Scholarlink

<p>Detecting cracks in concrete is critical for ensuring structural safety, as undetected flaws can lead to costly repairs or catastrophic failures. This research introduces a novel method that integrates graph representation learning with Gray-level co-occurrence matrices (GLCM) to detect cracks in concrete surfaces. Images are preprocessed using adaptive gamma correction with weighted distribution (AGCWD) to improve crack visibility under varying image quality and noise conditions. GLCM extracts texture features, which are then used to build graph representations-images as nodes linked by similarity weights-capturing intricate crack patterns. Graph convolutional networks (GCN) classify these graphs, achieving a remarkable 98.99% accuracy on a 40,000-image dataset. This outperforms traditional and many deep-learning methods, offering greater reliability with less reliance on extensive labeled data. The approach’s scalability and interpretability make it well-suited for practical applications, such as real-time bridge and pavement inspections. This work highlights the power of combining graph-based learning with texture analysis for robust crack detection.</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

Concrete crack detection via graph representation learning and texture analysis

  • 期刊论文

  • 发表日期：

    2025年7月29日

查看全文
  • Ramin Pourhanasa,
  • Ali Monadipour

摘要

Detecting cracks in concrete is critical for ensuring structural safety, as undetected flaws can lead to costly repairs or catastrophic failures. This research introduces a novel method that integrates graph representation learning with Gray-level co-occurrence matrices (GLCM) to detect cracks in concrete surfaces. Images are preprocessed using adaptive gamma correction with weighted distribution (AGCWD) to improve crack visibility under varying image quality and noise conditions. GLCM extracts texture features, which are then used to build graph representations-images as nodes linked by similarity weights-capturing intricate crack patterns. Graph convolutional networks (GCN) classify these graphs, achieving a remarkable 98.99% accuracy on a 40,000-image dataset. This outperforms traditional and many deep-learning methods, offering greater reliability with less reliance on extensive labeled data. The approach’s scalability and interpretability make it well-suited for practical applications, such as real-time bridge and pavement inspections. This work highlights the power of combining graph-based learning with texture analysis for robust crack detection.