Abstract: The classification of mental states based on motor imagery(MI) electroencephalograph(EEG) signal was investigated. All the states were classified according to the phenomenon of event-related synchronization and event-related desynchronization. The band power of the MI EEG signal was extracted as the input feature and then classified by using LS-SVM. The final classification accuracy is 92%， which shows that LS-SVM performs well for the classification of the band power feature of MI EEG signal. And compared to the standard SVM， the performance of LS-SVM is as good as that of the standard SVM， but has some advantage in computing time.

Key words: EEG(electroencephalograph), motor imagery, band power, least square, support vector machine(SVM)