Scholarlink

<p>Fishing and anchoring activities within exclusion zones can damage critical subsea infrastructure, posing significant risks. To enhance proactive protection, a novel vessel behavior recognition framework was developed by utilizing automatic identification system (AIS) trajectory data to classify four fishing vessel types (trawlers, gill netters, purse seiners, and canvas stow netters). We applied our refined framework and methodology to process 397891 AIS records from 24 vessels in the East China Sea and the Yellow Sea. This framework incorporates three key innovations: voyage-based data segmentation, feature optimization guided by large language models (LLMs), and automated hyperparameter tuning. Data segmentation preserves operational integrity. Feature optimization reduces the number of features from 45 initial indicators to 28 core discriminative features. The resulting refined model, termed light gradient boosting machine_lv, demonstrated excellent performance, reaching an accuracy of 94.1%. Model reliability was rigorously validated by examining <i>Precision</i>-<i>Recall</i> curves and through feature importance analysis. Notably, the strategic integration of LLMs accelerated the research process by approximately 80%. The proposed framework acts as a robust and efficient tool for the dynamic risk assessment and proactive safeguarding of subsea facilities.</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

A behavior recognition model for fishing vessels using limited AIS trajectory dataset

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月19日

查看全文
  • Tianzhi Wang,
  • Feng Lyu,
  • Chao Chen,
  • Qiang Jing,
  • Lei Han

摘要

Fishing and anchoring activities within exclusion zones can damage critical subsea infrastructure, posing significant risks. To enhance proactive protection, a novel vessel behavior recognition framework was developed by utilizing automatic identification system (AIS) trajectory data to classify four fishing vessel types (trawlers, gill netters, purse seiners, and canvas stow netters). We applied our refined framework and methodology to process 397891 AIS records from 24 vessels in the East China Sea and the Yellow Sea. This framework incorporates three key innovations: voyage-based data segmentation, feature optimization guided by large language models (LLMs), and automated hyperparameter tuning. Data segmentation preserves operational integrity. Feature optimization reduces the number of features from 45 initial indicators to 28 core discriminative features. The resulting refined model, termed light gradient boosting machine_lv, demonstrated excellent performance, reaching an accuracy of 94.1%. Model reliability was rigorously validated by examining Precision-Recall curves and through feature importance analysis. Notably, the strategic integration of LLMs accelerated the research process by approximately 80%. The proposed framework acts as a robust and efficient tool for the dynamic risk assessment and proactive safeguarding of subsea facilities.