Abstract :

Unmanned Aerial Vehicle (UAV)-based Road Inspection Has Emerged As An Efficient Solution For Large-scale Infrastructure Monitoring. However, Accurate Detection Of Road Damages Such As Cracks, Potholes, And Surface Deformities From Aerial Imagery Remains Challenging Due To Small Object Size, Varying Illumination, And Complex Backgrounds. In This Work, A Lightweight Hybrid Attention-based Framework Is Proposed For Real-time UAV Road Damage Detection. The Proposed Method Builds Upon The YOLOv8n Object Detection Architecture And Introduces Three Key Enhancements: (i) Squeeze-and-Excitation (SE) Attention For Channel-wise Feature Recalibration, (ii) Deformable Convolution-based Refinement To Improve Spatial Adaptability, And (iii) Test-time Augmentation (TTA) Combined With Weighted Box Fusion (WBF) To Enhance Detection Robustness. These Components Collectively Improve The Model’s Ability To Detect Fine-grained Damages Under Diverse Environmental Conditions While Maintaining Computational Efficiency. The Model Is Trained And Evaluated On The RDD2022 Dataset, Using A Lightweight Configuration Suitable For Real-time Deployment. Experimental Results Demonstrate That The Proposed Framework Achieves Improved Detection Accuracy Compared To Baseline YOLO Models, Particularly In Identifying Small And Irregular Road Damages. Additionally, The System Maintains A Favorable Balance Between Accuracy And Inference Speed, Making It Suitable For UAV-based Real-time Applications. The Proposed Framework Contributes An Effective And Practical Solution For Intelligent Road Monitoring Systems, Supporting Automated Maintenance Planning And Enhancing Transportation Safety In Smart City Environments.

Published:

07-6-2026

Issue:

Vol. 26 No. 6 (2026)

Page Nos:

422-431

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite