Electroencephalogram Emotion Recognition Method Based On Multi-Scale Analysis And Ensemble Tree Model And System Design

With the development of affective computing and emotional robotics,emotional intelligence is gradually integrated into human-computer interaction.The robot can obtain the emotional state of humans through speech,facial expression,gesture,and physiological signals,etc.,wherein the speech,facial expressions,gestures are susceptible to subjective factors,thus they can’t reflect human emotion truly,whereas the electroencephalogram(EEG)signal that records the variations in the cerebral cortical nerve potential can reflect human emotion truly.In this thesis,EEG signals are taken into account for emotion recognition,in which EEG signal preprocessing,feature extraction,feature selection,and emotion classification are studied,from which an EEG emotion recognition method based on multi-scale analysis and ensemble tree model is proposed.Firstly,to solve the problem of the nonlinearity and instability of EEG signals,an EEG feature extraction method based on multi-scale analysis is proposed,including EEG signal feature extraction based on empirical mode decomposition(EMD)and based on variational mode decomposition(VMD),in which the EEG signal is decomposed into multiple components on multiple scales,and the features of each component are extracted.To analyze the influence of the number of features in the feature subset on the emotion recognition results,a feature selection method based on sequence backward selection is presented.The experimental results show that the EEG signal feature extraction method based on multi-scale analysis can obtain higher emotion recognition accuracy than those based on single-scale analysis.Secondly,an ensemble tree model based on Bayesian optimization is introduced for EEG emotion recognition to solve the problems of the complex distribution of feature sets caused by the difference among different people.Xgboost that is an ensemble tree model composed of multiple decision trees is employed to learn and classify features.Furthermore,Bayesian optimization algorithm is used to optimize the model’s super-parameters.Experimental results show that EEG emotion recognition method based on Xgboost achieves better recognition performance.Finally,an EEG emotional recognition system is designed,in which the proposed methods for EEG emotion recognition is applied,in which multi-channel EEG data acquisition,signal preprocessing,feature extraction,and emotion classification are realized in real-time.