سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

صفحه اصلی / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Dual-View Data Representation and Contrastive Learning for Robust EEG-Based Person Identification

نویسندگان :

Mahdi Tabatabaei¹ Mohammad Bagher Shamsollahi²

1- Sharif University of Technology 2- ُSharif University of Technology

کلمات کلیدی :

Person identification،EEG signal،Dual-view representation،Contrastive learning

چکیده :

Person identification using EEG signals has emerged as a promising biometric modality due to the uniqueness and internal nature of brain signals. However, EEG signals are often noisy, task-dependent, and limited in quantity, posing challenges for reliable and generalizable identification. In this paper, we propose a novel dual-view learning framework that integrates both scalp-level spatial features and source-informed neural topology using convolutional and graph neural networks (CNN and GCN). The scalp view captures spatial-spectral representations through a CNN applied to differential entropy (DE) maps, while the neural view models inter-channel relationships via a GCN built on source-localized graphs. To improve generalization across tasks and sessions, we incorporate supervised contrastive learning, encouraging the model to cluster embeddings of the same individual regardless of task conditions. The proposed method is evaluated on the EEG Motor Movement/Imagery Dataset and achieves high accuracy across within-task and cross-task scenarios. Furthermore, it remains effective with reduced training data and benefits significantly from feature normalization. Experimental results and ablation studies confirm the contribution of each component and demonstrate the reliability of the approach for EEG-based biometric identification.