Study On Emotion Recognition And Channel Optimal Selection Based On EEG

Emotion and affective computing have been integrated with many disciplines such as psychology,physiology,behavioral science,and information science.There are also many ways to recognize emotions,and emotion recognition can be performed through facial expression features,speech,body characteristics,and various physiological signal indicators.Since most of these reference factors are measured by external manifestations of the body,it is impossible to accurately and truly observe human emotions.Therefore,more and more researches conduct emotion recognition research with Electroencephalogram(EEG)signal characteristics.EEG signals can measure the internal activity of the brain and have a high temporal resolution.Considering the advantages of EEG,this paper studies the method of emotion recognition based on EEG,and considers the requirement of portable devices in practical use,and analyses the method of channel optimization.The main work is as follows.(1)Based on the emotion-induced experiment of video,four experimental paradigms of emotional stimulation were designed and EEG emotional data were collected.After preprocessing the EEG data,11 features of the three domains of time domain,frequency domain and nonlinear domain were extracted.(2)Support vector machine(SVM),extreme learning machine(ELM),and K-nearest neighbor(KNN)algorithm are used to compare their performance in emotion classification accuracy.After selecting the optimal features,the optimal feature and differential entropy are studied in different rhythms.It is proved that the γ-band and β-band has the greatest contribution to emotion recognition.It is found that support vector machine performs better in emotion recognition and is chosen as the classifier of channel selection.(3)Considering the portability requirement of emotional recognition equipment,the top 10 channels with the largest weight are extracted based on Relief feature selection algorithm and optimal feature combination,and emotional recognition research is carried out on the top 7 channels.Because of the difference of individualization,the optimal channels differ greatly among different subjects.In order to solve this problem,this paper uses the method of weights addition to get the common effective channel,and uses 2st as the feature vector.The average accuracy of four emotional valence binary classification of the first 7 channels is 94.696%.And the distribution of common channel brain area is consistent with that of brain area related to emotional physiological mechanism.(4)Based on the selected features,seven channels of FP1,FP2,F3,F4,F7,F8 and FZ with fixed prefrontal lobe were studied to explore the classification accuracy of single feature and feature combination in emotion recognition.Considering the limitation of the processing ability of portable devices,seven channels of EEG signals are filtered into beta band,and 2st is used as feature vector.The average accuracy of emotional potency classification is 88.548%.It provides theoretical support for the subsequent emotional recognition portable devices.This paper completes the work of emotion recognition and channel optimization selection,and achieved some results.