Emotion Feature Extraction And Classification Based On EEG Signal

Emotion is an essential part of human physiological performance.At the same time,the research of emotion recognition is becoming more and more concerned by researchers in the field of artificial intelligence.Human brain activity changes reflect the emotional changes in many people,so a lot of emotional recognition studies are based on the EEG,how to get used to the interpretation accurately and effectively is a hot issue in the world on this field.This paper is mainly studied emotional EEG signal processing,feature extraction and classification problem and using EEG online public sentiment database(DEAP database)to establish two-dimensional emotion classification model,divided the emotions into pressure and calm two states.The main contents:1.The method of feature extraction based on EEG emotion recognition is studied.The characteristics of EEG are usually studied from three aspects: time domain,frequency domain and time frequency domain.The characteristics of EEG are mostly manifested in the frequency domain characteristics of signals.In this paper,7 electrodes in the forehead region of the brain are selected as the research objects.The EEG signals are divided into five frequency bands by using the DB4 wavelet decomposition and reconstruction algorithm,and the asymmetric features,approximate entropy and permutation entropy are extracted on each frequency band respectively.The thesis analyzes and compares the accuracy of each feature and combination feature in each frequency band using SVM classification.The experimental results show that using the combined features in this paper to classify emotions has a good classification effect,and draw the conclusions that the key frequency bands of EEG signals are in the ?-band.2.The classification algorithm of emotions,the total sample of this extract is 270,belong to a small sample of species,we classified the effect analysis to the feature extracted by K nearest neighbor algorithm and support vector machine,the experiment results show that the support vector machine algorithm more exactly than K nearest neighbor classification,kernel function support vector machine experiments selected for RBF kernel function.Then by using genetic algorithm and other optimization algorithms to optimize the parameters of SVM,comparing the optimization results through experiments can be drawn using genetic algorithm to optimize the parameters of support vector machine classification is better than the effect of several other optimization algorithms,the average recognition rate reached 83.88%.Most of the researchers who use the same database in this paper only use the entropy feature method to extract features,and use SVM classification,the recognition rate is mostly between 70% and 80%.Based on this,this paper combines the asymmetry of the brain,asymmetric features,approximate entropy and permutation entropy,and optimizes the SVM parameters by using the optimization algorithm,which improves the classification effect of emotion recognition significantly.