Skip to Main Content
The spatial weights for electrodes called common spatial pattern (CSP) are known to be effective in EEG signal classification for motor imagery-based brain-computer interface (MI-BCI). To achieve accurate classification in CSP, it is necessary to find frequency bands that relate to brain activities associated with BCI tasks. Several methods that determine such a set of frequency bands have been proposed. However, the existing methods cannot find the multiple frequency bands by using only learning data. To address this problem, we propose discriminative filter bank CSP (DFBCSP) that designs finite impulse response filters and the associated spatial weights by optimizing an objective function which is a natural extension of that of CSP. The optimization is conducted by sequentially and alternatively solving subproblems into which the original problem is divided. By experiments, it is shown that DFBCSP can effectively extract discriminative features for MI-BCI. Moreover, experimental results exhibit that DFBCSP can detect and extract the bands related to brain activities of motor imagery.