Scholarlink

<p>This study aims to address the inefficiency and delayed feedback inherent in traditional manual guidance for football technical training by proposing a digital image analysis solution based on the spatio-temporal graph convolutional network (ST-GCN). The research method employs a dual-stream adaptive graph convolutional model (spatial and temporal streams), which captures spatiotemporal features of movements through learnable weighted adjacency matrices and dynamic temporal convolution kernels. A four-dimensional evaluation system is constructed, encompassing spatial accuracy, temporal rhythm, kinetic chain coordination, and postural stability. The results demonstrate: 1) the model achieved 89.7% action recognition accuracy and a macro F1-score of 0.878 on the test set, with a real-time feedback delay of only 58&#xa0;ms; 2) a four-week empirical training period resulted in a 38.2% improvement in movement accuracy for the experimental group (<i>p</i> &lt; 0.05), outperforming the control group. The findings indicate that this system, through its multimodal real-time feedback mechanism (500&#xa0;ms response cycle), effectively enhances training efficiency and is expected to provide a valuable technical reference for intelligent training in football.</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

Digital image analysis technology in football technical training

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月18日

查看全文
  • Haibo Dai

摘要

This study aims to address the inefficiency and delayed feedback inherent in traditional manual guidance for football technical training by proposing a digital image analysis solution based on the spatio-temporal graph convolutional network (ST-GCN). The research method employs a dual-stream adaptive graph convolutional model (spatial and temporal streams), which captures spatiotemporal features of movements through learnable weighted adjacency matrices and dynamic temporal convolution kernels. A four-dimensional evaluation system is constructed, encompassing spatial accuracy, temporal rhythm, kinetic chain coordination, and postural stability. The results demonstrate: 1) the model achieved 89.7% action recognition accuracy and a macro F1-score of 0.878 on the test set, with a real-time feedback delay of only 58 ms; 2) a four-week empirical training period resulted in a 38.2% improvement in movement accuracy for the experimental group (p < 0.05), outperforming the control group. The findings indicate that this system, through its multimodal real-time feedback mechanism (500 ms response cycle), effectively enhances training efficiency and is expected to provide a valuable technical reference for intelligent training in football.