Multi-band Functional Connectivity Features Fusion Using Multi-stream GCN for EEG Biometric Identification

At present, due to the gradual increase in the demand for security, the identification technology based on EEG functional connectivity (FC) features is getting more and more research. It has been demonstrated that the FC features of different frequency bands have different classification capabilities, which implies that FCs in different frequency bands contain different identity information. However, the integrated use of FC features in different frequency bands in the identification system is still an urgent problem to be solved. In this paper, we propose a multi-stream GCN (MSGCN) network structure that simultaneously utilizes different frequency bands of FC. The multi-band FC feature is extracted through the multi-input network structure and then integrated with the later network. We evaluated the proposed MSGCN model on a classical public dataset and compared it with state-of-the-art methods. The results show that the proposed MSGCN achieves a classification accuracy of 98.05% on unprocessed data, which is an improvement of 3–7% over the existing optimal methods. We also discuss the sensitivity of the proposed model to the number of channels and demonstrate that our MSGCN model has better applicability for practical deployment.