Joint selection of time and frequency segments for classifying multiclass EEG data in motor imagery based BCIs

The time and frequency intervals of EEG signals have crucial influence on the classification performance of motor imagery based brain-computer interfaces (BCIs). This paper proposes a novel algorithm for joint selection of time segment and frequency band for multiclass motor imagery based BCIs. A task-independent sliding window method, which does not require any prior knowledge, is used for the choice. Approximation joint diagonalization (AJD) based multiclass common spatial pattern (CSP) algorithm is utilized for feature extraction and k-nearest neighbor (KNN) is used for classification. The algorithm was tested and compared with three typical methods on a four-class motor imagery data set The experimental results suggested that based on the chosen time segment and frequency band, the proposed algorithm achieved superior performance with regard to classification accuracy.