Cross-subject EEG-based Emotion Recognition Using Adversarial Domain Adaption with Attention Mechanism

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Cross-subject emotion recognition (ER) using electroencephalography (EEG) faces challenges due to significant individual differences.
• Subject-specific ER models are time-consuming and impractical for new users in real-world applications.
• Existing methods struggle to bridge the gap between different subjects' EEG data.

Purpose of the Study:

• To propose a novel method, ADAAM-ER, to decrease individual discrepancies in EEG-based emotion recognition.
• To enhance the transferability of ER models to new subjects without requiring subject-specific training.
• To develop a more robust and efficient cross-subject ER system.

Main Methods:

• Developed ADAAM-ER, integrating Graph Convolutional Neural Networks with CNNs (GCNN-CNNs) for feature extraction.
• Employed an Adversarial Domain Adaption with a Level-wise Attention Mechanism (ADALAM) to align feature distributions across subjects.
• Utilized GCNN-CNNs to create a broader feature space for more discriminative features.

Main Results:

• The proposed ADAAM-ER method demonstrated improved transferability of emotion recognition models.
• Achieved a mean accuracy of 86.58% on the SEED dataset, validating the method's effectiveness.
• Successfully decreased individual discrepancies by aligning transferable feature regions.

Conclusions:

• ADAAM-ER offers a more transferable and discriminative approach to EEG-based emotion recognition.
• The method addresses the limitations of subject-specific models in cross-subject scenarios.
• This research advances the development of practical, real-world EEG emotion recognition systems.