RLD-Net: an enhanced deep learning model for accurate pavement distress detection using UAV captured images

Accurate detection of pavement distresses, particularly cracks, is crucial for road maintenance and safety. However, traditional pavement damage detection methods are often difficult to effectively capture small-scale or subtle damage in complex and crowded environment, which limits the detection ac...

Full description

Saved in:
Bibliographic Details
Published in:Multiscale and Multidisciplinary Modeling, Experiments and Design Vol. 8; no. 8; p. 353
Main Authors: Cheng, Huisheng, Zhang, Junfei, Yan, Haohui, Wang, Yuhang, Li, Dong
Format: Journal Article
Language:English
Published: Cham Springer International Publishing 01.08.2025
Springer Nature B.V
Subjects:
Accuracy
Algorithms
Asphalt pavements
Background noise
Characterization and Evaluation of Materials
Complexity
Cracks
Damage detection
Deep learning
Efficiency
Engineering
Environmental conditions
Feature extraction
Inspections
Machine learning
Mathematical Applications in the Physical Sciences
Mechanical Engineering
Methods
Modules
Neural networks
Numerical and Computational Physics
Original Paper
Preventive maintenance
Recall
Road maintenance
Roads & highways
Simulation
Solid Mechanics
Traffic accidents & safety
Unmanned aerial vehicles
China
Unmanned aerial vehicle (UAV)
Road crack object-detection
Convolutional neural network (CNN)
Asphalt pavement distress
Object-detection algorithms
ISSN:2520-8160, 2520-8179
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Accurate detection of pavement distresses, particularly cracks, is crucial for road maintenance and safety. However, traditional pavement damage detection methods are often difficult to effectively capture small-scale or subtle damage in complex and crowded environment, which limits the detection accuracy and robustness. This study proposes RLD-Net, an enhanced detection framework based on the You Only Look Once Extended (YOLOX) model, aimed at improving the detection of pavement distresses, particularly road cracks. The RLD-Net incorporates three novel modules: the Regional Dynamic Perception (RDP) module for improving feature extraction, the Location Enhanced Attention Module (LEAM) for reducing background noise, and the Dynamic Context-Detail Fusion Module (DCFM) for multi-scale feature fusion. These modules work together to significantly improve the accuracy and efficiency of pavement damage detection. And compared with other models, it has better recognition and detection ability. The proposed framework was evaluated using unmanned aerial vehicle (UAV)-captured images under various environmental conditions. Experimental results demonstrate that RLD-Net significantly improves detection accuracy, achieving higher mean Average Precision at 0.5 Intersection over Union (mAP@0.5) scores and better Average Recall (AR) compared to the baseline YOLOX model. The fully integrated RLD-Net achieved an mAP@0.5 of 84.7% and an AR of 68.6%, marking a significant improvement in both precision and recall. Despite a slight increase in computational complexity, the RLD-Net maintains excellent performance with high speed.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2520-8160
2520-8179
DOI:10.1007/s41939-025-00950-9