Reproduce “J-DDL: Surface Damage Detection and Localization System for Fighter Aircraft” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jin Huang, Mingqiang Wei, Zikuan Li, Hangyu Qu, Wei Zhao, Xinyu Bai
J-DDL combines non-contact cameras and laser scanners with a lightweight YOLO-based detector to identify and spatially register visible damage on fighter-aircraft surfaces. On the AIRSD dataset, its AIR-YOLO model reaches 65.2% mAP at a reported 110 FPS, while the integrated workflow detects and localizes substantially more candidate defects in a shorter inspection session than a single manual inspector.
Key insights
- The system integrates recognition and geometric localization without requiring a learned multimodal network: 2D detection identifies the defect, while sensor calibration supplies its spatial position.
- Lightweight backbone changes reduce backbone parameters by about 30%, while attention and the modified regression loss recover and improve detection performance over the YOLOv8 baseline.
- Performance is strongly defect-dependent: missing fasteners and rivet damage are easier to detect, whereas scratches and paint peeling remain difficult because of their narrow, variable, or low-contrast appearance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.