Classification of Motor Imagery EEG Based on a Time-Frequency Analysis and Second-Order Blind Identification

In this paper, two methods for the classification of motor imagery electroencephalogram (EEG) recordings in a brain computer interface (BCI) task was described. The first one is based on a time-frequency analysis of EEG signals. The original EEG signals are converted to time-frequency signals by a function of short time Fourier transforms (STFTs). In another method, we applied second-order blind identification (SOBI), a blind source separation (BSS) algorithm to preprocess EEG data. Subsequently in both of two methods, Fisher class separability criterion was used to select the features. Finally, classification of Motor Imagery EEG evoked by a sequence of randomly mixed left and right motor imagery was performed by a linear classifier or back-propagation neural networks (BPNN), using as inputs the two STFTs time- frequency signals or the two SOBI-recovered SI components or the two EEG channels C3/C4. The results showed that classification accuracy of Motor Imagery EEG was significantly improved by STFTs or SOBI preprocessing.