Residual-Enhanced VAE-GAN for EEG Data Augmentation in Motor Imagery Classification

Motor imagery is a widely studied paradigm in Brain-Computer Interfaces (BCIs) systems, where electroencephalography (EEG) signals are used to decode brain activity. However, subject-specific variability leads to significant differences in EEG distributions across individuals, making it challenging...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:IEEE International Conference on Consumer Electronics-China (Online) s. 603 - 604
Hlavní autori: Tang, Haiyuan, Zhao, Wenshan
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: IEEE 16.07.2025
Predmet:
Accuracy
Autoencoders
Brain modeling
Brain-computer interfaces
Data augmentation
Data models
Domain Generalization
Electroencephalography
Image reconstruction
Motor Imagery
Motors
Residual Modeling
Training data
ISSN:2575-8284
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Motor imagery is a widely studied paradigm in Brain-Computer Interfaces (BCIs) systems, where electroencephalography (EEG) signals are used to decode brain activity. However, subject-specific variability leads to significant differences in EEG distributions across individuals, making it challenging for models trained on existing subject data to generalize well to unseen individuals. To improve the diversity of training data and enhance model generalization, this paper proposes an EEG data augmentation method, termed Residual-Enhanced Variational Autoencoder-Generative Adversarial Network (REVG). The framework combines a Variational Autoencoder for EEG reconstruction with a Wasserstein Generative Adversarial Network to generate realistic residual signals, enriching EEG data distribution. Experimental results on the BCI Competition IV 2b dataset demonstrate that REVG achieves an average classification accuracy of 76.83%, outperforming the baseline and other data augmentation methods.
ISSN:2575-8284
DOI:10.1109/ICCE-Taiwan66881.2025.11208108