Arrangements of Resting State Electroencephalography as the Input to Convolutional Neural Network for Biometric...

Area of Science:

• Neuroscience
• Computer Science
• Biometrics

Background:

• Biometrics are crucial for secure access to sensitive information.
• Electroencephalography (EEG) offers unique individual identification capabilities.
• Convolutional Neural Networks (CNNs) simplify biometric identification by reducing preprocessing steps.

Purpose of the Study:

• To investigate the most effective EEG data arrangement for CNN-based biometric identification.
• To compare various input formats for CNNs using EEG data.

Main Methods:

• EEG datasets from resting state eyes open (REO) and resting state eyes close (REC) conditions were used.
• Six distinct EEG data arrangements were evaluated: amplitude/energy vs. time matrices and images, with and without channel rearrangement.
• CNN models were trained and validated on these different input formats.

Main Results:

• The matrix format of amplitude versus time, with rearranged channels, using combined REC and REO data, yielded the best performance.
• This optimal arrangement achieved a validation accuracy of 83.21% and a test accuracy of 79.08%.

Conclusions:

• The arrangement of EEG data significantly impacts the performance of CNN-based biometric systems.
• A matrix of amplitude versus time for rearranged channels represents a highly effective input strategy for EEG biometrics.