EEG Emotion Recognition Based on Multi-View and Multi-Scale Fusion Network

In recent years, emotion recognition based on electroencephalogram (EEG) has become a prominent research hotspot. However, EEG signals are spatially discrete and nonstationary, posing challenges in representing spatiotemporal information and extracting more discriminative features from complex signals. This paper proposes a feature fusion model based on a multi-view Transformer and multi-scale dynamic convolution. The multi-view Transformer model is designed to extract spatial and frequency features. In contrast, the multi-scale dynamic convolution model captures temporal and frequency features, followed by the fusion of these spatiotemporal-frequency features. We conducted subject-dependent and subject-independent experiments on the DEAP, SEED, and SEED-IV datasets, and the experimental results validate the model's effectiveness.