High Resolution Common Spatial Frequency Filters for Classifying Multi-class EEG

Abstract

The Common Spatial Patterns (CSP) algorithm is a highly successful spatial filtering method for extracting spatial patterns related to specific mental tasks from electroencephalogram (EEG) signals. The performance of CSP highly depends on the selection of frequency band in the preprocess. However, the most discriminative frequency band features varies slightly with subjects and mental tasks. In order to provide high resolution in frequency domain, we propose an common spatial frequency patterns method to learn most discriminative spatial and frequency filters simultaneously for specific mental task. The results on EEG data during motor imagery (MI) tasks demonstrate the good performance of our method with decreased number of EEG channels.