Scholarlink

<p>Cycling is a common and effective home-based rehabilitation exercise. Accurate and accessible assessment of the onset of fatigue is essential to achieving optimal exercise benefits and preventing overuse injuries. To obtain fatigue-related parameters in different age groups, we applied deep learning algorithms and signal processing technology to analyze cycling movement features for the people aged over 45. 20 healthy adults aged over 45 and 20 aged 18–30 were recruited. Participants were asked to ride a stationary exercise bike at their self-regulated pedaling speeds for 10&#xa0;min and wear a COSMED K5 device to collect physiological signals. The Keypoint RCNN (KR) algorithm and three signal processing methods (Fourier transform, short-time Fourier transform, and multiscale entropy analysis were used to analyze the cycling movement data. Based on time–frequency analysis, subjects’ movement status change points were identified when fatigue occurred. Four movement status parameters were calculated, including the peak frequency before/after the movement status change point and the complexity index average (<i>CIA</i>) before/after the movement status change point. Inter-group and intra-group movement features, movement status, and physiological data were compared to determine fatigue-related features. Results showed that the peak frequency (<i>p</i> = 0.005), the peak frequency before/after the change point (<i>p</i> = 0.008/0.019), the&#xa0;<i>CIA&#xa0;</i>after the change point (<i>p</i> = 0.014), the maximum heart rate, maximal oxygen consumption, metabolic equivalents, and energy efficiency exhibited significant inter-group differences. The KR algorithm demonstrated outstanding performance in keypoint detection, achieving an accuracy of 86.5%, significantly outperforming OpenPose. With an inference speed of 30 FPS, it fulfills the demands for real-time monitoring. In addition, <i>CIA&#xa0;</i>valuses before and after change pointsshowed significant differences in the the middle-aged and elderly people group. After the change point<i>,</i> the <i>CIA</i> canidentify movement status changes in inter-group and intra-group comparisons, suggesting it can be used as a indicator of fatigue status, especially for people aged over 45.</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

The fatigue status feature of bicycle movement based on deep learning and signal processing technology

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月10日

查看全文
  • Yingchun He,
  • Yih-haw Jan,
  • Fan Yang,
  • Yunru Ma,
  • Xin-Yuan Chen,
  • Chun Pei

摘要

Cycling is a common and effective home-based rehabilitation exercise. Accurate and accessible assessment of the onset of fatigue is essential to achieving optimal exercise benefits and preventing overuse injuries. To obtain fatigue-related parameters in different age groups, we applied deep learning algorithms and signal processing technology to analyze cycling movement features for the people aged over 45. 20 healthy adults aged over 45 and 20 aged 18–30 were recruited. Participants were asked to ride a stationary exercise bike at their self-regulated pedaling speeds for 10 min and wear a COSMED K5 device to collect physiological signals. The Keypoint RCNN (KR) algorithm and three signal processing methods (Fourier transform, short-time Fourier transform, and multiscale entropy analysis were used to analyze the cycling movement data. Based on time–frequency analysis, subjects’ movement status change points were identified when fatigue occurred. Four movement status parameters were calculated, including the peak frequency before/after the movement status change point and the complexity index average (CIA) before/after the movement status change point. Inter-group and intra-group movement features, movement status, and physiological data were compared to determine fatigue-related features. Results showed that the peak frequency (p = 0.005), the peak frequency before/after the change point (p = 0.008/0.019), the CIA after the change point (p = 0.014), the maximum heart rate, maximal oxygen consumption, metabolic equivalents, and energy efficiency exhibited significant inter-group differences. The KR algorithm demonstrated outstanding performance in keypoint detection, achieving an accuracy of 86.5%, significantly outperforming OpenPose. With an inference speed of 30 FPS, it fulfills the demands for real-time monitoring. In addition, CIA valuses before and after change pointsshowed significant differences in the the middle-aged and elderly people group. After the change point, the CIA canidentify movement status changes in inter-group and intra-group comparisons, suggesting it can be used as a indicator of fatigue status, especially for people aged over 45.