Scholarlink

<p>Road cracks constitute a predominant form of pavement distress, significantly impacting durability and traffic safety. However, crack detection remains challenging due to interference from significant scale variations and complex backgrounds. Timely detection is a critical prerequisite for effective road maintenance. To address these challenges, this paper presents ReLA-YOLO, an enhanced road crack detection model based on the YOLOv11n architecture. First, the proposed model integrates a structure-reparameterized multi-scale parallel convolution module coupled with a hybrid attention mechanism (ReLA). This integration effectively captures global contextual information of cracks across diverse scales while suppressing background interference. Second, within the backbone network, receptive field attention convolution (RFCAConv) is introduced to optimize the downsampling process, thereby enhancing crack feature extraction. Finally, a simplified spatial pyramid pooling fast module (SimSPPF) is employed in the feature pyramid layer to improve computational efficiency. Experimental validation on the RDD2022 dataset demonstrates that ReLA-YOLO achieves a mean average precision (mAP) of 87.8%, surpassing the baseline model by 2.6%. These results indicate that the proposed ReLA-YOLO model provides an effective solution to the challenges posed by significant scale variations and complex backgrounds in road crack detection, exhibiting high potential for practical engineering 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

Road crack detection algorithm based on fusion structure re-parameterization with multi-scale parallel convolutions and hybrid attention mechanism

  • 期刊论文

  • 发表日期：

    2025年9月17日

查看全文
  • Shanqiang Li,
  • Zhiqiang Lin,
  • Yujing Shi,
  • Junjie Lan,
  • Yu Zhuo

摘要

Road cracks constitute a predominant form of pavement distress, significantly impacting durability and traffic safety. However, crack detection remains challenging due to interference from significant scale variations and complex backgrounds. Timely detection is a critical prerequisite for effective road maintenance. To address these challenges, this paper presents ReLA-YOLO, an enhanced road crack detection model based on the YOLOv11n architecture. First, the proposed model integrates a structure-reparameterized multi-scale parallel convolution module coupled with a hybrid attention mechanism (ReLA). This integration effectively captures global contextual information of cracks across diverse scales while suppressing background interference. Second, within the backbone network, receptive field attention convolution (RFCAConv) is introduced to optimize the downsampling process, thereby enhancing crack feature extraction. Finally, a simplified spatial pyramid pooling fast module (SimSPPF) is employed in the feature pyramid layer to improve computational efficiency. Experimental validation on the RDD2022 dataset demonstrates that ReLA-YOLO achieves a mean average precision (mAP) of 87.8%, surpassing the baseline model by 2.6%. These results indicate that the proposed ReLA-YOLO model provides an effective solution to the challenges posed by significant scale variations and complex backgrounds in road crack detection, exhibiting high potential for practical engineering applications.