EEG single-trial classification of different motor imagery tasks using measures of dispersion and power in frequency bands

摘要

This paper proposes a novel combination of features for the four types of wrist movement discrimination (extension, flexion, pronation and supination) on left and right wrist respectively. The features are based on dispersion measures (Inter Quartile Range (IQR) and Median Absolute Deviation (MAD)), entropy and band power in the EEG signal. These features are input to the RBF classifier to test classification accuracy. The classification rate was up to 92% with an average of over 90% in the four subjects. The reduced computational intricacy and the resulting acceleration in speed obtained were other hallmarks of this method. These results show further improvement in recognition rate when compared with the groups' earlier effort on same database (Khan and Sepulveda, 2010; Hubais et al., 2006).