A Personalized Channel Selection and Spatial filtering Model for Brain-Computer Interface

Abstract

Brain-computer interface (BCI) systems are new human-computer interaction technology, and the electroencephalography (EEG) signals can be translated as the control commands. For more operational dimensions, a hybrid experimental paradigm with motor imagery and speech imagery has been proposed in our previous study. To improve the practicality of BCIs, a personalized channel selection and spatial filtering model is proposed in this paper. Correlated channels are chosen by Pearson's correlation coefficient, and spatial filters are obtained by common spatial pattern (CSP) from these channels. The features of EEG signals are extracted and classified by the spatial filters and support vector machine (SVM), respectively. The average classification accuracy of ten subjects is 73.9%, and it is 2.1% higher than the accuracy without channel selection. Suitable channels can reduce the complexity of BCIs, and the classification results of EEG are also improved.