MEERNet: Multi-source EEG-based Emotion Recognition Network for Generalization Across Subjects and Sessions.

Area of Science:

• Human-Computer Interaction
• Neuroscience
• Machine Learning

Background:

• Electroencephalogram (EEG)-based emotion recognition is crucial for human-machine interaction.
• Practical application is hindered by significant inter-subject and inter-session variability in EEG data.
• Existing domain adaptation methods often fail due to differing marginal distributions across EEG sources.

Purpose of the Study:

• To develop a novel approach for robust EEG-based emotion recognition across different subjects and sessions.
• To address the limitations of traditional domain adaptation by accounting for domain-specific features.
• To improve the practicality and accuracy of EEG emotion recognition systems.

Main Methods:

• Proposed the multi-source EEG-based emotion recognition network (MEERNet).
• MEERNet extracts both domain-invariant and domain-specific features using multiple branches for each data source.
• Domain adaptation is performed individually between the target and each source, with final inference from all branches.

Main Results:

• MEERNet demonstrated superior performance compared to single-source methods in cross-session and cross-subject transfer tasks.
• Achieved an average accuracy of 86.7% for cross-session transfer and 67.1% for cross-subject transfer.
• Evaluated on the SEED and SEED-IV datasets for recognizing three and four emotions, respectively.

Conclusions:

• The proposed MEERNet effectively handles EEG data variability by incorporating domain-specific feature extraction.
• This multi-source domain adaptation strategy significantly enhances the accuracy and generalization of EEG emotion recognition.
• MEERNet offers a promising solution for more reliable and practical brain-computer interfaces.