EEG Emotion Recognition Based on Graph Signal and Stable Learning

In order to extract effective information from Electroencephalography (EEG) data, this paper proposes an EEG emotion recognition method based on graph signal and stable learning. To simulate the integration of information by the brain, the EEG data is endowed with spatial structure, that is, it is modeled as a graph signal, and the graph Fourier transform is used to transform the initial graph signal into a graph signal in the frequency domain, and then the graph convolution neural network is used to automatically extract the more discriminating features of the data with graph structure for emotion classification. In addition, in order to extract the causal features in EEG channels, we used the weight learning of decorrelation to remove confounding factors and exclude correlation features. The proposed method is tested on two public datasets for binary valence and arousal, and the classification accuracy is better than other similar methods.