Research On Feature Selection And Lead Optimization Methods For Emotion Regulation Based On EEG Motion

Brain-computer interface is a technology that transforms traditional human-computer interaction.Emotional brain-computer interface is an important type of brain-computer interaction.It is expected to provide a quantitative method for the regulation,monitoring or evaluation of emotions.It is potentially important Value.However,the extraction and recognition of emotion-related brain signal features has not been completely resolved,and many challenges are faced.Therefore,this article explores appropriate emotional induction experimental paradigms based on the way that exercise regulates emotions and extracts the time domain and frequency domain of emotion-related EEG signals.,Time-frequency domain and spatial domain,and carry out the feature screening of emotion-related EEG signals,so as to screen out the EEG features closely related to emotions.Finally,SVM,KNN,CNN are used to compare and compare the classification results.The lead optimization of emotion-related EEG signals can screen out channels closely related to emotions.The main research contents of the thesis are as follows:(1)Based on the study of feature selection and lead optimization of EEG regulation of emotions,the experimental paradigm of positive emotion and negative emotion is designed,and the emotion-related EEG signals of subjects induced by pictures are collected,and after the signals are preprocessed,Extract time domain features: statistical features,energy features,power features,instability index;frequency domain features: power spectral density and Hilbert-Huang transform features;time-frequency domain features: STFT and wavelet features;spatial features: DASM,RASM,DCAU features;use SVM,KNN,CNN to classify the extracted features,and compare the classification results.The results show that:frequency domain and time-frequency domain can better carry emotion-related EEG features.(2)On the basis of feature extraction,feature selection and lead optimization are performed.The sampling mutual information selection algorithm performs feature selection on the extracted features,and uses the classification accuracy as an indicator to verify the effectiveness of feature selection.The results show that: using SVM for classification,on the self-acquired data set,the classification accuracy of using MI for feature screening is increased by 2.02%,1.29%,and 0.04%,respectively,compared with the classification accuracy without feature screening;on the SEED data set,The classification accuracy of feature screening using MI is improved respectively than the classification accuracy without feature screening:2.09%,0.43%,3.95%;KNN is used for classification,and on self-collected data sets,the classification accuracy ratio of using MI for feature screening is higher The classification accuracy of the non-feature screening has been improved respectively:3.68%,2.05%,3.7%;on the SEED data set,the classification accuracy of the feature screening using MI is improved respectively than the classification accuracy of the non-feature screening: 1.57%,1.95 %,2.1%;CNN is used for classification.On the self-collected data set,the classification accuracy of feature screening using MI is improved respectively than the classification accuracy without feature screening:3.92%,2.51%,5.41%;in the SEED data set Above,the classification accuracy of feature selection using MI is increased by 1.69%,2.25%,and 2.49%,respectively,compared with the classification accuracy without feature selection,which proves the effectiveness of feature selection.Sampling F-score algorithm for lead optimization,and using EEG topographic map for verification,so as to verify whether the lead selection is correct.