Scholarlink

<p>Crack detection is a critical approach to ensuring the structural health of dams. However, challenges like uneven underwater lighting, sediment interference, and complex backgrounds often hinder traditional detection methods, leading to feature loss and false detections. To address these issues, this paper proposes Crack-ConvT Net, a U-Shape architecture that integrates Convolutional Neural Networks (CNNs) and Transformers for underwater dam crack segmentation. Firstly, a Global Information Aggregation Block is introduced to enhance the model’s ability to capture the spatial distribution of cracks by leveraging multiscale pooling and channel expansion strategies, which improve global context awareness while preserving high-resolution details. Secondly, to address the limitations of traditional skip connections in crack segmentation under complex environments, an Adaptive Feature Fusion Module is designed to optimize the interaction and integration of multi-level features. Finally, a Deep Prediction Head is developed, incorporating cascaded <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="40747_2025_1957_Article_IEq1.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="39" /> </InlineMediaObject> <EquationSource Format="TEX">\(3\times 3\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>3</mn> <mo>×</mo> <mn>3</mn> </mrow> </math></EquationSource> </InlineEquation> convolutions and Leaky ReLU activation functions to enhance the network’s capacity for modeling intricate crack features. Experimental results demonstrate that the proposed framework significantly outperforms existing methods in underwater dam crack segmentation, effectively improving segmentation accuracy under noise interference.</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

Crack-ConvT Net: A Convolutional Transformer Network for Crack Segmentation in Underwater Dams

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月23日

查看全文
  • Pengfei Shi,
  • Hongzhu Chen,
  • Zaiming Geng,
  • Xinnan Fan,
  • Yuanxue Xin

摘要

Crack detection is a critical approach to ensuring the structural health of dams. However, challenges like uneven underwater lighting, sediment interference, and complex backgrounds often hinder traditional detection methods, leading to feature loss and false detections. To address these issues, this paper proposes Crack-ConvT Net, a U-Shape architecture that integrates Convolutional Neural Networks (CNNs) and Transformers for underwater dam crack segmentation. Firstly, a Global Information Aggregation Block is introduced to enhance the model’s ability to capture the spatial distribution of cracks by leveraging multiscale pooling and channel expansion strategies, which improve global context awareness while preserving high-resolution details. Secondly, to address the limitations of traditional skip connections in crack segmentation under complex environments, an Adaptive Feature Fusion Module is designed to optimize the interaction and integration of multi-level features. Finally, a Deep Prediction Head is developed, incorporating cascaded \(3\times 3\) 3 × 3 convolutions and Leaky ReLU activation functions to enhance the network’s capacity for modeling intricate crack features. Experimental results demonstrate that the proposed framework significantly outperforms existing methods in underwater dam crack segmentation, effectively improving segmentation accuracy under noise interference.