Research On Emotional Feature Extraction And Classification Based On EEG

Human emotions not only include psychological reactions and physiological reactions,but also reflect people's own needs and subjective attitudes.Studies have shown that the generation or activity of human emotions is highly correlated with the activity of the cerebral cortex,which provides a basis for studying emotional classification through EEG signals.With the rapid development of brain science and in-depth study of various disciplines,emotional recognition through EEG signals has gradually become a hot topic.However,the influence of differences in EEG signals induced by different subjects at different times on sentiment classification,and how to extract effective EEG emotional features to ensure higher accuracy and better robustness of emotion recognition are still an urgent problem to be solved in EEG emotional classification research.In response to these problems,In response to these problems,this paper has carried out the following research:(1)This paper studied how to reduce the influence of day difference and fluctuation of EEG signal on the performance of emotion classification,and to improve the robustness and accuracy of emotion classification based on EEG.The data set used was to collect EEG data of 12 subjects for 5 consecutive days.This data set facilitates the study of the effects of fluctuations and differences between EEG signals induced by the same subject on different days on sentencing.Two methods were used for the study.One method was to process the EEG dataset using data space adaptive and common spatial patterns algorithms.Firstly,the data space adaptive algorithm and the common spatial patterns algorithm were used to process EEG data sets successively,so that the difference between the EEG signals collected on different days was minimized,and the difference between the classes was maximized.Finally,the power spectral density characteristics,differential side and differential causal characteristics of EEG signals were extracted.Another method was to use the common spatial patterns combined with the wavelet packet decomposition algorithm to process the EEG data set.Firstly,the EEG emotion data was used to find the optimal subspace by using the common spatial patterns algorithm.Then the EEG emotion signal was decomposed by time-frequency domain using wavelet packet decomposition algorithm,and finally the wavelet packet energy characteristics were extracted.For the features extracted by these two methods,Bagging tree,support vector machine,linear discriminant analysis and Bayesian linear discriminant analysis algorithm were used to classify emotions.The experimental results of the two methods show that the common spatial patterns combined with the wavelet packet decomposition EEG emotion classification can alleviate the impact of EEG daytime differences on sentiment classification,and the classification accuracy is as high as 86.20%,which improves the effectiveness of classification performance.(2)This paper studied how to improve the robustness and efficiency of emotion classification of EEG signals by multi-feature extraction and combination method.Data set 2 that was from the DEAP data set was used.It not only contained EEG signals,but also contains peripheral physiological signals,which contain a large amount of data,so as to study the multi-feature extraction and feature combination methods to improve the robustness and efficiency of EEG emotional classification.EEG emotion data were analyzed by empirical mode decomposition and wavelet packet decomposition.Then,the average energy,fluctuation index,sample entropy,approximate entropy,multi-scale permutation entropy and Hurst exponent were extracted from the processed data.These feature sets were emotionally classified using four classification algorithms.According to the accuracy of each feature classification,it was analyzed which features can better distinguish EEG emotional signals,selected features for feature combination,and classified the combined feature sets to find the best performance combination.It can be seen from the experimental results that after the EEG signal was decomposed by wavelet packet,the feature combination of the sample entropy feature and the approximate entropy feature can better characterize the main features of the EEG emotion signal,and the classification accuracy is as high as 95.54%.