Brainprint Measurement Under Complex Cognitive Load Scenarios With a Spatially Aware Polyspectral Retention Network

The brainprint, based on electroencephalography (EEG), offers unique identification through dynamic physiological patterns. However, EEGs may concurrently capture cognitive load as a potential interference to brainprint. On the contrary, traditional EEG-based deep models, designed for diagnostic purposes, often treat brainprint variations as noise. To address this mismatch, we introduce the brainprint database under complex cognitive load scenarios (BOOLS) and a novel spatially aware polyspectral retention network (SAPR-Net). The BOOLS database records EEG signals under four different cognitive load levels from 30 participants, which is made publicly available. The SAPR-Net amplifies individual-specific EEG features by improving the connectivity within and between cortical regions, utilizing a new electrode-position encoding method. The peak performance of the SAPR-Net on the BOOLS database was achieved at 94.70%. Through analysis of channel selection strategies and window size choices, we demonstrate that the brainprint can be consistently preserved despite cognitive load fluctuations by leveraging interpretable EEG connectivity.