A WPSD-Based Feature Extraction Method of EEG Signal for Motor Imagination

In order to improve the performance of the Brain-Computer Interface (BCI) system, it is important to extract the feature vector which reflects nonlinearity and non-stationary of Electroencephalogram (EEG) signals. This paper proposed an adaptive feature extraction method for BCI system based on multiclass motor imagery EEG signals. The proposed method extracts a feature vector related with time-frequency-spatial information of multiclass EEG signals from the continuous wavelet transform power spectral density (CWT-PSD) difference weighted by weight matrices which reflect the time-varying of EEG signals, in some channels which reflect well multiclass motor imagery. This method was evaluated on the EEG signals of BCI Competition IV Datasets
$2\mathrm{a}$
using SVM as classifier, and the performance results were compared with the results obtained by the Common Spatial Pattern (CSP) and Filter Band CSP (FBCSP) feature extraction method used Naïve Bayesian Parzen Window (NBPW) classifier. The average classification accuracy of the proposed method is 90.22% for the 4-class motor imagery based BCI, and the kappa coefficient of the proposed method is higher 1.38 times and 1.21 times than one of the CSP and FBCSP method, respectively.