Human Recognition Using Deep Neural Networks and Spatial Patterns of SSVEP Signals.

Area of Science:

• Neuroscience
• Biometrics
• Machine Learning

Background:

• Traditional biometrics face challenges; brain biometrics, particularly electroencephalography (EEG), offer unique individual identification properties.
• EEG signals exhibit distinct individual-specific features, making them promising for biometric applications.

Purpose of the Study:

• To propose and evaluate a novel brain biometric approach for individual identification.
• To leverage spatial patterns in brain responses to visual stimuli and deep learning for enhanced discrimination.

Main Methods:

• Utilizing common spatial patterns (CSP) to design personalized spatial filters for EEG data.
• Applying specialized deep-learning neural networks to map CSP features into deep representations for identification.
• Experimenting with steady-state visual evoked potential (SSVEP) datasets across various flickering frequencies.

Main Results:

• The proposed method combining CSP and deep neural networks achieved a high average correct recognition rate of 99%.
• Demonstrated superior performance compared to several classical methods on two SSVEP datasets.
• Validated the approach's effectiveness across a wide range of visual stimulus frequencies.

Conclusions:

• The novel brain biometric method integrating CSP and deep learning is highly effective for individual identification.
• The approach shows significant potential for secure and reliable person identification systems.
• The study highlights the utility of spatial patterns in EEG for advanced biometric applications.