Nonlinear analysis for motor imagery EEG based kernel partial least squares

A brain-computer interface (BCI) is a system that should in its ultimate form translate a subject´s intent into a technical control signal without resorting to the classical neuromuscular communication channels. However, electroencephalogram(EEG ) signal is non-stationary signals, linear analysis methods is not well performance for feature extraction of EEG. Nonlinear analysis methods based kernel partial least squares(KPLS) was proposed to use for classification of motor imagery. The coefficients of AR model for C3, C4, Cz electrodes were computed, which were transformed as the independent variables. After numbers of factor extraction was assessed by the analysis for cross-validation, Linear and nonlinear PLS were used for classification of the motor imagery. It shows that the satisfactory results are obtained and high performance of kernel partial least squares compares against linear partial least squares.