Research on EEG Emotion Recognition Based on Multi-scale Residual Convolutional Neural Network

Emotion recognition based on EEG signals has become an essential method in fields such as human-computer interaction and disease assessment. To further improve the recognition effect of human emotions, this paper fused the spatial and frequency information in EEG signals. It proposed a multi-scale residual convolutional neural network (ECA-Resnet-CNN) with an ECA attention mechanism to perform multi-scale information fusion on the constructed original features. The method first divides the EEG signals into six frequency bands; then, the differential entropy features are extracted from each frequency band and converted into a spatial-band three-dimensional feature matrix; then, the deep information of the three-dimensional feature matrix is extracted through a multi-scale residual convolutional neural network; finally, the weights of different channels of the feature map are adjusted by introducing the ECA attention mechanism, and the emotion classification is completed. The results show that on the Deap dataset, the arousal dimension and valence dimension binary classification accuracy of this method can reach $97.16 \%$ and $\mathbf{9 6. 3 0 \%}$ respectively, which is a significant improvement compared with the existing mainstream emotion recognition models, proving the effectiveness of this method.