Joint disentangled representation and domain adversarial training for EEG-based cross-session biometric recognition
Honggang Liu1,2, Xuanyu Jin1,2, Dongjun Liu1,2
1School of Computer Science, Hangzhou Dianzi University, Hangzhou, 310018 Zhejiang China.
Cognitive Neurodynamics
|January 27, 2025
Summary
This study introduces a new framework for electroencephalogram (EEG) biometrics, improving accuracy in cross-session recognition by disentangling identity features and using adversarial training for robust performance.
Area of Science:
- Biometrics
- Neuroscience
- Machine Learning
Background:
- Wearable technologies increase the use of electroencephalogram (EEG) for biometric recognition.
- Cross-session variability and single-task protocol diversity challenge EEG biometric model stability and generalization.
Purpose of the Study:
- To propose a novel framework, Joint Disentangled Representation with Domain Adversarial Training (JDR-DAT), for robust EEG-based cross-session biometric recognition.
- To enhance longitudinal robustness and model generalization within single-task protocols.
Main Methods:
- The JDR-DAT framework disentangles identity-specific features using mutual information estimation.
- Domain adversarial training is incorporated to improve robustness across different sessions.
- Experiments were conducted on RSVP-based and MI-based single-task protocol datasets.
Main Results:
- The JDR-DAT framework demonstrated significant efficacy in cross-session biometric recognition.
- Average accuracies of 85.83% on RSVP-based data and 96.72% on MI-based data were achieved.
- The proposed method shows strong performance in longitudinal EEG data analysis.
Conclusions:
- The JDR-DAT framework effectively addresses challenges in EEG-based cross-session biometric recognition.
- The method provides a robust solution for maintaining model stability and generalization in longitudinal EEG data.
- This work advances the application of EEG signals in biometric identification systems.
