PAPER P02 · 2018
Autonomous Structural Visual Inspection Using Region-Based Deep Learning for Detecting Multiple Damage Types
Cha, Y.-J., Choi, W., Suh, G., Mahmoudkhani, S., Büyüköztürk, O. — Computer-Aided Civil and Infrastructure Engineering 33(9):731-747
Abstract read Citations: openalex: 1,540 · crossref: 1,413 · semanticScholar: 1,454 (2026-10-22)
Adapts Faster R-CNN object detection to find several damage types at once rather than cracks alone: concrete cracks, medium and high steel corrosion, bolt corrosion and steel delamination. Trained on a labelled image set, it runs fast enough per image to underpin a video-based damage detection framework. The abstract does not frame it as a UAV study.
Key findings
- Mean average precision 87.8% across five damage classes (class APs 90.6, 83.4, 82.1, 98.1, 84.7%)
- 0.03 s per 500x375 image, enabling a near real-time video framework
- Dataset of 2,366 labelled images
Limitations
- Abstract only reviewed
- Not described as UAV imagery; robustness test used only 11 new images
- Bounding boxes, not damage size or severity measurement