Research On EEG Feature Selection And Channel Optimization Algorithm For Emotion Recognitio

Emotion recognition refers to the process in which a computer analyzes and processes the collected signals to obtain the current emotional state of the subject.It is an important research topic in the field of Affective computing,which is widely used in health medical treatment,daily social communication and other fields.In recent years,with the rapid development of Brain-Computer interaction,emotion recognition based on Electroencephalogram(EEG)has become the focus of research.However,there are still several problems in data processing and wearable portability,such as incomplete collection,high feature dimension and redundant channels.Thus,the dissertation will carry out research from the aspects of EEG acquisition and preprocessing,multi-domain feature extraction and selection,channel optimization.The main research work is listed as follows:(1)EEG acquisition and preprocessing.With visual and audio stimuli,EEG signals of happiness,calmness and sadness emotions are selected in this research.Then,the EEG signal is preprocessed to reduce the noise mixed.The 50 Hz notch filter is used to remove the power frequency interference and the Ensemble Empirical Mode Decomposition(EEMD)is used to remove the eye artifact.(2)The Discrete Wavelet Transform(DWT)and the CART-Adaboost algorithm are proposed for the multi-domain feature extraction and selection of EEG signals.Aiming at solving the problem that a single category of EEG features contains less emotional information,the dissertation proposes a way that extracting multi-domain features such as time domain,frequency domain and nonlinear dynamics.From the perspective of EEG detail,DWT algorithm is used to analyze the role of high frequency and low frequency multi-domain features,a total of 24 features and their normalization processing.Aiming at the problems of strong feature correlation and high model complexity,CART-Adaboost integration strategy is proposed for feature selection.The Gini coefficient is used to calculate the significance of a single feature,and different feature subsets is combined according to its importance to build the CART-Adaboost classification model.Compared with before feature selection,the emotion classification accuracy is improved by9.46%,and obtained the optimal fusion features,the classification accuracy is as high as87.7%.At the same time,the optimized CART-Adaboost classification model proposed in the dissertation is compared with random forest,K-nearest neighbor and support vector machine classifier,and it is found that its classification performance is much better than the other three kinds of category.(3)Aiming at the problems of channel redundancy and fuzzy emotional brain region,the dissertation proposes an optimization algorithm of EEG channels based on Relief F-FGSBS.Firstly,the Relief F algorithm is used to assign weights to each channel;Secondly,FGSBS is sorted according to the weights of channels;Finally,the classification accuracy is used as the evaluation standard to completing channel optimization.When the number of channels reaches 12,the classification effect is best,and the next sequentially best 12 channels are: T8,P3,F2,FP1,O2,PO8,FPz,FC4,Oz,F3,PO7 and PO4;By locking in the brain region where emotions are located,it is concluded that emotional channels are mainly distributed in the frontal and occipital regions.