Research On Emotion Recognition Method Based On Improved Reservoir Computing And Transfer Learning

Emotion,which significantly affects people’s behavior,is the overall expression of consciousness.Emotion recognition is the key for computer to have the ability of expression,which makes artificial intelligence to move towards emotional intelligence.Emotion recognition,as the research hotspot in the field of artificial intelligence,has important research significance.Electroencephalogram(EEG)is the main basis for emotion recognition because it is not easy to camouflage,is realistic and objective,and has high real time resolution.At present,the classification of EEG by neural networks is the main method of emotion recognition.However,the traditional emotion recognition methods based on neural networks have not considered the characteristics of correlation,timing and nonlinearity of EEG.In addition,the emotion recognition methods based on neural networks are limited by the quantity and quality of EEG samples,and the quantity of EEG samples is often insufficient and the quality of EEG samples is often irregular.In view of the above problems,this thesis proposes an emotion recognition method based on improved reservoir computing and transfer learning.The main works of this thesis are as follows:(1)In view of the characteristics of correlation,timing and nonlinearity of EEG,an emotion recognition method based on improved reservoir computing is proposed.First,an EEG feature extraction algorithm based on reservoir computing with brain network is proposed to take advantage of the correlation of EEG.Second,an EEG feature enhancement algorithm based on ridge regression is proposed to take advantage of the timing of EEG.Finally,an EEG feature classification algorithm based on multilayer perception is proposed for the nonlinearity of EEG.Experimental results on DEAP database shows that,compared with traditional methods,the proposed method can comprehensively consider the characteristics of correlation,timing and nonlinearity of EEG,and improve the accuracy of emotion recognition.(2)In view of the problems of insufficient quantity and low quality of EEG samples,an emotion recognition method based on improved transfer learning is proposed,on the basis of the emotion recognition method based on improved reservoir computing.First,an EEG sample selection algorithm based on average Frechet distance is proposed to improve the sample quality.Second,an EEG sample feature transfer algorithm based on transfer component analysis is proposed to enlarge the sample quantity.Experimental results on DEAP database shows that,compared with the emotion recognition methods based on other transfer learning algorithms,the proposed method can enlarge the quantity of high-quality EEG samples and further improve the accuracy of emotion recognition.