Maximum entropy based common spatial patterns for motor imagery classification

The common spatial pattern (CSP) is extensively used to extract discriminative feature from raw Electroencephalography (EEG) signals for motor imagery classification. The CSP is a statistical signal processing technique, which relies on sample based covariance matrix estimation to give discriminative information from raw EEG signals. The sample based estimation of covariance matrix becomes a problem when the number of training samples is limited, which causes the performance of CSP based brain computer interface (BCI) to degrade significantly. In this paper, we present a maximum entropy based CSP algorithm that incorporates principle of maximum entropy while estimating the sample based covariance matrix. The proposed algorithm is evaluated on publicly available data set samples. The classification results indicate that the proposed algorithm outperforms the traditional CSP algorithm by 13.38% on average.