Research On Emotion Recognition Of EEG Signals Based On Machine Learning

Emotion is closely related to people’s lives,and the broad application prospect of emotion recognition is focused on in many fields.EEG can reflect the information of brain cognitive activity,and emotion recognition using EEG signals has become a new research hot spot.On the one hand,electroencephalogram(EEG)contains a lot of information closely related to emotion.On the other hand,compared with speech and facial expressions,EEG is not easy to disguise,true and reliable,which makes it show greater advantages in the field of emotion recognition.This thesis focuses on emotional recognition of EEG signals.Based on SEED emotional EEG data set,the spatial correlation feature information of EEG signals is extracted,and emotion classification is carried out by machine learning.Improve the depth residual network model to improve the accuracy of emotion recognition and model performance.On this basis,an emotion recognition system based on EEG signals is developed.The specific research contents include:(1)Extracting the spatial correlation information of EEG signals for machine learning emotion recognition.Aiming at the problems of traditional emotional feature extraction,such as feature redundancy and lack of EEG time-space correlation feature information,Pearson correlation coefficient of EEG signals with different leads is calculated;the wavelet transform is used to obtain the wavelet coherence coefficient between EEG of different leads;the Hilbert transform is used to extract the instantaneous phase and phase synchronization index of EEG in each lead.This thesis explores the corresponding relationship between the spatial correlation characteristics of EEG and different emotional states,and designs a support vector machine classifier to effectively classify positive,negative and neutral emotional states.The simulation results show that the spatial correlation features of EEG are effective in emotion recognition,and the average recognition accuracy can reach 91.5%.The average classification accuracy of 93.7% can be obtained by Pearson correlation coefficient of EEG differential entropy for emotion recognition.(2)An improved model of EEG emotion recognition based on deep residual convolution network with attention mechanism is proposed to classify positive,neutral and negative emotions.By selecting residual blocks to avoid the degradation of deep neural network,the influence of residual network structure at different depths on the model performance is studied,and spatial attention,channel attention and CBAM are added respectively to extract EEG features that are more related to emotion and improve the accuracy of emotion recognition.The experimental results show that the average recognition accuracy of positive,negative and neutral emotions can reach 95.49% by using the deep residual network model.Adding the residual network model of spatial attention,channel attention and CBAM,the accuracy of emotion classification is increased by 1.85%,1.78% and 3.03% respectively,which shows that the improved deep residual network model can effectively improve the accuracy of EEG emotion recognition.(3)Develop the emotion recognition system of EEG signals.Design an experimental paradigm of emotion-evoked,and use EEG equipment to collect video-induced emotional EEG signals.Based on Py Qt5 platform in Pycharm,the system operation interface is designed by using Designer plug-in,and the system functions are designed by using MATLAB and Python software to realize EEG signal preprocessing,feature extraction and emotion classification.The system can record EEG measurement,save and call the past detection results,and display the user interface,providing convenient and quick user emotion analysis.The research results of this thesis have important theoretical and application value in the fields of emotional brain-computer interaction,medical health and emotional robots.