Research On Emotion Recognition Of Effective Brain Network Based On Transfer Entropy

Emotion is an advanced cognitive activity.The activation of emotion requires the division of labor and cooperation of different brain regions.In recent years,complex network analysis has been applied to neuroscience and neuroimaging research,and has a better understanding of the brain at the system level.However,considering that EEG signals are caused by the dynamic interaction of complex neural activities,it is necessary to use nonlinear dynamics method to fit the dynamic characteristics of EEG signals.In the research of complex network,the study of emotion based on effective network is still at the development stage,and the research of emotion with directed information transfer is relatively rare.As a model free nonlinear effective connectivity form,transfer entropy can well describe the dynamic characteristics of brain networks.This thesis calculates the transfer entropy of multi-channel EEG signals under emotional activation,analyzes the network characteristics of transfer entropy under different emotional states,and applies the network characteristics to emotion recognition.The main work of this thesis is as follows:(1)Transfer entropy is used to model the complex brain network and quantify the nonlinear activities of the brain in emotional response.Considering the influence of the past information between the EEG signal channels,the state space of the time series is reconstructed according to the embedding dimension and the embedding delay,and the transfer entropy of the multi frequency band is estimated to construct the emotional effective brain network.(2)The network connection modes in different emotional states are analyzed.When the response of the subjects tends to be positive,the causal information transfer of most frequency bands is strengthened.The global attribute,local attribute and network type parameters in different frequency bands are calculated to analyze the characteristics of the complex network of transfer entropy in different emotional states.Then the regional attribute and left-right asymmetry of the brain are discussed according to the in degree and out degree.(3)A classification method based on fusion of network features is proposed.The machine learning classifier using KNN and SVM achieves the highest accuracy of 76.86% in the potency dimension.The research of emotion classification shows that the method of integrating asymmetry,connection coupling and graph theory features can be effectively applied to emotion recognition of EEG.