Emotion Recognition Based On EEG By Multivariate Pattern Analysis

At present,it is difficult to define emotion exactly,since a lot of factors affect emotion.However,emotional computing is one of the key technologies to achieve HCI.Some earlier signals about emotion recognition are non-physiological signals.Computer vision techniques to recognize non-physiological affective features,such as facial expressions,gestures,and voice.These external features are easily disguised and this can easily lead to unreliable results.In contrast,various physiological indicators obtained from physiological responses,such as EEG,skin electrical reactions,blood circulation,and respiratory activity are not disguised and are stable.At present,emotion recognition technology is widely used in business,polygraph,security and other fields.Thus,it is a very worthy study on identifying emotion effectively.The main work of this paper is as follows:1.The method of network is widely used in the pattern recognition,polygraph,magnetic resonance and so on.On the basis of the existing power spectrum research,firstly propose the emotion recognition method of network analysis.Secondly,we propose a method which combines the power spectrum feature and the network feature to improve the accuracy of classification.We use the power spectrum under the five bands(theta,slow alpha,alpha,beta,gamma)as a feature classify three emotions(the positive,neutral,negative)and find that the classification of beta and gamma frequency bands are more accurate,and they are 62.8% and 64.2% respectively.And then,we extract the network attributes under the five frequency bands as the characteristics of the emotional classification.The results also show that the classification of beta and gamma frequency bands are more accurate,and they are 56% and 67% respectively.It shows that we can use the network analysis method to study the emotion recognition.Finally,we combine the power spectrum characteristics and network characteristics and classify further the accuracy rates in the beta and gamma bands are 63.3% and 68.2% respectively,which are higher than the rate of power spectrum and network.It illustrates that the combination of different types of features can improve the classification accuracy.2.With the rapid development of artificial intelligence,machine learning has become a hot research.As an important branch of machine learning,a superior in-depth learning plays a significant role,builds a learning network and shows a superior performance of classification.Convolutional neural network(CNN)is a relatively mature depth-learning depth learning model.At present,CNN is widely used in image recognition and speech recognition.CNN has been used in EEG.In this paper,we obtain the accuracy rate of 75.2% by using CNN model to study the emotion recognition in EEG signals.The accuracy of most subjects is higher than that of power spectrum and network attributes and the classification accuracy of fusion features.The average classification accuracy ratio is 11% higher than the accuracy of power spectrum,and the average classification accuracy ratio is 8.2% higher than the network feature.What is more,the average classification accuracy ratio is 7% higher than the ratio of fusion feature.Therefore all results indicate that CNN can be used for emotional recognition and can get a better classification effect.