Research On EEG Emotion Classification Based On Convolutional Neural Network

Human-computer interaction is an important part of artificial intelligence.In recent years,with the continuous development of computer technology,the way of human-computer interaction has also been expanding.Because of its high authenticity,meeting the needs of people with disabilities and other factors,brain-computer interface(BCI)has been widespread concern in many areas.In order to achieve human-computer natural interaction,we must give the computer the ability to recognize people's emotions.Therefore,it is very important to recognize people's emotions from the BCI.Research on EEG has matured in recent years and many methods have been applied to emotion recognition based on EEG.However,as an emerging technology,convolutional neural network(CNN)is still relatively rare in the application of EEG.This paper explores the application of CNN in emotion recognition based on EEG.The main work of this paper is as follows:In this paper,we firstly designed the emotion stimulation materials,collected the positive and negative EEG samples,and then used the methods of bad block rejection,ocular artifact reduction,artifact rejection and filter to pre-process the EEG in order to better extract EEG features and classify them.Next,we classified the processed EEG data using five models:WT-SVM,WT-CNN,CSP-VAR-SVM,CSP-VAR-CNN and CSP-CNN.Among them,WT-SVM and WT-CNN use wavelet transform to extract EEG features,and then classify by using support vector machine and CNN respectively.CSP-VAR-SVM and CSP-VAR-CNN use the Common Spatial Pattern to reduce dimension,and then select the standardized variance as features,and finally use SVM and CNN to classify.CSP-CNN model use the public space model for EEG dimension reduction,then the CNN is used to extract and classify the dimension reduction data.In this paper,when using WT-SVM model for classification,we compared the four features after wavelet transform and When classifying by CSP-VAR-SVM model,we chose the appropriate number of dimension reduction.After classifying EEG with five kinds of classification models,we compared the classification results of these classification models and found that:1)When we classify the emotional characteristics,the classification effect of convolution neural network is similar to that of support vector machines based on Gaussian radial basis function,but the convolution neural network is slightly better.For example,the average classification accuracy of WT-CNN model is 86.90%,which is 0.39%higher than that of WT-SVM model.The classification accuracy of CSP-VAR-CNN model is 69.84%,which is 0.79%higher than CSP-VAR-SVM.2)The average classification accuracy of CSP-CNN model is 80.56%,which is 11.51%higher than CSP-VAR-SVM model and 10.72%higher than CSP-VAR-CNN model.This result shows that CNN can be directly used to EEG feature extraction and classification.It improved significantly compared with the standardized variance.3)Among the five classification models in this paper,WT-CNN model has the best classification effect,and the average classification accuracy of the six participants is 86.90%.