Scholarlink

<p>Gait analysis has become a systematic assessment process for diagnosing and monitoring neuro-musculoskeletal abnormalities. This process relies on extracting comprehensive spatial-temporal and frequency-domain features from multi-modal gait data. It is observed that traditional feature extraction methods (statistical, machine learning, and convolutional neural networks) have limitations in learning complex features with long-range dependencies directly from gait signals. To overcome these limitations, we employ multi-domain feature extraction techniques, including the discrete wavelet transform (DWT), the continuous wavelet transform (CWT), fast Walsh–Hadamard transform (FWHT), and power spectral density (PSD). This multi-domain feature preserves the spatial-temporal properties of the gait signal. A hybrid neural architecture with temporal convolutional networks (TCN) and Siamese networks is proposed for the classification of gait abnormality on the standard inertia measurement unit (IMU) dataset. The TCN model captures long-term dependencies in the gait signal. The proposed hybrid model achieves a 97.74% accuracy, outperforming the state-of-the-art algorithms.</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

Hybrid neural model for the classification of gait abnormality using multi-domain features

  • 期刊论文

  • 发表日期：

    2026年5月19日

查看全文
  • Mainak Ghosh,
  • Anup Nandy,
  • Bidyut Kr Patra,
  • R Anitha,
  • K Mohanavelu

摘要

Gait analysis has become a systematic assessment process for diagnosing and monitoring neuro-musculoskeletal abnormalities. This process relies on extracting comprehensive spatial-temporal and frequency-domain features from multi-modal gait data. It is observed that traditional feature extraction methods (statistical, machine learning, and convolutional neural networks) have limitations in learning complex features with long-range dependencies directly from gait signals. To overcome these limitations, we employ multi-domain feature extraction techniques, including the discrete wavelet transform (DWT), the continuous wavelet transform (CWT), fast Walsh–Hadamard transform (FWHT), and power spectral density (PSD). This multi-domain feature preserves the spatial-temporal properties of the gait signal. A hybrid neural architecture with temporal convolutional networks (TCN) and Siamese networks is proposed for the classification of gait abnormality on the standard inertia measurement unit (IMU) dataset. The TCN model captures long-term dependencies in the gait signal. The proposed hybrid model achieves a 97.74% accuracy, outperforming the state-of-the-art algorithms.