Complex common spatial patterns on time-frequency decomposed EEG for brain-computer interface

Highlights：

• We propose a novel approach that extracts complex common spatial patterns on time and frequency decomposed EEG signals.

• EEG signal is decomposed into reaction, action, and completion stages; each stage is decomposed into frequency components.

• We show that frequency decomposition not only leads to sharper spatial filters but also generates rich informative features.

• We propose to weight and regularize all CSP features instead of selecting leading CSP features to avoid information loss.

• The algorithm yields a higher classification rate compared to state-of-the-art methods on multiple BCI competition datasets.