Cross-session SSVEP brainprint recognition using attentive multi-sub-band depth identity embedding learning network

Area of Science:

• Biometrics
• Neuroscience
• Machine Learning

Background:

• Brainprint recognition using electroencephalogram (EEG) faces challenges due to signal variability and low signal-to-noise ratio.
• Steady-state visual evoked potentials (SSVEP) offer a higher signal-to-noise ratio and frequency locking, making them suitable for brainprint recognition.
• Extracting time-invariant identity information from SSVEP EEG signals is crucial for reliable biometric systems.

Purpose of the Study:

• To develop a robust method for stable cross-session SSVEP brainprint recognition.
• To address the challenge of reduced recognition accuracy across different recording sessions.
• To propose a novel deep learning architecture for enhanced identity information extraction from SSVEP signals.

Main Methods:

• Proposed an Attentive Multi-sub-band Depth Identity Embedding Learning Network.
• Introduced a Sub-band Attentive Frequency mechanism to integrate frequency-domain characteristics and explore depth-frequency identity information.
• Employed Attentive Statistic Pooling to improve the stability of frequency domain feature distributions across sessions.

Main Results:

• The proposed method achieved superior performance compared to state-of-the-art models on 2-second SSVEP samples across sessions.
• Demonstrated enhanced stability and accuracy in cross-session brainprint recognition.
• Validated the approach on two multi-session SSVEP benchmark datasets.

Conclusions:

• The Attentive Multi-sub-band Depth Identity Embedding Learning Network provides stable cross-session SSVEP brainprint recognition.
• The proposed mechanisms effectively address the limitations of low signal-to-noise ratio and time-varied brain signals.
• The approach shows potential as a benchmark for multi-subject biometric recognition systems.