Investigation of EEG-Based Biometric Identification Using State-of-the-Art Neural Architectures on a Real-Time...

Area of Science:

• Neuroscience
• Biometrics
• Machine Learning

Background:

• Electroencephalogram (EEG) signals are increasingly explored for subject identification.
• Deep learning (DL) models have advanced neurological signal analysis, but their application to EEG biometrics is limited due to signal variability.
• Existing systems often only incorporate partial EEG biometric identification processes.

Purpose of the Study:

• To investigate the efficacy of state-of-the-art DL models (ResNet, Inception, EEGNet) for EEG-based subject identification.
• To develop a portable, low-cost, real-time system for end-to-end EEG biometric identification.
• To address the challenge of high EEG feature variability across sessions.

Main Methods:

• Utilized the BED dataset with EEG recordings from 21 individuals.
• Applied deep learning models including ResNet, Inception, and EEGNet.
• Developed a Raspberry Pi-based system for real-time EEG signal acquisition and identity prediction.

Main Results:

• Achieved accuracies of 63.21% (ResNet), 70.18% (Inception), and 86.74% (EEGNet).
• EEGNet outperformed previous best efforts (83.51%) on the BED dataset.
• Demonstrated successful real-time EEG biometric identification using the developed portable system.

Conclusions:

• State-of-the-art DL models, particularly EEGNet, are effective for EEG-based subject identification.
• The developed system enables practical, real-time EEG biometrics.
• Further research can leverage these models to overcome EEG signal variability challenges in biometrics.