A Spatial Attention-based Graph Convolutional Network for EEG-based Motor Imagery Classification

Brain-computer interface technology is now widely used in rehabilitating patients with physical movement disorders. However, the limitation of the decoding ability of motor imagery EEG signals still hinders the development of brain-computer interface technology. In this paper, we propose a spatial attention-based graph convolutional network (SAGCN) for motor imagery EEG signal classification. First, we initialize the adjacency matrix based on the functional connectivity of the raw EEG signals and extract the DE features of the original EEG signals as input. Subsequently, the channels that are of greater significance to the motor imagery EEG signals are identified through the spatial attention mechanism. Finally, the graph convolutional network is combined with the spatial attention matrix to obtain the classification results of EEG signals. We conduct experiments on two public datasets and the experimental results show that our proposed model outperforms current state-of-the-art methods.