Graph Convolutional Neural Network for EEG Emotion Recognition

Because emotion recognition is vital in improving human-computer interaction, emotion recognition is becoming increasingly critical in artificial intelligence. Deep learning is widely used to analyze frequency or time domain features in emotion recognition research. Few researchers consider the spatial features of Electroencephalogram (EEG) and the correlation between EEG channels. In this paper, the graph convolutional neural network is used to analyze the emotion of EEG signal. Firstly, the edge and point features of EEG were extracted to construct the graph structure, which made full use of the spatial and frequency features of EEG. Then, the time-series information of EEG was extracted and classified by using the long short-term memory neural network. In the DEAP dataset, the accuracy of the proposed method in the valence and arousal dimension are 83.23% and 85.82%, respectively. The research can be applied to the emotional brain computer interface system.