Research On Affective Eeg Recognition Method Based On Graph Convolutional Neural Network

In recent years,with the rapid development of machine learning and deep learning,automatic emotion recognition based on EEG signals has attracted more and more attention.EEG has been widely used in most EEG interface applications and researches due to its low cost and high temporal and spatial resolution.At the same time graph convolutional neural network as a popular neural network to extract graph domain information has been more and more widely used.At present,eeg emotion recognition based on graph convolutional neural network model has the following problems and proposed solutions to these problems :(1)simply using the simplest graph convolutional neural network can not make the effect of the model better than the existing hybrid model.In this paper,we propose a combination of graph convolutional neural network(GCN)and long and short-term memory neural network(LSTM)for emotional eeg recognition.Since non-European spatial structure can be naturally formed between electrodes of brain channels,GCN is used to model EEG data first to extract the graph domain features of EEG data,then LSTM network is used to extract time information,and finally emotion classification results are obtained by dense layer.The average accuracy of the model reached 93.98% under the condition of a single subject.(2)As graph convolution is more complex than normal CNN convolution,the overhead of the model in the process of graph convolution is relatively large,and the information used in the process of constructing the graph from the original EEG data is too single,resulting in some information loss.The GC-GCN model is proposed in this paper.In order to verify the validity of spatial information,the average accuracy is 92.48% under the experimental condition of single subject.The graph coarding algorithm proposed in this chapter has outstanding performance in improving the model efficiency.Compared with the neural network model without graph coarding module,the training efficiency of the model has been improved by 23.27%.Finally,compared with other neural network models with the same system sensitivity,the GCGCN model proposed in this chapter also takes the lead in terms of accuracy.(3)Most of the studies on emotional EEG are completed in the case of a single subject,but the accuracy rate is low in the case of cross-subjects.This paper presents a dynamic graph convolutional neural network model based on Top-K strategy.The retention-one cross-validation strategy is used to train and test.In order to verify the effectiveness of top-k strategy,this chapter describes in detail the influence of different values of K on the whole model,and proves that the model gets the best result when the value of K is 10.Then,in order to verify the effect of the model,the whole SEED data set was compared with RGNN,DAN and other models,proving that the effect of the model was the best in cross-subjects,reaching 87.42%.At the same time,because top-K strategy can construct sparse graph,the efficiency of this model is better than the above model.