A treatment of EEG data by underdetermined blind source separation for motor imagery classification

Brain-Computer Interfaces (BCI) controlled through imagined movements cannot work properly without a correct classification of EEG signals. The difficulty of this problem consists in low signal-to-noise ratio, because EEG may contain strong signal components that are not related to motor imagery. In this paper, these artifact components are to be suppressed using a recently proposed underdetermined blind source separation method and a novel MMSE beamformer. We use these tools to remove unwanted components of EEG to increase the classification accuracy of the BCI system. In our experiments with several datasets, the classification is improved by up to 10%.