Answers are grounded in this paper and cite their sources.
Unmanned aerial vehicle (UAV) imaging offers a non-contact approach to wind-turbine bladesurface assessment, but partial blade views make reliable defect localization difficult from visual evidence. Structural damage may be truncated by blade boundaries or image cropping, whereas small contamination defects often produce weak local responses. These conditions require compact detectors to preserve fine texture cues while incorporating larger blade context. We propose a feature-pyramid diffusion network (FPD-YOLO) for UAV image-based blade-defect localization. The network combines multi-receptive-field encoding, progressive context aggregation, and direct cross-scale feature exchange, so that local defect evidence is retained during feature aggregation and incomplete damage regions receive broader contextual support. Evaluation on a DTU-derived blade-defect dataset and the Wind Turbine Blade Defect dataset demonstrates consistent defect retrieval and image-space localization under partial views, weak textures, scale variation, and blade-background interference. Compared with representative detection methods, FPD-YOLO retrieves small contamination defects more reliably and localizes incomplete structural damage more stably. The method provides spatially reviewable defect evidence for UAV-assisted non-contact blade-surface assessment, strengthening the image-based measurement stage of wind-turbine inspection.
No takes yet. Share an insight, caveat, or question.
Shi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: