Scholarlink

<p>Improper procedures during oil tank unloading operations can lead to catastrophic accidents. To address these issues of low real-time performance and poor interpretability in existing oil tanker unloading detection systems, we propose a novel framework: YOLO-ESC. This framework monitors safety compliance via multimodal feature fusion and knowledge reasoning, enabling it to determine whether ongoing oil unloading operations comply with prescribed safety protocols and generate explainable warnings when violations are detected. Specifically, YOLO-ESC leverages a Multimodal Feature Processing Network (MFPN) with visual-semantic alignment to improve fine-grained recognition of equipment states, a Human-Equipment Interaction Semantic Modeling Module (HEISMM) to identify structured interaction events through spatial-temporal analysis, and a Compliance Judgment Expert System (CJES) that performs rule-based reasoning with explainability. It achieves real-time performance at 31 FPS, a rate that ensures seamless operation in safety-critical environments, while maintaining 92.3% compliance classification accuracy. Unlike many approaches that rely on large, complex vision-language models, YOLO-ESC instead leverages explicit knowledge encoding, making it both efficient and interpretable.</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

Safety compliance monitoring for oil tank unloading based on multimodal feature fusion and knowledge reasoning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月9日

查看全文
  • Tiechao Liu,
  • Chao Sun,
  • Bo Wang,
  • Lichao Yao

摘要

Improper procedures during oil tank unloading operations can lead to catastrophic accidents. To address these issues of low real-time performance and poor interpretability in existing oil tanker unloading detection systems, we propose a novel framework: YOLO-ESC. This framework monitors safety compliance via multimodal feature fusion and knowledge reasoning, enabling it to determine whether ongoing oil unloading operations comply with prescribed safety protocols and generate explainable warnings when violations are detected. Specifically, YOLO-ESC leverages a Multimodal Feature Processing Network (MFPN) with visual-semantic alignment to improve fine-grained recognition of equipment states, a Human-Equipment Interaction Semantic Modeling Module (HEISMM) to identify structured interaction events through spatial-temporal analysis, and a Compliance Judgment Expert System (CJES) that performs rule-based reasoning with explainability. It achieves real-time performance at 31 FPS, a rate that ensures seamless operation in safety-critical environments, while maintaining 92.3% compliance classification accuracy. Unlike many approaches that rely on large, complex vision-language models, YOLO-ESC instead leverages explicit knowledge encoding, making it both efficient and interpretable.