Automated Multiclass Damage Detection in RC Bridges Using Drone Imagery and Deep Learning Models

Bridges are critical infrastructure components that require continuous inspection to ensure their safety and longevity. Traditional visual inspection methods are time-consuming, costly, and prone to subjectivity. This study proposes an automated multiclass damage detection framework for reinforced concrete (RC) bridges using drone imagery and deep learning models. High-resolution images captured by Unmanned Aerial Vehicles (UAVs) were processed using YOLOv11 and RT-DETR architectures to detect and classify three primary damage types: concrete cracks, rebar corrosion, and spalling. The dataset, consisting of 596 high-resolution images from bridges, was meticulously annotated to enhance model accuracy. Comparative analysis revealed that RT-DETR with the Stochastic Gradient Descent (SGD) optimizer outperformed YOLOv11, achieving the highest mean Average Precision (mAP) of 0.40. While YOLOv11 demonstrated strong precision, RT-DETR excelled in recall and robustness, particularly for detecting small and dispersed damage regions. These findings highlight the effectiveness of transformer-based architectures in structural health monitoring and highlight the potential of UAV-assisted inspections to improve efficiency and accuracy in bridge maintenance. Future work will explore transitioning from rectangular bounding boxes to pixel-level segmentation for even greater damage characterization accuracy.