Scholarlink

<p>The accurate detection of pavement deterioration is critical for evaluating road conditions and serves as a fundamental component for decision-making within Pavement Management Systems (PMS), particularly amidst escalating traffic volumes. Unmanned Aerial Vehicle (UAV) photogrammetry provides a highly efficient approach for mapping critical flexible pavement distresses, enabling the high-resolution acquisition and spatial analysis of various crack morphologies. This study proposes a deep learning framework to autonomously detect asphalt pavement cracks using UAV-acquired imagery. The developed model leverages an encoder-decoder convolutional architecture, incorporating an innovative feature fusion mechanism between the encoding and decoding pathways to optimize detection precision. Consequently, this methodology achieves robust pixel-level crack segmentation, yielding superior accuracy compared to conventional architectures in the field. The proposed network was trained and evaluated using the publicly available CrackDataset. Furthermore, this research benchmarked the developed model against several established semantic segmentation networks, specifically CrackForest, SegNet, U-Net, PSPNet, DeepCrack, and CrackSeg. The performance assessment incorporated quantitative metrics, including Overall Accuracy (OA), Precision, Recall, F1-score, and mean Intersection over Union (mIoU). The empirical results demonstrate that the proposed framework attains an exceptional crack segmentation accuracy ranging from 96% to 99%, substantially outperforming existing methods in the automated extraction and condition assessment of crack-induced pavement distresses.</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

Deep learning-based pavement crack segmentation using UAV imagery

  • 期刊论文

  • 发表日期：

    2026年3月18日

查看全文
  • Seyed Arya Fakhri,
  • Mehran Satari,
  • Mohammad Taherkhani

摘要

The accurate detection of pavement deterioration is critical for evaluating road conditions and serves as a fundamental component for decision-making within Pavement Management Systems (PMS), particularly amidst escalating traffic volumes. Unmanned Aerial Vehicle (UAV) photogrammetry provides a highly efficient approach for mapping critical flexible pavement distresses, enabling the high-resolution acquisition and spatial analysis of various crack morphologies. This study proposes a deep learning framework to autonomously detect asphalt pavement cracks using UAV-acquired imagery. The developed model leverages an encoder-decoder convolutional architecture, incorporating an innovative feature fusion mechanism between the encoding and decoding pathways to optimize detection precision. Consequently, this methodology achieves robust pixel-level crack segmentation, yielding superior accuracy compared to conventional architectures in the field. The proposed network was trained and evaluated using the publicly available CrackDataset. Furthermore, this research benchmarked the developed model against several established semantic segmentation networks, specifically CrackForest, SegNet, U-Net, PSPNet, DeepCrack, and CrackSeg. The performance assessment incorporated quantitative metrics, including Overall Accuracy (OA), Precision, Recall, F1-score, and mean Intersection over Union (mIoU). The empirical results demonstrate that the proposed framework attains an exceptional crack segmentation accuracy ranging from 96% to 99%, substantially outperforming existing methods in the automated extraction and condition assessment of crack-induced pavement distresses.