L21-Norm-Based Common Spatial Pattern with Regularized Filters

Common spatial pattern (CSP) is one of the most popular feature extraction techniques that performs well in the field of brain-computer interface (BCI). However, the classical CSP based on the squared Frobenious norm is known to be sensitive to noise and prone to overfitting. In recent years, efforts have been made to solve the problem of robust and sparse modeling. Therefore, we describe the formulation of a CSP criterion by using the L21-norm and add the L21-norm regularization term. The regularized version of the L21-norm-based common spatial pattern is termed RCSP-L21 in this paper. An iterative algorithm is designed to obtain the spatial filters of the regularized objective function. Experimental results on EEG dataset IIa of BCI Competition IV verify the effectiveness of our new method.