Research And Application Of EEG Emotion Based On Deep Learning

As an advanced function of the human brain,emotion has a great impact on people’s mental health and personality characteristics.Electroencephalography(EEG)can non-invasively collect induced emotional EEG signals,and then use EEG.The classification of signals can provide further theoretical and practical basis for real-time monitoring of the emotions of normal people or depressed patients in the future.In order to obtain an efficient and stable EEG signal emotion classification method,this paper starts with the deep learning algorithm,using Differential Asymmetry(DASM),Differential Caudality(DCAU)and Differential Entropy(DE)features as the cut-in Point,explored the EEG signal emotion classification method based on Back Propagation Neural Network(BPNN)and Convolutional Neural Network(CNN),and designed two network classification models based on emotion EEG signal,Implemented a method to apply the trained model to We Chat applet for sentiment classification.The main work and results of this paper are as follows:(1)Aiming at the problem that the traditional machine learning method has strict requirements on feature engineering and the parameters are not easy to adjust,an emotion recognition method based on the BPNN model is proposed.The DASM and DCAU features of the EEG signal were selected to input into the BPNN model respectively,and the validity and stability of the model were verified.For the public EEG dataset,the experimental results show that the accuracy of EEG features extracted based on DASM reaches 80.11%,and the precision rate,recall rate and F1 score are 80.06%,80.18% and 80.12%,respectively.The accuracy of EEG features extracted based on DCAU reached 86.27%,and the precision,recall and F1 scores were 87.55%,87.06%and 87.30%,respectively.In addition,in order to determine the batch size and epoch with better effect,a comparative experiment was carried out for the two features using different batch sizes and epochs.Finally,the performance of the proposed BPNN model is compared with the k-Nearest Neighbor(k NN)machine learning algorithm.The experimental results show that the proposed BPNN model has better classification accuracy and running time than k NN.Performance.(2)There is still room for improvement in the classification effect of the BPNN model.Using the differential entropy features extracted from the public EEG emotion dataset,a CNN-based EEG emotion classification model is designed.The CNN model includes 4 convolutional layers,4max pooling layers,2 fully connected layers and 1 Softmax layer,and adopts batch normalization to ease the parameter search problem and suppress model overfitting.The experimental results show that the average accuracy rate of the three kinds of emotion recognition on the SEED dataset using this model reaches 98.73%,and the precision rate,recall rate and F1 score are 99.69%,98.12% and 98.86%,respectively.The area under the Receiver Operating Characteristic(ROC)curve was 0.998.In order to verify the effectiveness of the method,a comparison experiment was done with traditional machine learning methods,such as decision tree,k NN and Support Vector Machine(SVM).The experimental results show that the method has a high emotion recognition rate.Compared with recent similar work,the proposed CNN structure has certain advantages for EEG emotion classification.(3)In order to improve the convenience of EEG emotion analysis,an application method is proposed to apply the trained model to the We Chat applet.The above-mentioned well-trained CNN model is applied to the We Chat applet through Tensor Flow.js,which realizes that the user uploads the EEG signal in the specified format,the server obtains the uploaded EEG signal,calls the CNN model for classification,and finally feeds back the subject.The function of positive,neutral or negative emotions,with better human-computer interaction experience.In addition,the functions of user login,uploading EEG signals,testing records,uploading records,and server deployment are designed and implemented.