Related Experiment Video
Updated: Mar 3, 2026
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
5.5K
A Fast, Efficient Domain Adaptation Technique for Cross-Domain Electroencephalography(EEG)-Based Emotion Recognition
Xin Chai1, Qisong Wang2, Yongping Zhao3
1School of Electrical Engineering and Automation, Harbin Institute of Technology, Harbin 150001, China. 11b901011@hit.edu.cn.
Sensors (Basel, Switzerland)
|May 4, 2017
Summary
This study introduces Adaptive Subspace Feature Matching (ASFM), a novel domain adaptation method for electroencephalography (EEG) emotion recognition. ASFM effectively reduces performance degradation across subjects and sessions, improving accuracy in real-time applications.
Area of Science:
- Neuroscience
- Machine Learning
- Biomedical Engineering
Background:
- Electroencephalography (EEG)-based emotion recognition is crucial for psychiatric health diagnosis.
- Non-stationary EEG signals across sessions/subjects degrade classification performance.
- Existing domain adaptation methods struggle with both marginal and conditional distribution mismatches in EEG data and often require high computational cost.
Purpose of the Study:
- To propose a novel, computationally efficient domain adaptation strategy, Adaptive Subspace Feature Matching (ASFM), for EEG-based emotion recognition.
- To address both marginal and conditional distribution discrepancies in EEG data without requiring labeled target samples.
- To enable effective real-time emotion recognition by reducing performance degradation across subjects and sessions.
Main Methods:
- Developed a linear transformation function to match marginal distributions of source and target subspaces without regularization, reducing computational complexity.
- Integrated both marginal and conditional distribution adaptation within a unified framework.
- Applied logistic regression (LR) on the aligned source domain for classification in the target domain.
Main Results:
- Subject-to-subject offline experiments showed ASFM achieved 80.46% mean accuracy, outperforming the subspace alignment auto-encoder (SAAE) at 77.88%.
- Online analysis demonstrated ASFM's average accuracy of 75.11%, significantly improving upon the baseline LR at 56.38%.
- ASFM exhibited superior computational efficiency, making it suitable for real-time classification.
Conclusions:
- ASFM effectively reduces domain discrepancies and performance degradation in EEG-based emotion recognition across subjects and sessions.
- The method offers a computationally efficient and effective solution for real-time emotion recognition applications.
- ASFM proves to be a valuable tool for enhancing the robustness and accuracy of EEG emotion recognition systems.
