Set-pMAE: spatial-spEctral-temporal based parallel masked autoEncoder for EEG emotion recognition.

Area of Science:

• Affective Computing
• Machine Learning
• Neuroscience

Background:

• Electroencephalography (EEG) is a key tool for emotion recognition in affective computing.
• Supervised learning methods for EEG emotion recognition often struggle with limited labeled data and poor feature generalizability.
• EEG signals exhibit complex temporal, spatial, and spectral correlations with human emotional states.

Purpose of the Study:

• To propose a novel Spatial-spEctral-Temporal based parallel Masked Autoencoder (SET-pMAE) model for robust EEG-based emotion recognition.
• To leverage self-supervised learning to overcome limitations of labeled data and improve feature generalizability.
• To capture comprehensive spatial-temporal and spatial-spectral features from EEG signals.

Main Methods:

• Developed a dual-branch self-supervised learning model, SET-pMAE, for EEG emotion recognition.
• The spatial-temporal branch reconstructs EEG signals to learn contextual dependencies.
• The spatial-spectral branch reconstructs spectral features to capture inter-regional associations.

Main Results:

• The SET-pMAE model effectively learns generalized spatial-temporal and spatial-spectral representations.
• Simultaneous learning in dual branches reduces the risk of overfitting.
• Experiments on DEAP and DREAMER datasets demonstrate the model's ability to capture discriminative and generalized features.

Conclusions:

• The proposed SET-pMAE model, utilizing self-supervised learning, significantly enhances EEG emotion recognition.
• The model achieves excellent performance by capturing more discriminative and generalized features.
• This approach offers a promising direction for advancing affective computing with EEG data.