Research On EEG Emotion Recognition Based On Brain Functional Network And Common Spatial Pattern

Electroencephalogram(EEG)signals record the changes of electric waves generated by the interaction between brain neurons,which contains a lot of physiological and emotional information.Researchers often achieve human-computer interaction by extracting and classifying the different features of EEG signals in different emotional states.It is widely used in BCI such as clinical medicine and engineering application.EEG brain functional network analysis method based on complex network and graph model is used to effectively perceive the coordination and separation mechanism of human brain when dealing with different emotional states.It is of great significance to study the functional changes of human brain.However,the existing EEG brain function network mainly estimates the connection relationship of the network by observing the sparsity range between electrodes,and extracts some information of network attributes by manual setting,which is easy to cause information loss and difficult to obtain high recognition performance.To solve the above problems,this thesis proposed an EEG Emotion Classification based on Common Spatial Patterns of Brain Networks Topology(EEC-CSP-BNT).The main research contents are as follows:1.An EEC-CSP-BNT model is proposed.Use mutual information to calculate the functional connection matrix of the EEG signal,and directly spatially filter the matrix on the common spatial pattern algorithm to obtain different types of emotional features of the EEG signal.Different from the traditional brain function network feature extraction method,this method can effectively reduce the errors caused by network threshold setting and network attribute selection,and has strong feasibility and development prospects.2.Using Elastic Net constraints to perform feature selection,realize EEG emotion recognition based on multi-band information fusion.Through the Elastic Net constraint algorithm,the brain network features and differential entropy features of multiple sub-bands are further integrated to realize the automatic selection and classification of features.Compared with the EEC-CSP-BNT model,this method has an average accurary of 70.8% and 80.5% on the valence dimension of the DEAP and SEED data,respectively,which have improved 4.6% and 3.8%,respectively;The average recognition rate is 74.5%on the DEAP data arousal dimension,which have improved 5.6%.3.Laboratory data collection and analysis.The EEG data collection experiment related to emotion recognition is designed using laboratory EEG equipment,and the above algorithm is used to conduct experiments on EEG data collected in the laboratory.The experimental results show that the average accuracy rate of Valence is 71.3±2.3(%).