EEG Individual Identification Using Multi-Dimensional Attention Graph Convolutional Network

Currently, the field of individual identification utilizing coded modulation visual evoked potentials (cVEP) is gaining significant attention. However, existing methods face challenges due to the EEG signals' low signal-to-noise ratio and inter-session variability among participants. To address these issues, we introduce a Multi-Dimensional Attention Graph Convolutional Network (MDAGCN) tailored for identifying cVEP biomarkers in personal identification tasks. Our approach first employs a Temporal-Attention Enhanced Convolutional Neural Network (TEEGNet) to extract time-frequency domain features, which are subsequently integrated with functional connectivity and spatial distance matrices to construct functional connectivity and spatial distance brain networks. Subsequently, a spatial attention mechanism coupled with graph convolution is utilized to extract deep features from these networks. The MDAGCN model is optimized by minimizing the categorical cross-entropy loss, ensuring an optimal fit to the data. Experimental evaluation on a cVEP EEG dataset comprising 20 subjects validates the model's efficacy, achieving a cross-session recognition accuracy of 72.17%, surpassing benchmark methods. Our novel MDAGCN framework presents a promising avenue for cVEP biomarker individual identification.