Improving the efficiency of an EEG-based brain computer interface using Filter Bank Common Spatial Pattern

Brain Computer Interface (BCI) systems are popular due to their ability to improve the quality of life of disabled people. They are proper tools to translate the movement intentions of human into control's signals for out devices. We use Filter Bank Common Spatial Pattern (FBCSP) which leads to extracting discriminative spatial patterns of Electroencephalogram (EEG) signals. In the FBCSP algorithm, the EEG signals are decomposed into multiple frequency bands using a filter bank. In order to exploit the time-varying feature of EEG signals and extract the unknown features, we partitioned the EEG signals into a number of time windows. Then, CSP features are extracted from all time windows of all frequency bands. Finally, feature selection and classification has been done using Mutual Information (MI) algorithm and Naive Bayesian classifier, respectively. We used EEG dataset BCI Competition IV dataset 2a to evaluate the proposed method. This dataset is a four-class two-session motor imagery dataset. The experimental results show that the mean kappa value has improved to 0.60.