Comparison of three motor imagery EEG signal processing methods

Abstract

Feature extraction and classification of EEG signals is core issues on EEG-based brain computer interface (BCI). Motor imagery EEG signals can be difficult to classification because EEG sensor signals are mixtures of effective signals and noise, which has low signal-to-noise ratio. So signal processing methods should be used to improve classification performance. In this paper, three methods were used to process motor imagery EEG data respectively, and the Fisher class separability criterion was used to extract features. Finally, classification of Motor Imagery EEG evoked by a sequence of randomly mixed left and right image stimulations was performed by multilayer back-propagation neural networks (BPNN). The results showed that using of the three methods significantly improved classification accuracy of Motor Imagery EEG, and SOBI method had done a best job in this situation. © 2011 Springer-Verlag Berlin Heidelberg.