal for Motor Imagery Classification through Bispectral Analysis

Electroencephalograph (EEG) signals associated with motor imagery (MI) are highly non-Gaussian, non-stationary and have non- linear characteristics. Bispectral analysis is an advanced signal processing technique that quantifies quadratic non-linearities (phase-coupling) among the components of a signal and holds promise for characterizing MI-related EEG. Studies have been reported on the applicability of bispectrum for MI classification; often with different choice of high order spectra features. Question remains as to which of the different features of non-linear interactions over frequency components are best suited for MI classification. In this paper, an analysis based on bispectrum is reported to extract multiple high order spectra features of EEG for MI classification. MI signals from C3 and C4 channels for two tasks are used in the analysis. Based on bispectrum analysis, four high order spectra features are extracted. The classification results indicate that the extracted features could differentiate the two MI tasks with an accuracy of 90±4.71%.