EEG Emotion Recognition Based On Self-attention Mechanism

Emotion is the external manifestation of human complex psychological state,which is closely related to people’s health and life.The effective recognition of human emotions by computers through artificial intelligence technology has broad application prospects and social value in the fields of medical care and human-computer interaction.In recent years,emotion recognition,as one of the important research contents in the field of artificial intelligence,has received more and more attention and research.In emotion recognition research,because EEG is closely related to neural activity,it can objectively reflect people’s emotional state,and has the advantages of low acquisition cost and convenient use,therefore,EEG emotion recognition research has always attracted much attention.At present,the research on EEG emotion recognition has made some progress,but in the existing research,there are still the following problems:(1)in the research of EEG channel selection and optimization,the division and screening of EEG channels still rely on manual division and manual feature extraction.The model lacks the ability of independent channel selection of the original EEG,and can not guarantee the performance of channel inhibition;(2)In the emotion classification model,the model lacks the autonomous learning ability of local correlation of brain regions,and does not make full use of the global connection information between brain regions.To solve the above problems,this paper has carried out the following work:(1)Aiming at the problems existing in the selection and optimization of EEG channels,this paper proposes an EEG emotion recognition neural network based on channel and temporal self-attention.The network uses the raw EEG data as input,learns the importance of different EEG channels through channel-wise attention mechanism,and obtains important channels related to different emotional tasks.The multi-head self-attention mechanism in the model further ensures the performance of the model in the channel suppression experiment,which verifies the rationality of channel selection.The model achieves the recognition rates of 95.70%and 96.23% respectively on the two classification tasks of valence and arousal on DEAP data set,and an optimized channel configuration scheme based on raw EEG data is proposed.(2)Aiming at the problem that the model lacks the ability of autonomous learning for the local correlation of brain regions,this paper proposes an EEG emotion classification model based on the brain region-like local-global attention.Through the position coding and multihead self-attention mechanism,considering the local correlation of brain regions,the global connectivity information of EEG is effectively extracted.The model combined the local and global correlations of brain regions and successfully learned the connectivity characteristics of brain regions.The model achieves 98.21% and 98.47% accuracy in EEG emotion recognition tasks based on valence and arousal on the DEAP dataset,respectively,which improved the emotion classification performance of the model.