Abstract:In brain-computer interface applications, such as neural prosthesis and wheelchair control, the accurate motor task classification is an important basis for the efficient operation of the system. Meanwhile, magnetoencephalography as a brain neuroimaging method with higher spatial resolution is attracting more and more attention. Comparing to traditional brain signal classification schemes which usually require delicate design of feature extraction, emerging deep learning methods represented by convolutional neural network provide another choice without specific feature extraction. In this paper, a CNN-based motor task classification scheme in magnetoencephalography is designed, which can automatically learn the spatial and temporal features of magnetoencephalography layer by layer and carry out feature combination. The scheme has the advantages of simple network architecture and fast processing speed. The results of within-subject and cross-subject experiments on BCI competition dataset and ConnectomeDB dataset show that the classification accuracy and generalization performance of the proposed scheme are superior to traditional schemes based on features extraction.

A Novel Motor Task Classification Scheme in Magnetoencephalography Based on CNN