Robust Brainprint Recognition via Session-wise Riemannian Alignment: Overcoming Non-stationary Brain Dynamics

The non-stationarity of electroencephalography (EEG) signals constitutes a fundamental barrier to their application in reliable biometric identification. Due to fluctuations in cognitive states and environmental conditions, cross-session EEG signals exhibit significant distributional drift on the Riemannian manifold. This study proposes a robust brainprint recognition framework that integrates Session-wise Riemannian Centerization (S-RCT) with a global channel optimization strategy. By aligning the covariance matrices from different sessions to a common geometric reference, the proposed method effectively mitigates temporal variability. Experimental results on the BCI Competition IV 2a dataset demonstrate that this approach achieves a cross-session identification accuracy of 98.33% using 22 channels. Furthermore, topological analysis of channel configurations identifies an optimal 5-channel array centered on the parieto-occipital region. This minimal configuration maintains an accuracy of 90.89%, revealing the inherent dynamical stability of this brain region for identity representation. This research provides a geometric solution for constructing neuro-biometric systems robust to temporal drift.