Scholarlink

<p>Electroencephalogram (EEG) signals, reflecting synaptic activity across the brain’s surface, offer a promising avenue for biometric authentication due to their resilience to spoofing attacks and immunity to coercion. This study leverages the BCI IV 2a dataset, where participants imagine movements of four body parts: right hand, left hand, both foot, and tongue. A two-dimensional Convolutional Neural Network (CNN) with six convolutional layers, combined with a self-distillation technique, was employed to maximize classification accuracy. The model uses channel-wise convolutional kernels to capture localized spatial features from individual EEG channels, enabling analysis of inter-channel relationships during the distillation process. Various scenarios were analyzed based on task type, signal length, frequency band, and the number of EEG channels. Optimal results were achieved using a combination of frequency bands and the maximum number of channels, with a 4-second signal input yielding 100% accuracy and a 2-second input achieving 99.8397% accuracy for left-hand or tongue movement tasks. Despite these advancements, EEG-based authentication still faces challenges requiring further research to match the reliability and security of traditional biometric methods like fingerprint authentication.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

EEG-based biometric authentication: advancing security through motor imagery and deep learning

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Mohammadreza Mostafavi,
  • Ahmad Ayatollahi,
  • Sivakumar Rajagopal,
  • Seok-Bum Ko

摘要

Electroencephalogram (EEG) signals, reflecting synaptic activity across the brain’s surface, offer a promising avenue for biometric authentication due to their resilience to spoofing attacks and immunity to coercion. This study leverages the BCI IV 2a dataset, where participants imagine movements of four body parts: right hand, left hand, both foot, and tongue. A two-dimensional Convolutional Neural Network (CNN) with six convolutional layers, combined with a self-distillation technique, was employed to maximize classification accuracy. The model uses channel-wise convolutional kernels to capture localized spatial features from individual EEG channels, enabling analysis of inter-channel relationships during the distillation process. Various scenarios were analyzed based on task type, signal length, frequency band, and the number of EEG channels. Optimal results were achieved using a combination of frequency bands and the maximum number of channels, with a 4-second signal input yielding 100% accuracy and a 2-second input achieving 99.8397% accuracy for left-hand or tongue movement tasks. Despite these advancements, EEG-based authentication still faces challenges requiring further research to match the reliability and security of traditional biometric methods like fingerprint authentication.