Feature extraction and classifier evaluation of EEG for imaginary hand movements

Accurate classifying EEG signals for imaginary left-right hand movements is a crucial issue in brain-computer interface (BCI) technology. Aiming to the nonstationarity and nonlinearity of electroencephalogram (EEG) signal, in this work, based on the analysis of EEG time-frequency characteristics by wavelet packet transform and EEG uncertainty analyzing by information entropy, and EEG features for motor imagery from single trial are extracted. Then, the feature data is classified by the support vector machines (SVM), and an optimal search method is proposed to determine the kernel parameter V and penalty parameter C. Finally, some evaluation criteria including mutual information (MI) and misclassification rate (MR) are utilized to evaluate the performance of classifier. The classification accuracy could reach 90%, the MI was 0.65 bit. The test results have shown that the proposed method could accurately extract EEG substantial feature and provide an effective means to classify the motor mental tasks. It can be applied in BCI system of imaging movements.