Subject-independent emotion recognition of EEG signals using graph attention-based spatial-temporal pattern learning

Electroencephalography (EEG) reveals human brain activities and becomes an essential solution for exploring human intrinsic emotional states. In this study, we proposed a graph attention-based spatial-temporal pattern learning method called TAGAT to take full advantage of spatial structure of EEG ch...

Full description

Saved in:
Bibliographic Details
Published inChinese Control Conference pp. 7070 - 7075
Main Authors Zhu, Yiwen, Guo, Yeshuang, Zhu, Wenzhe, Di, Lare, Yin, Zhong
Format Conference Proceeding
LanguageEnglish
Published Technical Committee on Control Theory, Chinese Association of Automation 25.07.2022
Subjects
Adaptation models
Brain modeling
Computational modeling
Deep learning
electroencephalography
Emotion recognition
Learning systems
pattern learning
transfer learning
Online AccessGet full text

Cover

More Information
Summary:Electroencephalography (EEG) reveals human brain activities and becomes an essential solution for exploring human intrinsic emotional states. In this study, we proposed a graph attention-based spatial-temporal pattern learning method called TAGAT to take full advantage of spatial structure of EEG channels and take the nonstationary of emotions into consideration. The attention mechanism is applied to compute weight coefficients of different spatial and temporal patterns. To alleviate differences between subjects, a domain discriminator is added to the model based on domain adaptation to tackle subject-independent EEG emotion recognition tasks. Based on the DEAP database, our method achieves the accuracy of 56.56% and 58.91% of arousal and valence dimensions for subject-independent emotion classification. The effectiveness of T-AGAT method has been demonstrated when compared to other existing common methods.
ISSN:1934-1768
DOI:10.23919/CCC55666.2022.9901838