Emotion recognition based on dynamical graph convolutional neural networks and BiLSTM
Due to the spatial dependencies among electrode channels evolving over time during the entire process of emotion occurrence,this paper proposes a model for emotion recognition based on dynamic graph con-volutional neural network-bidirectional long short-term memory(DGCNN-BiLSTM).Firstly,DGCNN dynamical-ly learns the connections between different electrode channels by training the neural network,thereby dynamical-ly updating and optimizing the adjacency matrix.Secondly,BiLSTM can learn the temporal correlations of fea-ture sequences,thereby enhancing the network's ability for emotion recognition.Experimental results on the SEED dataset and DEAP dataset show that the model achieves the highest average accuracy of 92.03％and the highest accuracy of 96.56％for arousal dimension and 95.22％for valence dimension,respectively.The results indicate that the model is beneficial for improving emotion recognition accuracy,and compared with other meth-ods,there is also an improvement in emotion classification accuracy to varying degrees.
万方数据
维普
