Emotion Recognition Based On Deep Learning Of A Few EEC Signals

Emotion is a complex psychological phenomenon,which is closely related to the survival of human beings as a high function of human brain.With the continuous improvement of brain imaging technology and the gradual maturity of machine learning algorithms,electroencephalogram(EEG)has become a popular choice in emotion recognition research,due to its advantages such as high temporal resolution,not easy to be controlled by subjective consciousness and strong correlation with emotional state.In order to satisfy the real application scenario,explore the emotional recognition is convenient and real-time systems become a problem to be solved,my research group has done part of the work,by the F-score selection algorithm and support vector machine(SVM)classification algorithm,selected a combination of electrodes from 64 EEG electrode channels,the classification accuracy rate of the filtered five channels was 81.1523%,which were FT7,T7,FC4,TP 10,O1.In order to verify the effectiveness of the five EEG channels in the study of emotion recognition,a five EEG channels based on emotion recognition study was constructed on the basis of channel optimization.The research is divided into three stages:The first stage,based on the emotional images evoked paradigm,induce 50 male subjects to produce Low Valence Low Arousal(LVLA),High Valence Low Arousal(HVLA),Low Valence High Arousal(LVHA),High Valence High Arousal(HVHA)four types of different emotions and use SAM to assess the emotional arousal effects,at the same time,through the five EEG channels(FT7,T7,FC4,TP10,O1)wireless EEG acquisition equipment acquisition EEG signals.The second stage,under the different emotional states of EEG data preprocessing,wavelet packet transform 8 layer decomposition of EEG data,extract before 21 nodes energy and the mean,standard deviation and 8 order autoregressive parameters as features,through Keras package on R,use TensorFlow,with Feed-Forward Deep Neural Network(FFDNN)training to build emotional state prediction model.In the third stage,the basic functions of the emotion recognition system are designed based on the existing research basis,and the prototype of its presentation interface is designed and displayed.The main results of the study are as follows:(1)The results of SAM self-evaluation in the emotional induction stage showed that the four kinds of emotions had significant differences in the scores of Valence and Arousal degree(P<0.001),among which the Low Valence High Arousal emotion(LVHA)had the best effect.(2)For the four categories of emotion recognition problems,the average accuracy of the emotion classification model constructed in this study is about 61%.During the training,the optimal parameter combination is as follows:the number of cells in hidden layer 1 is 100,the number of cells in hidden layer 2 is 70,the number of epoch is 250,and the sample size used for each iteration is 400.Therefore,this study is based on a few EEG channels combination,wireless EEG signal acquisition and deep learning modeling to carry out the research on the recognition of four categories of emotions,and five EEG channels achieve the recognition accuracy rate of 61%in the research on the recognition of emotions.This is a new attempt to realize real-time emotion recognition.Although it did not achieve the similar prediction effect compared with the prediction model established in the previous experiments,the analysis of the reasons for the differences provided some practical reference significance for the follow-up research.