Research On Emotion Recognition Method Based On Attention Mechanism For EEG Signals

Emotion is the attitude and reaction of human beings to objective things.It is also the advanced function of the human brain,which can reflect people’s personality characteristics and emotional characteristics.Emotion recognition plays an important role in the field of mental health and other research.Electroencephalogram(EEG)is a physiological signal of the human brain’s central nervous system,which can objectively and accurately reflect a human’s emotional state.Compared with other physiological signals,EEG signals are immune to external interference and are difficult to disguise,so they can better represent people’s emotions.Therefore,EEG has been widely used in the field of emotion recognition in recent years.By analyzing and processing EEG signals,researchers extract the representative features of EEG signals and conduct emotion recognition.However,there are still some problems in the depth model in the existing studies,such as ignoring the correlation between EEG channels in feature extraction,and the features extracted are not comprehensive and accurate enough.To solve the above problems,two EEG emotion recognition models are proposed in this thesis to improve the effect of EEG emotion classification.The main research contents are as follows:(1)An emotion recognition model based on Graph Convolutional Network(GCN)is proposed.The model uses standardized mutual information entropy to calculate the correlation between channels,and then constructs the feature map.Next,GCN is used for feature extraction,and the channel attention mechanism is introduced to form the final feature vector.Finally,the depth model is used to classify emotions.The experimental results show that the accuracy of this model is higher than that of other models on the two common data sets,which indicates that this model has better performance in emotion recognition.(2)An emotion recognition model based on Transformer and Long-Short Term Memory(LSTM)is proposed.Based on the Transformer model,the proposed model improves the attention mechanism and feed-forward network in the Transformer model.By using the improved Transformer and LSTM to extract the frequency domain features and time domain features of EEG data,the comprehensiveness of EEG feature extraction is improved,and the accuracy of model classification is increased.The proposed model was tested on SEED and Seed-IV datasets to evaluate its performance of the proposed model.The results show that the accuracy of the proposed model reaches 93.31% and 80.56% on two data sets,respectively.