Scholarlink

<p>With the advancement of urbanization and the increase in transportation demand, the maintenance of urban infrastructure has become increasingly important. As a core component of urban transportation networks, road surfaces directly affect traffic safety and the service life of facilities. However, traditional road surface inspection often relies on manual inspection, which is inefficient and prone to errors. As a result, there is an urgent need to develop efficient technologies to detect road damage, which can achieve automation and improve accuracy. A new road crack detection method is developed based on the “You Only Look Once” version 5 (YOLOv5) architecture. The contextual information fusion module is applied to optimize the model ability to understand the relationships between different regions in the image. The attention aggregation module focuses on key areas that may contain damage. The adaptive refinement module dynamically adjusts the predicted bounding boxes of the network to better fit the actual damage boundaries. Experiments are carried out using the PD-Dataset and CRACK2000 datasets. The accuracy increased from 83.4 to 97.9%. The recognition accuracy of the testing set reached 98.0%. The recall rate increased from 0.882 to 0.964, and the F1 score enhanced from 0.885 to 0.972. In addition, compared with traditional deep learning networks, the improved model had an accuracy of over 96% in detecting construction joints and mesh cracks. This indicates that the method has significant advantages in accuracy, robustness, and real-time detection capability, which can better cope with complex pavement damage detection tasks.</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

Pavement Damage Detection Based on Embedded Design and Computer Vision

  • 期刊论文

  • 发表日期：

    2025年6月23日

查看全文
  • Xue Gao

摘要

With the advancement of urbanization and the increase in transportation demand, the maintenance of urban infrastructure has become increasingly important. As a core component of urban transportation networks, road surfaces directly affect traffic safety and the service life of facilities. However, traditional road surface inspection often relies on manual inspection, which is inefficient and prone to errors. As a result, there is an urgent need to develop efficient technologies to detect road damage, which can achieve automation and improve accuracy. A new road crack detection method is developed based on the “You Only Look Once” version 5 (YOLOv5) architecture. The contextual information fusion module is applied to optimize the model ability to understand the relationships between different regions in the image. The attention aggregation module focuses on key areas that may contain damage. The adaptive refinement module dynamically adjusts the predicted bounding boxes of the network to better fit the actual damage boundaries. Experiments are carried out using the PD-Dataset and CRACK2000 datasets. The accuracy increased from 83.4 to 97.9%. The recognition accuracy of the testing set reached 98.0%. The recall rate increased from 0.882 to 0.964, and the F1 score enhanced from 0.885 to 0.972. In addition, compared with traditional deep learning networks, the improved model had an accuracy of over 96% in detecting construction joints and mesh cracks. This indicates that the method has significant advantages in accuracy, robustness, and real-time detection capability, which can better cope with complex pavement damage detection tasks.