A study of emotion recognition methods incorporating functional brain network features and self-attention mechanisms

Aiming at the problem that a single feature is insufficient to characterize the emotional information in the emotion classification of EEG data, while the fusion of multiple features will lead to information redundancy, an emotion classification method that fuses brain functional network features and self-attention mechanism is proposed. The method introduces the median centrality feature (BC) and clustering coefficient feature (CC) to enhance the spatial information representation of brain regions; and uses the self-attention mechanism to establish a feature fusion model, through which the hidden relationship within the feature sequence is increased, which not only highlights the information coupling between the channels but also weakens the interfering information; and then uses the SVM classifier to classify the feature vectors for emotion classification, which is validated in the DEAP dataset. Validation. The experimental results show that the accuracy of the multi-feature fusion method in sentiment classification through self-attention reaches 85.13%. Compared with previous feature fusion methods, this method has better classification effect, proving its effectiveness and feasibility.