Research On EEG Emotion Recognition Method Based On Brain Network

Emotion is closely related to people’s cognition,decision making and life state,and its generation is influenced by the information interaction of different brain regions.The study of neural mechanisms of emotions and emotion recognition methods,on the one hand,can help computers have the ability of understanding human emotion,making human-computer interaction more intelligent.On the other hand,it can be applied to the auxiliary diagnosis and treatment of mental illnesses such as depression,etc.However,EEG-based emotion recognition is a complex task,and it is difficult to obtain a better accuracy of emotion recognition by single-dimensional features.Therefore,it is necessary to extract the features that can characterize the emotional state and adopt suitable fusion strategies to maximize the accuracy of emotion recognition.The paper proposed three EEG-based emotion recognition methods by combining brain network with feature extraction method.(1)An EEG emotion recognition method based on brain network and sample entropy is proposed.This method combines global features of brain network with nonlinear local features for identifying valence and arousal emotional states,and achieves high accuracy of emotion recognition on the publicly available dataset DEAP.The comparison of accuracy on different frequency bands indicates that the feature components in the high frequency bands contain more important information for distinguishing emotional states.(2)An EEG emotion recognition method based on bilevel brain network is proposed.The brain network is constructed with minimum spanning tree(MST)and threshold selection(TS),respectively,and the features extracted from the bilevel brain network are fused using the Bayesian fusion(BF)method based on weighted average.The experimental results show that this method can more effectively differentiate between valence and arouse emotional states.Experimental analysis has shown that negative emotions have higher brain network connectivity and faster information exchange rates in the parietal and occipital lobes compared to positive emotions,which provides theoretical support for negative bias,while the frontal and parietal lobes are the main brain regions responsible for emotion processing in the brain.The results have important implications for the analysis of emotion-related brain areas and the development of effective human-computer interaction systems.(3)An EEG emotion recognition method based on the MST brain network and FVMDGAMPE is proposed.On the one hand,the MST-based brain network is constructed on four frequency bands and seven brain network features are extracted.On the other hand,Fast Variational Mode Decomposition(FVMD)and Wavelet Packet Transform are used to decompose EEG data into more refined modes and frequency bands,and genetic algorithms is utilized to optimize the parameter of multi-scale permutation entropy(GAMPE),and finally the brain network features are fused with the non-linear MPE feature.The experimental results show that this method has better emotion recognition accuracy compared with existing studies,which provides a new research idea and perspective for EEG emotion recognition.