In this paper, a novel method based on Hilbert-Huang transform (HHT) is presented to select optimal timefrequency patterns for single-trial motor imagery electroencephalograph (EEG). The method comprises three progressive steps: 1) employ Empirical Mode Decomposition (EMD) method to decompose EEG signal into a superposition of components or functions called IMFs, and then apply Hilbert transform to the IMFs to calculate the instantaneous frequency and instantaneous amplitude; 2) select the IMFs including the most useful frequency components 3) the optimal timefrequency patterns can be selected according to the instantaneous frequency and instantaneous amplitude of the selected IMFs. After selecting the optimal timefrequency patterns, the features extracted by different methods are classified by Fisher linear discriminator. The results showed that the proposed method could improve the classification accuracy.
Learning Common Time-Frequency-Spatial Patterns for Motor Imagery Classification