Research On Emotion State Recognition Of Eeg Signal Based On Channel Graph Fusion

In recent years,the field of human-computer interaction represented by braincomputer interfaces has achieved rapid development,and it is an important part of artificial intelligence technology and neuroscience.A machine with good emotional interaction capabilities can make correct responses based on the subject’s brain electrical signals.Realizing the emotional state recognition of EEG signals and digging deeply into the emotional information contained in EEG signals can help enhance machine intelligence and have important scientific research value and practical application prospects.Existing methods mainly use convolutional neural network and frequency domain graph convolutional neural network methods to recognize the emotional state of the EEG signals.The emotion classification method based on convolutional neural network needs to determine the structure of the EEG signal multi-channel in advance,but there is no such explicit structure for the EEG signal multi-channel;the emotion classification method based on frequency domain graph convolutional neural network has defects due to the insufficient depth of the graph structure.To this end,this article studies the EEG signal emotional state recognition method,from two aspects,the EEG signal channel graph fusion method and the self-attention mechanism for EEG signals,achieved good emotional classification performance,while satisfying the disorder and individual differences of EEG channels.This article first proposes an EEG signal emotional state recognition method based on the EEG signal channel partial graph.This method uses the k-nearest neighbor algorithm to construct a local graph for each EEG channel and calculates the edge feature vectors of all neighbor channels in the neighborhood through a shared multilayer perceptron,and then aggregates all edge feature vectors through a weighted adaptive pooling layer and updates the channel feature vector.The experimental results on the SEED and DREAMER datasets show that the emotional state recognition method based on the channel partial graph has a significant performance improvement compared with the baseline method.At the same time,it can make good use of the potential channel structure and stack multiple local graphs to explore more channel neighborhood information.Considering the lack of global graph information in the channel local graph,it is impossible to consider the lack of individual differences in EEG signals.In this paper,the feature fusion of the channel local graph and the global graph is carried out,and the self-attention mechanism is introduced to learn individual differences.Finally,an emotional state recognition method based on the channel graph fusion method and the self-attention mechanism is proposed.This method introduces the channel global graph,treats each EEG channel as a node on the global graph,and updates the feature vectors of all channels through the Chebyshev graph convolution method,and then fuses with the channel local graph information.In addition,considering the individual differences of EEG signals,a self-attention mechanism for EEG signal channels and the frequency bands are introduced.By learning the differences of signal channels and frequency bands,the self-attention mechanism can consider the individual differences of EEG signals.The experimental results show that the classification accuracy rate of this method on the DREAMER data set for valence,arousal and dominance increased by 3.86%、2.48%、3.21%,respectively,and the accuracy rate for sentiment classification on the SEED data set increased by 0.54%,reaching the highest performance currently known.