Emotion Recognition And Optimization Of EEG Signal Based On RVM

Emotion recognition technology plays a decisive role in the harmonious development of human-computer interaction and the realization of artificial intelligence.At the same time,through the study of human emotional state,human physical and mental health can be protected.The EEG signals have a certain authenticity and reliability because of its producing mechanism and the differences in different emotional states,so it has become the main research object in the field of emotion recognition.However,There are still some difficulties in emotional study of EEG signals because of the randomness and non-stationarity of EEG signals.Therefore,good feature parameters and the sparse,robust and computational complexity of the recognition algorithm are very important for the efficiency and accuracy of the emotion recognition.In this paper,the characteristics of EEG signals and common recognition methods are analyzed.Firstly,the RVM which is similar to the SVM but the model is more sparse is applied to the emotion recognition of EEG signals.Secondly,three nonlinear features of EEG are extracted,and the optimal feature set is selected by fusion and dimension reduction.Finally,the multi-pattern classification algorithm based on RVM is improved to improve the accuracy of emotional EEG classification.The work done is as follows:(1)The structure of emotional EEG recognition system is introduced indetail,in which the commonly used features and recognition networks are emphatically analyzed.In view of the high computational complexity and the limitation of kernel function selection in SVM,considering the similarity between RVM and SVM and the excellent characteristics of RVM itself,the application of RVM in EEG emotional analysis is proposed.The EEG signals collected in the laboratory are used as data,and the performances of RVM and SVM in the recognition of two kinds of emotional EEG signals are compared through experiments.The EEG signals collected in the laboratory are used as data,and the performances of RVM and SVM in the recognition of two kinds of emotional EEG signals are compared through experiments.At the same time,the comparison experiment with BP neural network is added to strengthen the verification of the superiority of the RVM algorithm.The results showed that RVM is more suitable for two classification of emotional EEG.(2)This paper introduced two multi-pattern classification algorithms: OAO and OAA based on RVM.Two multi-pattern classification algorithms based on RVM,OAO and OAA,are introduced in this paper.Considering the nonlinear characteristics of EEG signals and the insufficient representation of emotional Information of EEG signals by a single feature,the power spectral entropy,sample entropy and Hurst index are extracted.Different features are combined and inputted into the OAO-RVM model,and the optimal feature set is selected.The results showed that the recognition effect is the best when the above three nonlinear features were fused and dimensionally reduced.(3)The optimal feature set is used as input vector of OAO-SVM and OAO-RVM,and the emotion recognition ability of the two classifiers for EEG signals was compared.The results showed that the emotion recognition ability of RVM is always better than that of SVM,which proved the validity of RVM in the field of emotion recognition of EEG.(4)In order to solve the problem that large computation and invalid voting affect the final decision in OAO-RVM algorithm,this paper proposed a new two-layer multi-classification model: OAA-OAO-RVM model,which combines the characteristics of OAA and OAO.And experiments were designed to compare the performance of OAO-SVM,OAO-RVM and OAA-OAO-RVM recognition networks for emotional EEG signals.The results showed that the improved model can effectively optimize the classification accuracy of emotional EEG signals,and the recognition performance was significantly better than that of OAO-RVM and OAO-SVM.It was proved that the improved model is an effective multi-classification model.