Scholarlink

<p>The accurate detection of pavement distresses is often hindered by challenges such as multi-scale feature variation, complex backgrounds, and real-time constraints. To address these issues, this study proposes YOLO-ELC, an enhanced version of the YOLOv8 algorithm tailored for UAV-captured images of road surfaces. The novelty of YOLO-ELC lies in three innovations: (1) the Efficient Multi-Scale Convolution module (C2f_EMC), which strengthens scale-sensitive feature extraction; (2) the Local Spatial Kernels Aggregation module (SPPF_LA), which improves robustness against background interference; and (3) the Content-Aware Reassembly of Features module (CRF), which preserves semantic detail during upsampling. Together, these improvements enable more accurate recognition of diverse crack patterns under complex real-world conditions. Experiments on a UAV dataset confirm that YOLO-ELC achieves substantially higher accuracy and recall than the baseline YOLOv8, while maintaining real-time processing capability. These contributions provide a practical and scalable solution for infrastructure monitoring, supporting timely maintenance decisions and enhanced road safety.</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

YOLO-ELC: Enhanced Detection of Pavement Distresses Using UAV-Captured Images

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Junfei Zhang,
  • Huisheng Cheng,
  • Yinhang Gao,
  • Yuhang Wang

摘要

The accurate detection of pavement distresses is often hindered by challenges such as multi-scale feature variation, complex backgrounds, and real-time constraints. To address these issues, this study proposes YOLO-ELC, an enhanced version of the YOLOv8 algorithm tailored for UAV-captured images of road surfaces. The novelty of YOLO-ELC lies in three innovations: (1) the Efficient Multi-Scale Convolution module (C2f_EMC), which strengthens scale-sensitive feature extraction; (2) the Local Spatial Kernels Aggregation module (SPPF_LA), which improves robustness against background interference; and (3) the Content-Aware Reassembly of Features module (CRF), which preserves semantic detail during upsampling. Together, these improvements enable more accurate recognition of diverse crack patterns under complex real-world conditions. Experiments on a UAV dataset confirm that YOLO-ELC achieves substantially higher accuracy and recall than the baseline YOLOv8, while maintaining real-time processing capability. These contributions provide a practical and scalable solution for infrastructure monitoring, supporting timely maintenance decisions and enhanced road safety.