Scholarlink

Power engineering environments are usually unstructured and the environment is constantly changing, traditional manual monitoring means can no longer meet the demand for real-time monitoring of personnel actions in complex environments of modern power systems, existing intelligent detection methods are difficult to accurately extract useful behavioral information, there is a delay problem in the real-time monitoring system, and it is difficult to guarantee the accuracy and efficiency of behavior recognition at the same time. To address these challenges, this paper proposes a methodological framework for power construction personnel behavioral trajectory recognition, including data mining, improved MMPose pose estimation method and behavioral recognition based on spatio-temporal decoupling-compression contrast learning. The data mining part extracts effective data of workers’ activities through cleaning, preprocessing, correlation analysis and cluster analysis. The posture estimation part annotates the skeleton information of workers’ behaviors by improving the MMPose framework, optimizing the inference process, reducing the processing time and enhancing the stability of the algorithm. Experimental results show that our proposed solution generally outperforms other algorithms in terms of runtime, with a behavior recognition AP energy guarantee of 82.37% and a low deployment consumption, and is able to effectively monitor and analyze the behavioral trajectories of workers at power engineering sites.

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

A Deep Learning-Based Framework for Data Mining and Recognition of Workers’ Dynamic Behavioral Trajectories

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yong Liu,
  • Zifeng Zhang,
  • Jian Zhu,
  • Fang Xie,
  • Yingxue Sun,
  • Jingya Li,
  • Yue Wang,
  • Zhiyuan Yu

摘要

Power engineering environments are usually unstructured and the environment is constantly changing, traditional manual monitoring means can no longer meet the demand for real-time monitoring of personnel actions in complex environments of modern power systems, existing intelligent detection methods are difficult to accurately extract useful behavioral information, there is a delay problem in the real-time monitoring system, and it is difficult to guarantee the accuracy and efficiency of behavior recognition at the same time. To address these challenges, this paper proposes a methodological framework for power construction personnel behavioral trajectory recognition, including data mining, improved MMPose pose estimation method and behavioral recognition based on spatio-temporal decoupling-compression contrast learning. The data mining part extracts effective data of workers’ activities through cleaning, preprocessing, correlation analysis and cluster analysis. The posture estimation part annotates the skeleton information of workers’ behaviors by improving the MMPose framework, optimizing the inference process, reducing the processing time and enhancing the stability of the algorithm. Experimental results show that our proposed solution generally outperforms other algorithms in terms of runtime, with a behavior recognition AP energy guarantee of 82.37% and a low deployment consumption, and is able to effectively monitor and analyze the behavioral trajectories of workers at power engineering sites.