Research On EEG Emotion Recognition Based On Space-time Neural Networ

Emotion is the basis of daily human life and plays an essential role in human cognitive functions,rational decisions,and interpersonal communications.EEG is an electrophysiological signal generated by the human central nervous system,which has been widely used in the research of human emotion recognition in recent years.The recognition of human emotion through EEG signals has the advantages of easy acquisition,good stability,and wide application range.This thesis aiming at the large amount of computation,cumbersome steps,and the large differences between samples in the cross-subject EEG emotion recognition task of traditional EEG emotion classification tasks.Neural network are used to model in the two dimensions of spatial and temporal,extract effective features in the original EEG signal,pay attention to the differences between different samples,and propose effective network models to solve the above problems.The research content of this thesis mainly includes the following parts:(1)Aimming at the problem of cumbersome feature extraction steps and large calculation in EEG emotion recognition.This chapter proposed an end-to-end spatial-temporal neural network model for subject-dependent EEG emotion recognition.First,baseline correction is performed on the original EEG signal to remove noise interference;Secondly,considering the spatial and temporal information contained in EEG signals,an end-to-end EEG emotion recognition neural network E2 ENNet is proposed.The network uses a convolutional neural network based on deep separable convolution to extract features from the spatial level,and finally uses a long short-term memory network to extract time series features from the time level,and fuses the two features for emotion recognition classification.The model runs end-to-end,which effectively reducing the steps of EEG emotion recognition,preserving the emotional features in the original signal,and reducing computational complexity.(2)Aiming at the problem of differences between different individual samples in cross-participant EEG emotion recognition.This chapter proposed an attention-based spatial-temporal neural network model for cross-subject EEG emotion recognition.The model introduces the attention mechanism into the neural network,so that it can pay attention to the differences between different samples in the process of feature extraction,extract representative features for emotion recognition,and remove redundant information.Firstly,the EEG signal with the baseline signal removed is extracted from the spatial features through the CNN module,and the spatial attention module is used to suppress irrelevant feature information and retain valid information.Then the temporal features of EEG signals are extracted through the RNN module,and the temporal information is advanced through the self attention module.Through the weighting processing of the temporal information in the self attention module,the differences between different samples are eliminated,and the final features can be better applied to the sentiment classification task between different subjects.The models proposed in this thesis has been verified by extensive number of experiments on the three public datasets of DEAP,DREAMER and MPED,and the experimental results show the effectiveness of the EEG emotion recognition methods proposed in this thesis.