One step feature extraction and classification with Tikhonov regularization for BCI

The Common Spatial Patterns algorithm is a highly successful feature extraction algorithm used in classification of motor imagery signals in brain computer interface. For short data sets and noise contaminated data sets Tikhonov regularized variant of CSP is better. Both the methods are followed by a feature classification stage, usually Linear Discriminant Analysis. Here a one step feature extraction and classification for Tikhonov regularized common spatial pattern (CSP) is proposed, where the features are automatically learned, selected and combined through a convex optimization problem. The method reduces over fitting and sensitivity to noise contaminated signals.