Scholarlink

<p>Intrabody communication (IBC) channels offer physiological diversity that may support future wearable biometric identification. Recent reports of over 99 per cent identification accuracy have frequently resulted from data leakage, where samples from the same subject are seen in both training and evaluation, yielding inflated and unreliable metrics. In this work, we establish a public, leakage-free benchmark for IBC biometrics built on a 30-subject open dataset, using strict subject-wise 80/20 splits repeated five times to ensure reproducibility. We systematically compare frequency-domain and time-frequency representations, including resampled spectra, discrete wavelet transform (DWT) statistics, and their fusion. Under the subject-wise embedded-friendly benchmark, the strongest classical configuration, Scattering + LightGBM, reaches 54.0 per cent accuracy, while db4-DWT and lifting-based wavelet statistics with Random Forest improve over the Simple-3 baseline (49.3 and 51.6 per cent versus 39.0 per cent). Separately, closed-set neural analyses provide exploratory upper bounds rather than leakage-free subject-wise results: a Raw MLP reaches 83.7 per cent accuracy, whereas adding DWT statistics does not improve this result (81.2 per cent for Combined MLP), and SpectralCNN reaches 74 per cent. Confusion matrix analysis reveals that residual errors are concentrated among subject pairs with statistically overlapping signatures, suggesting the presence of intrinsically hard users and a potential biometric ceiling for this modality. Embedded profiling on an STM32F446RE Cortex-M4 microcontroller indicates that lifting-based wavelet features enable low-latency, low-energy scoring, requiring approximately 0.55 ms and 18 micro-J per 256-point spectrum for Lift-bior feature extraction plus Random Forest inference (versus approx. 33 micro-J for the equivalent db4-DWT pipeline). All code, data split scripts, and Jupyter notebooks are released open source to facilitate reproducibility and enable rigorous future comparisons.</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

Reproducible benchmark of wavelet-enhanced intrabody communication biometric identification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月28日

查看全文
  • Seungmin Jin,
  • Mikhail M. Komarov

摘要

Intrabody communication (IBC) channels offer physiological diversity that may support future wearable biometric identification. Recent reports of over 99 per cent identification accuracy have frequently resulted from data leakage, where samples from the same subject are seen in both training and evaluation, yielding inflated and unreliable metrics. In this work, we establish a public, leakage-free benchmark for IBC biometrics built on a 30-subject open dataset, using strict subject-wise 80/20 splits repeated five times to ensure reproducibility. We systematically compare frequency-domain and time-frequency representations, including resampled spectra, discrete wavelet transform (DWT) statistics, and their fusion. Under the subject-wise embedded-friendly benchmark, the strongest classical configuration, Scattering + LightGBM, reaches 54.0 per cent accuracy, while db4-DWT and lifting-based wavelet statistics with Random Forest improve over the Simple-3 baseline (49.3 and 51.6 per cent versus 39.0 per cent). Separately, closed-set neural analyses provide exploratory upper bounds rather than leakage-free subject-wise results: a Raw MLP reaches 83.7 per cent accuracy, whereas adding DWT statistics does not improve this result (81.2 per cent for Combined MLP), and SpectralCNN reaches 74 per cent. Confusion matrix analysis reveals that residual errors are concentrated among subject pairs with statistically overlapping signatures, suggesting the presence of intrinsically hard users and a potential biometric ceiling for this modality. Embedded profiling on an STM32F446RE Cortex-M4 microcontroller indicates that lifting-based wavelet features enable low-latency, low-energy scoring, requiring approximately 0.55 ms and 18 micro-J per 256-point spectrum for Lift-bior feature extraction plus Random Forest inference (versus approx. 33 micro-J for the equivalent db4-DWT pipeline). All code, data split scripts, and Jupyter notebooks are released open source to facilitate reproducibility and enable rigorous future comparisons.