EEG-Based Emotion Recognition Using Dynamic Spatial-Spectral-Temporal Network

EEG-based emotion recognition has demonstrated significant potential in various domains, for example, emotion detection of patients with disorders of consciousness, cognitive load assessment, emotion detection of drivers and so on. In the realm of brain-computer interface research and development, the ability to accurately and effectively recognize emotions is of significant importance. In this paper, we propose a multi-channel EEG emotion recognition method called dynamic spatial-spectral-temporal network (DSSTNet). It comprises three components: (1) Spatial features extractor consists of a graph converter and a graph convolutional network. It first converts the EEG signal into graph-structured data. Then, the graph convolutional network is used to dynamically optimize the adjacency matrix to obtain the spatial features between the channels. (2) Band attention module composed of semi-global pooling, local cross-band interaction, and adaptive weighting for further extracting spectral information. (3) Temporal features extractor that extracts deep temporal information by stacking multiple one-dimensional convolution layers. The results of comparative experiments show that the classification accuracy of the proposed method is better than that of a series of comparison methods, and the robustness is relatively strong. The results of ablation experiments validate the effectiveness of each module proposed in the method.