Decoding of motor intention from Brain EEG signal via local spatial sparse pattern

In the brain-computer interfaces (BCI) systems of motor imagery, the spatial pattern on global EEG channels were commonly used to identify the motor intention from EEG signal. However, some channels are more important than the others in motor imagery tasks. Thus, the spatial pattern from global EEG channels can not reflect the difference of channels. To enhance the classification performance of motor imagery system, we proposed local sparse common spatial pattern (CSP) method for addressing the problem of channel’s difference frequently arising in BCIs. It constructs local channels based on Euclidean distance and performs joint diagonalization on each local channels to obtain multiple local spatial features. Lastly, we used the group sparse model to select discriminative features from different channels. Experimental evaluations on motor imagery dataset show that the proposed algorithm has higher classification performance than the competing methods.