Research On Feature Extraction And Selection Algorithm Of Emotional EEG

Emotion is the attitude experience produced by the influence of objective things in the interaction between human beings and external things.It affects all aspects of mankind all the time.Therefore,accurate recognition of human emotional state can better help human life,study and work.In terms of emotion recognition,human physiological signals such as electroencephalogram(EEG),electro-oculogram(EOG),and electromyography(EMG)can more accurately reflect people’s true emotions than external behaviors such as facial expressions and language.Among physiological signals,EEG signals are generated by the central nervous system of the human body,and are closely related to the generation and recognition of emotions.Therefore,the analysis of human emotions through EEG signals has become a current research hotspot.EEG signals are non-stationary signals with prominent frequency domains and large energy gaps in frequency bands.This thesis mainly analyzes EEG signals from three aspects: time domain,frequency domain,and time-frequency domain,researches and extracts better features for emotion classification.The main research contents and results are summarized as follows:(1)Based on the power spectrum intensity with good frequency band balance ability,this thesis proposes the characteristics of balanced power spectrum intensity.The balanced power spectrum intensity feature has excellent recognition ability,and is the frequency domain feature with the best classification performance among the features studied.At the same time,compared with the complex time-frequency domain characteristics of the processing process,the balanced power spectrum intensity can achieve the same classification ability,and it has obvious advantages in terms of time efficiency.(2)The amount of EEG signal data collected by the EEG acquisition device is large,and the signals of some brain parts have no obvious correlation with the emotional state.Directly adopting all EEG data for emotion analysis and recognition will prolong the training of the learning model.And the judgment time and classification effect.In addition,when processing EEG data,traditional feature selection algorithms have shortcomings in dimensionality reduction.Some algorithms will significantly reduce the accuracy rate while significantly reducing the dimensionality.Some algorithms can maintain or even improve the accuracy rate,but the dimensionality reduction effect is very poor.Based on this,this thesis proposes a feature fusion algorithm FFS.FFS combines the two methods of Relief and m RMR,combines the advantages of the two types of selection algorithms,and obtains new fusion features that take into account the dimensionality reduction effect and the recognition accuracy.Select,and finally obtain a feature vector that can fully express the emotional state of the EEG signal.(3)This thesis conducted experiments on the DEAP dataset.Using the balanced power spectrum intensity as the experimental feature,and using the feature fusion algorithm FFS for feature extraction and feature selection,the feature dimension is reduced from 160 to 67,with a dimensionality reduction effect of up to 58%,and the support vector machine algorithm is used for classification,The classification accuracy rate on the V(non-pleasurepleasure)dimension reaches 88.89%,and the classification accuracy rate on the A(inactiveactive)dimension reaches 87.73%.The experimental results show that the feature fusion algorithm FFS has achieved good results regardless of the recognition performance or the dimensionality reduction effect,which is better than the existing results.The research results of this thesis provide references and references for the study of feature extraction and feature selection of emotional EEG.The proposed feature fusion algorithm framework has good scalability and can be extended to other applications in the field of machine learning.