The single-channel dry electrode SSVEP-based biometric approach: data augmentation techniques against overfitting for RNN-based deep models


Gorur K., Eraslan B.

Physical and Engineering Sciences in Medicine, vol.45, no.4, pp.1219-1240, 2022 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 45 Issue: 4
  • Publication Date: 2022
  • Doi Number: 10.1007/s13246-022-01189-1
  • Journal Name: Physical and Engineering Sciences in Medicine
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.1219-1240
  • Keywords: Biometric, Data augmentation techniques, Recurrent neural networks, Single-channel, SSVEP
  • Ankara Yıldırım Beyazıt University Affiliated: No

Abstract

Biometric studies based on electroencephalography (EEG) have received increasing attention because each individual has a dynamic and unique pattern. However, classic EEG-based biometrics have significant deficiencies, including noise-prone signals, gel-based electrodes, and the need for multi-training/multi-channel acquisition and high mental effort. In contrast, steady-state visually evoked potential (SSVEP)-based biometrics have the important advantages of high signal-to-noise ratio and untrained usage. Dynamic brain potential responses are a natural subconscious activity and can be elicited by flickering lights having distinct frequencies, such as cell phone flashes, without extra physical or mental effort. Few studies involving multi-channel/multi-trial SSVEP-based biometric research are available in the current literature. Moreover, there is a lack of research comparing them to the single-channel single-trial dry electrode-implemented SSVEP-based biometric approach using Recurrent Neural Networks (RNN). Furthermore, to the best of our knowledge, no prior work has proposed an SSVEP-based biometric comparison of the RNNs using data augmentation strategies against overfitting. It was observed that the biometric recognition results were promising, achieving up to 100% accuracy and > 97% sensitivity and specificity scores for 11 subjects. F-scores were also yielded as > 97% values. This single-channel SSVEP-based biometric approach using RNN deep models may offer low-cost, user-friendly, and reliable individual identification authentication, leading to significant application domains.