The Research Of Eeg Emotion Recognition Based On The Emotional Lateralization Mechanism With The Left And Right Hemispheres

In the process of human cognitive activities,emotion has an important impact in feelings,judgment,and decision of human.And effective emotional interaction will give humans a more pleasant interaction experience.External behavior and internal physiological signals are the important methods to understand the emotional state of human.Due to it is easily controlled by human subjective consciousness,external behavior is often deceptive and misleading,which makes emotion recognition research based on physiological signals captures more and more attention from researchers.The basic problems of occurring,rousing,and affecting emotion related to physiological signals are still uncertain.Hence,how to give machine brain-like intelligence and realize more natural human-computer interaction is a problem to be solved in the field of affective computing.Comparing with other physiological signals,electroencephalogram(EEG)records the electrical activity of subject brain neurons in real time.It makes EEG become an important method to realize emotion recognition.At present,a large number of EEG emotion recognition methods based on machine learning have been proposed,but there are still some problems to be solved to further improve the performance.The first one is that most of methods the cognitive law of human brain.They only construct the emotion model from EEG data level,or excessively rely on handcraft features.The above methods may loss some emotional information in EEG,which hindering the emotional intelligence system to further capture the delicate emotion changes.And the second is that due to the conduction effect of brain,one-channel EEG signal is the linear combination of several potential source signals,which leads to the highly correlated but redundant features in multi-channel EEG data.These redundant features do nothing to improve the performance of classifier,but easily cause the over-fitting of the classifier.To solve the above problems,this paper focus on the cognitive law of human brain,and model the hemispheric emotional lateralization mechanism(the different responses of the left and right hemispheres to emotional stimuli)for EEG emotion recognition.The main research contents of this paper include:(1)For the first problem,this paper proposes a bi-hemisphere discrepancy convolutional neural network model(Bi DCNN)for EEG emotion recognition,in which three different EEG representations is constructed based on the 10-20 system to simulate the emotional lateralization finding.Then the differential convolutional neural network with a three-input single-output network structure is designed to amplify the lateralization information for EEG emotion recognition.Different from previous methods based on handcraft features,this work starts from the cognitive law of emotional lateralization to construct the lateralization features using raw EEG signals.Then the classifier is used to extract temporal and spatial features from these features for emotion recognition.(2)Fort the second problem,this paper a channel selection method named emotional lateralization-inspired spatiotemporal neural networks(ELSTNN)for EEG emotion recognition.In ELSTNN,the highly correlated but redundant channels is removed by spectral clustering,meanwhile the emotion lateralization information is retained.After that,extracting the temporal and spatial features from the selected channels is done by the classifier for EEG emotion recognition.Specifically,in order to remove the highly correlated but redundant channels,unsupervised spectral clustering is used to cluster the channels in the left and right hemispheres,respectively.Then,the clustering results of all samples is combined and the channels appeared in different groups is selected.Finally,the temporal and spatial features of the selected channel are mined using the spatio-temporal neural network model for emotion recognition.Facing the challenge of EEG emotion recognition,this paper uses the mechanism of left-right hemispheric emotion lateralization to model the cognitive law of human brain,so as to realize emotion recognition.It includes EEG feature extraction,channel selection and the final classifier design combined with the cognitive principles.All experiments are carried out in the public dataset,DEAP.And the experimental results show the feasibility and effectiveness of cognitive principles in EEG emotion recognition.