A Spatial-Spectral-Temporal Representation Method Based on Riemannian Manifold for EEG Individual Identification.

Area of Science:

• Neuroscience and Computer Science
• Biometric Identification

Background:

• Electroencephalography (EEG) is a promising biometric modality due to its inherent characteristics.
• A significant challenge in EEG-based identification is maintaining stable, temporally robust inter-individual features.
• Effective cross-time EEG identity recognition is crucial for practical biometric systems.

Purpose of the Study:

• To propose a novel EEG-based identification framework using symmetric positive definite (SPD) manifolds.
• To develop an enhanced feature extraction method (E-SPD-M) for capturing temporal, spatial, and spectral EEG characteristics.
• To improve the accuracy and reliability of cross-time EEG-based identity recognition.

Main Methods:

• Utilized spatial covariance matrices of EEG signals to represent individual differences.
• Employed an enhanced feature extraction method (E-SPD-M) within Riemannian manifolds.
• Constructed personalized classification models and integrated their outputs for identification.
• Validated the approach on a new multi-task, cross-time EEG dataset and a public dataset (M3CV).

Main Results:

• The proposed framework achieved superior cross-time identification performance.
• Demonstrated the effectiveness of SPD manifolds in constructing discriminative representation spaces for EEG data.
• The E-SPD-M method successfully captured essential temporal, spatial, and spectral features.

Conclusions:

• The study presents a novel pathway for enhancing EEG-based biometric algorithms.
• Highlights the potential of Riemannian geometry in advancing EEG identification.
• Offers a more reliable and practical solution for cross-time EEG identity recognition.