Research On Emotional EEG Recognition Based On Integration Thinkin

Intelligent recognition of human complex emotional states is the basis for achieving higher levels of human-machine interaction.It is of great significance to research the technical route of analyzing the subject’s biological signals based on artificial intelligence technology to obtain the emotional state.Based on the pattern recognition theory,this paper conducted research on extracting emotional information from electroencephalogram(EEG)signals.Three classification models were designed for emotion recognition of EEG based on the ensemble thought.The base classifier of the proposed model involved a variety of machine learning and deep learning algorithms.The validity of the models was discussed according to different needs of emotion recognition tasks of EEG and different indicators for evaluation.Emotion recognition experiments were performed on the SEED datasets and the self-collected datasets.The work progress mainly included the following five aspects:(1)The characteristics and progress of emotion recognition research of EEG were analyzed and summarized.Based on the experimental results of open source datasets,the important influence of the time extracted conditions of datasets on the experimental results during task design was demonstrated;(2)The proposed “support vector machine-K nearest neighbor(SVM-KNN)”ensemble model solved the multiclass classification problem,and was applied to the rapid recognition task of a small number of emotional EEG samples,with an average generalization accuracy of 67.57% in the three-class emotion classification of 405 samples and 74.63% in the three-class emotion classification of 1000 samples;(3)The proposed “multi-scale convolution network-random forest(MSCNN-RF)”ensemble model combined the recognition ability of random forest on the basis of multi-scale convolution operation,which realized the secondary feature extraction of samples and achieved an average generalization accuracy of 98.74% in the three-class emotion classification of subject-dependence EEG data;(4)The proposed “dilated convolution network-bi-directional long-short term memory networks(Dilated CNN-Bi LSTM)” ensemble model combined the multi-scale spatial feature extraction capability achieved by dilation rate and the time series feature extraction capability achieved by long-term and short-term memory operations.The average generalization accuracy of the three-class cross-time emotion recognition was85.33%(EEG feature sequence input).(5)An emotional EEG self-collection experiment was designed and implemented.The validity of the ensemble model was further verified based on the four-class self-collected EEG datasets.An application software “Emotion Recognition” was designed and developed based on the characteristics of the EEG-based emotion recognition task,which is easy to deploy different algorithms.Finally,on the basis of summarizing different experimental designs and various ensemble strategies,this paper rethinked and prospected the EEG-based emotion recognition work in order to provide reference value for future research.