EEG Emotion Recognition Based on Horizontal-Vertical Spatial Structure and Timing Modeling

In the field of brain-computer interface, emotion recognition research is of great importance. We propose a novel electroencephalography (EEG) emotion recognition method based on the fusion of graph convolution and temporal convolution, named STSG-Net (Spectral-Temporal and Spatial Graph fusion Network). First, the relative power spectral density (rPSD) features of EEG signals are extracted using the spectral-temporal convolution module to effectively capture the localized characteristics in the frequency domain. Subsequently, based on the international standard 10-20 electrode system, the horizontal and vertical connected EEG structures were constructed, the spatial neighborhood relationship was modeled using graph convolution network (GCN), respectively, and the horizontal and vertical spatial feature representations were integrated through the feature fusion mechanism. Finally, the fused spatial features are input to the fully connected layer for emotional state regression. We conducted extensive experiments on the MAHNOB-HCI dataset, and the results show that the proposed method outperforms existing mainstream methods in terms of root mean square error (RMSE), Pearson's correlation coefficient (PCC), and consistency correlation coefficient (CCC), which verifies the effectiveness of the strategy of combining spatial structural modeling with temporal dynamics modeling in the task of EEG emotion recognition.