Optimal-less channel based mental task brain-computer interfaces

To decrease the number of channels of brain-computer interfaces, the optimal-less channel based common spatial pattern (CSP) algorithm is proposed to extract the eigenvalues of the electroencephalography (EEG) features of different mental tasks. First, the temporal-frequency features are represented by event-related (de)synchronization. Then, the separability of each individual channel is measured by entropy criterion. Finally, according to the rank of the separability, the eigenvalues of different channel groups are extracted and classified by the optimal-less channel CSP algorithm and the support vector machine algorithm to obtain the optimal channels. The results demonstrate that during the mental arithmetic task and the spatial rotation task, the EEG signals exhibit significant different powers in central and occipital lobe. The electrodes with the highest separability of all the subjects are located in these two areas. Compared with the traditional signal processing algorithm of EEG, the optimal-less channels based algorithm can reduce the number of the channels to 3.3 and increase the classification accuracy by 5.4%. Therefore, the optimal-less channel based algorithm can reduce the number of channels and improve the performance of the mental task brain-computer interfaces.