Scholarlink

<p>Road cracks not only pose traffic safety hazards but also further damage pavement structures and increase maintenance costs. Current mainstream crack detection methods struggle with accuracy in complex environments, high computational requirements, and real-time deployment challenges. To address these issues, this paper proposes an improved RT-DETR-based detection algorithm. A new PCAM (PKI C2f Attention Module) Backbone is introduced to enhance the model’s ability to perceive multiscale features. In addition, CGA (Cascaded Group Attention) replaces Multi-Head Self-Attention to aggregate features layer by layer, reducing computational load and enhancing local feature representation. The DBBC3 (Diverse Branch Block C3) module optimizes feature extraction and improves robustness against objects with varying angles and scales. Finally, combining FoclarIoU and MPDIoU into a new Foclar-MPDIoU replaces the original GIoU, accelerating convergence and improving the accuracy of fracture boundary box localization. We validated our method on the RDD2022 dataset (China subset). The experimental results showed that the improved model achieved an <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4165_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="50" /> </InlineMediaObject> <EquationSource Format="TEX">\(\text {mAP}_{50}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>mAP</mtext> <mn>50</mn> </msub> </math></EquationSource> </InlineEquation> of 84.1%, representing a 2.44% increase over the original model, while reducing the number of parameters by 12.5% and significantly improving the inference frame rate, thus meeting real-time detection requirements.</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

Road crack detection based on improved RT-DETR

  • 期刊论文

  • 发表日期：

    2025年5月13日

查看全文
  • Guangyuan Zhao,
  • Weilin Zhang,
  • Rui Sun,
  • Tong Wei

摘要

Road cracks not only pose traffic safety hazards but also further damage pavement structures and increase maintenance costs. Current mainstream crack detection methods struggle with accuracy in complex environments, high computational requirements, and real-time deployment challenges. To address these issues, this paper proposes an improved RT-DETR-based detection algorithm. A new PCAM (PKI C2f Attention Module) Backbone is introduced to enhance the model’s ability to perceive multiscale features. In addition, CGA (Cascaded Group Attention) replaces Multi-Head Self-Attention to aggregate features layer by layer, reducing computational load and enhancing local feature representation. The DBBC3 (Diverse Branch Block C3) module optimizes feature extraction and improves robustness against objects with varying angles and scales. Finally, combining FoclarIoU and MPDIoU into a new Foclar-MPDIoU replaces the original GIoU, accelerating convergence and improving the accuracy of fracture boundary box localization. We validated our method on the RDD2022 dataset (China subset). The experimental results showed that the improved model achieved an \(\text {mAP}_{50}\) mAP 50 of 84.1%, representing a 2.44% increase over the original model, while reducing the number of parameters by 12.5% and significantly improving the inference frame rate, thus meeting real-time detection requirements.