Scholarlink

<p>Monitoring surface damage in hydraulic structures is critical for safety management, yet inspection outputs often remain difficult to standardize and trace because detected cracks are rarely linked to specific structural components. This paper presents an Intelligent Damage Monitoring System (IDMS) that converts unmanned aerial vehicle (UAV) and ground-based red–green–blue (RGB) images into component-attributed and auditable crack records. The main contributions are: (1) a detection–segmentation collaborative framework that runs structural component detection and crack instance segmentation in parallel; (2) a post-inference association and validation module that assigns each crack instance to its most plausible component using IoU-based spatial consistency and engineering-guided semantic validity rules to reject implausible component–damage pairs; and (3) a structured reporting pipeline that further merges redundant crack reports within the same component and exports evidence-preserving visualization overlays with machine-readable outputs for documentation and review. Experiments on reservoir inspection imagery demonstrate near-real-time processing (0.38&#xa0;s per image). On a real-crack test set, the system achieves a 100% collaborative activation rate, 83.33% component-level localization accuracy, and 53.57% end-to-end instance-level overall accuracy, supporting practical and traceable crack inspection for hydraulic-structure maintenance.</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

Intelligent damage monitoring system for hydraulic structures with the detection–segmentation collaborative framework

  • 期刊论文

  • 发表日期：

    2026年7月17日

查看全文
  • Xinxin Jin,
  • Chang Zhou,
  • Mengyao Shen,
  • Ding Nie,
  • Xiaorong Xu,
  • Hui Jiang

摘要

Monitoring surface damage in hydraulic structures is critical for safety management, yet inspection outputs often remain difficult to standardize and trace because detected cracks are rarely linked to specific structural components. This paper presents an Intelligent Damage Monitoring System (IDMS) that converts unmanned aerial vehicle (UAV) and ground-based red–green–blue (RGB) images into component-attributed and auditable crack records. The main contributions are: (1) a detection–segmentation collaborative framework that runs structural component detection and crack instance segmentation in parallel; (2) a post-inference association and validation module that assigns each crack instance to its most plausible component using IoU-based spatial consistency and engineering-guided semantic validity rules to reject implausible component–damage pairs; and (3) a structured reporting pipeline that further merges redundant crack reports within the same component and exports evidence-preserving visualization overlays with machine-readable outputs for documentation and review. Experiments on reservoir inspection imagery demonstrate near-real-time processing (0.38 s per image). On a real-crack test set, the system achieves a 100% collaborative activation rate, 83.33% component-level localization accuracy, and 53.57% end-to-end instance-level overall accuracy, supporting practical and traceable crack inspection for hydraulic-structure maintenance.