Brain Network-based EEG Emotion Classification Research

In recent years,with the continuous maturity of artificial intelligence technology,the use of EEG signals for emotion recognition has become a research hotspot in the field of artificial intelligence.Electroencephalogram(EEG),as an overall response of the physiological activities of brain nerve cells in the cerebral cortex,is often used in various aspects of neuroscience research.Since it can reflect human emotions objectively and accurately,it has a positive effect on the recognition of human emotions.These characteristics have a positive effect on the recognition of human emotions.However,there is always a challenge in the field of EEG emotion recognition,that is,how to improve the recognition rate and generalization ability of emotion classification models.In addition,in research on brain networks,the results of pattern analysis on neuroimaging data show that emotion dimensions and emotion categories can be detected in the activities of the distributed nervous system across the cortex and subcortical regions.In this article,we use EEG signals to construct a brain network and obtain the topological structure of the brain network,so as to analyze the connections and relationships between different nodes,and use the connection differences between nodes as the basis for judging emotional differences.Finally,the deep learning network algorithms are combined to improve the accuracy of EEG signal recognition.The main work of this paper is as follows.In this paper,we first collected the positive and negative emotional EEG signals of the subjects by designing video stimulus materials as the experimental data of this article.Then preprocess the collected EEG signals to remove the noise in EEG signals as the first step to improve the accuracy of classification.Next,no artificial feature extraction is performed on the preprocessed EEG data,but a deep neural network is used to directly perform feature extraction and classification.This paper constructs four models for processing the preprocessed data.The specific process is as follows:1)Input all preprocessed EEG signals into CNN-based and LSTM-based emotion classification models for emotion recognition.Experimental results show that the accuracy of EEG signal classification based on the LSTM model is 83.69%higher than that based on the CNN model.The classification accuracy rate of 80.05%is increased by 3.6%,which shows that the LSTM network has a better effect on signals with strong temporal characteristics such as EEG signals.2)Aiming at some deficiencies of convolutional neural networks in processing EEG signals,the convolutional recurrent neural network architecture CNN-LSTM is proposed and designed.The network sends the EEG signal characteristics extracted by the convolutional neural network to the recurrent neural network,combining the advantages of the two networks(CNN and LSTM).Experimental results show that the algorithm improves the recognition rate to 85.25%.3)It is proposed to construct a real-time brain network of EEG signals to record the changes in the beginning and end of emotions during the signal acquisition process,intercept the EEG signals according to the changes in the brain network,and send the intercepted EEG signal data to CNN-LSTM network model,we call it BN-CNN-LSTM(Brain network-CNN-LSTM,BN-CNN-LSTM).The experimental results show that the EEG signal is intercepted by the brain network,and then sent to the convolutional recurrent neural network for classification,the accuracy rate is increased to 88.32%.Based on the above experimental results,the best emotion recognition effect in this article is the BN-CNN-LSTM model,which has an average classification accuracy of 88.32%among 12 subjects.It shows that the CNN model has a very good effect in signal feature extraction,and the LSTM model can fully learn the timing characteristics of the EEG signal.On this basis,it also proves that the brain network changes have a reference effect on the moment of emotion generation.