Scholarlink

<p>Crack detection plays a vital role in ensuring the structural integrity and safety of infrastructures, such as roads and buildings. However, traditional manual inspection methods are inefficient, and current machine learning approaches require large amounts of labeled data, which is costly and time-consuming to obtain. In this paper, we propose a Mixed Supervised Learning (MSL) algorithm that reduces the need for extensive labeled data sets by combining several learning techniques, and an Efficient Multiscale Transformer version 2 (EMTv2) model, that incorporates multiscale feature extraction and novel attention mechanisms to enhance crack classification and segmentation performance. The MSL algorithm achieves competitive segmentation results using only <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11554_2025_1678_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="31" /> </InlineMediaObject> <EquationSource Format="TEX">\(25\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>25</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> of the image-level labeled data required by traditional weakly supervised learning methods, outperforming existing models on four data sets in multiple metrics. Our approach demonstrates significant improvements in both efficiency, and segmentation quality, offering a more practical solution for real-world crack detection applications.</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 detection with minimal labels: a mixed supervision approach with multiscale transformers

  • 期刊论文

  • 发表日期：

    2025年5月4日

查看全文
  • Omar Al-maqtari,
  • Bo Peng,
  • Zaid Al-Huda,
  • Ali Rahman

摘要

Crack detection plays a vital role in ensuring the structural integrity and safety of infrastructures, such as roads and buildings. However, traditional manual inspection methods are inefficient, and current machine learning approaches require large amounts of labeled data, which is costly and time-consuming to obtain. In this paper, we propose a Mixed Supervised Learning (MSL) algorithm that reduces the need for extensive labeled data sets by combining several learning techniques, and an Efficient Multiscale Transformer version 2 (EMTv2) model, that incorporates multiscale feature extraction and novel attention mechanisms to enhance crack classification and segmentation performance. The MSL algorithm achieves competitive segmentation results using only \(25\%\) 25 % of the image-level labeled data required by traditional weakly supervised learning methods, outperforming existing models on four data sets in multiple metrics. Our approach demonstrates significant improvements in both efficiency, and segmentation quality, offering a more practical solution for real-world crack detection applications.