S³T-Net: A novel electroencephalogram signals-oriented emotion recognition model

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Electroencephalogram (EEG) signals exhibit significant intra-individual variability, posing challenges for accurate emotion recognition.
• Existing models often struggle with generalization due to these inherent differences in EEG data.
• Robust emotion recognition is crucial for advancing human-computer interaction (HCI) and intelligent sentiment analysis.

Purpose of the Study:

• To develop a novel deep learning model, the Skipping Spatial-Spectral-Temporal Network (S³T-Net), for robust and generalized emotion recognition from EEG signals.
• To effectively handle intra-individual differences in EEG data for improved model performance.
• To enhance the accuracy and reliability of emotion recognition systems in HCI applications.

Main Methods:

• A multi-branch architecture was employed to extract spatial-spectral cross-domain representations from 4D EEG features.
• A bi-directional long-short term memory module with an attention mechanism was utilized to capture temporal dependencies and integrate context.
• A skip-change unit was introduced to mitigate the vanishing gradient problem in deep spatial-temporal networks.

Main Results:

• The proposed S³T-Net demonstrated superior performance compared to existing advanced models in emotion recognition accuracy.
• Performance improvements of 0.23%, 0.13%, and 0.43% were achieved over the sub-optimal model across three test scenarios.
• Experimental validation confirmed the effectiveness and superiority of the S³T-Net's key components.

Conclusions:

• The S³T-Net provides a reliable and competent solution for emotion recognition from EEG signals, effectively addressing intra-individual differences.
• The model's ability to learn generalized spatial-spectral-temporal representations contributes to more accurate and robust sentiment analysis.
• This work advances the development of intelligent sentiment analysis within the human-computer interaction (HCI) domain.