Emotion Recognition Based On EEG Signal

Emotion is an innate attribute of human beings.It can be reflected by expressions,actions,and language,but it is disguised due to the influence of human control.EEG signals directly reflect changes in brain activity,and are closely related to emotions,and cannot be disguised.Therefore,the use of EEG signals for emotion recognition research has great prospects.Based on the emotional EEG data set DEAP,this paper realizes the classification of different emotional states through baseline processing and data enhancement of EEG signals.The main research work of this paper is as follows:(1)Aiming at the problem that the role of baseline data in the DEAP data set is always neglected,an EEG signal preprocessing algorithm based on de-baseline features is proposed.Wavelet packet transform(WPT)was used to separate the frequency band of EEG signal,and the Pearson correlation coefficient(PCC)of baseline data band was extracted as the EEG feature under the condition of heartless fluctuation,and the Pearson correlation coefficient of experimental data band was extracted as the feature with emotion.The feature matrix of the experimental data is subtracted from the feature matrix of the baseline data to obtain the difference value of the EEG emotion feature,and the difference value is input into the convolutional neural network as a new feature for classification.The experimental results show that compared to the scheme of discarding the baseline data and directly using the experimental data,the new method WPT-PCC improves the accuracy by 4% on average.(2)Aiming at the problem of insufficient emotional EEG data and imbalance of emotional category samples,a data enhancement method based on generative confrontation network is proposed.On the basis of Wasserstein GAN-gradient penalty(WGAN-GP),tag information is added as the input of the generator to generate data of the specified tag,and sequence backward selection(SBS)is introduced to evaluate the quality of the generated samples.Keep samples with high recognition accuracy and add them to the training set.By comparing other different evaluation indicators,the sample generated by CWGAN-GP-SBS improves the accuracy by 2%,which proves the effectiveness of the proposed CWGAN-GP-SBS method in data enhancement.