Regularized CSP with Fisher's criterion to improve classification of single-trial ERPs for BCI

摘要

A brain-computer interface (BCI) based on the combination of oddball paradigm and face perception has been introduced. Such BCI mainly exploits three event-related potential (ERP) components, namely vertex positive potential (VPP), N170 and P300 instead of only P300. With different temporal and spatial distributions of the three ERP components, a regularized common spatial pattern (CSP) with Fisher's criterion (FC), named FCCSP, is proposed to extract the most discriminative features for single trial classification of ERP components. With linear discriminant analysis (LDA) classifier, the proposed FCCSP spatial filtering method yields an average classification accuracy of 95.4% on seven healthy subjects for single-trial ERP components, which outperforms no spatial filtering, the CSP and the FC.