Investigation of methods for extracting features related to motor imagery and resting states in EEG-based BCI system

Methods for extracting features of motor imagery from 1-channel bipolar EEG were evaluated. The EEG power spectrums which were used as feature vectors were calculated with filter bank, FFT and AR model, and were then classified by linear discriminant analysis (LDA) to discriminate motor imagery and resting states. It was shown that the extraction method using AR model gave the best result with the average true positive rate of 83% (σ = 7%). Furthermore, when principal component analysis (PCA) was applied to the feature vectors, the dimension of the feature vectors could be reduced without decreasing accuracy of discrimination.