Spatial prior and defect co-occurrence constraint fusion for vibration damper defect detection in power lines
Measurement Science and Technology · 06.11.2026 · 37 (44) · 446105
DOI: 10.1088/1361-6501/aeb03c ↗Kurz gefasst
Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 05.11.2026. Verfasst von Zhiwei Huang und 5 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben. Gefördert von: Joint Fund of the National Natural Science Foundation of China and the Karst Science Research Center of Guizhou Province, The Natural Science Foundation of Shandong Province.
Abstract (Originalsprache)
In UAV-based power line inspection, vibration dampers occupy only a tiny fraction of pixels and are surrounded by cluttered backgrounds, leading to under-utilized spatial priors, high false-positive localization errors, and severe category confusion. To address these challenges, we present a novel framework, termed PGMD-Net, which integrates spatial layout priors and defect co-occurrence constraints to systematically infuse domain knowledge into fine-grained object detection. First, the Spatial Prior Aware Hybrid Extraction Module translates physical power line layout rules into deterministic spatial heatmaps that explicitly highlight damper-probable regions. Second, the Hybrid Attention Fusion Localization module employs an attention-gated mechanism to merge these geometric priors with data-driven convolutional features, effectively injecting the physical constraints of damper distributions into the network hypothesis space to achieve prior-driven, precise target localization. Finally, the Co-occurrence Adjacency Graph Classification module leverages an empirical defect co-occurrence matrix to initialize a prior-weighted adjacency matrix, performing relation-aware graph convolutions to propagate inter-defect semantic relational constraints for robust fine-grained recognition. Extensive experiments on our self-constructed UAV dataset demonstrate that PGMD-Net attains 86.9% precision, 83.9% recall, 89.0% mAP@0.5, and 61.4% mAP@[0.5:0.95], achieving competitive performance against existing object detection methods and confirming the effectiveness of physical knowledge injection for complex industrial defect detection tasks.
Förderung
- Joint Fund of the National Natural Science Foundation of China and the Karst Science Research Center of Guizhou Province
- The Natural Science Foundation of Shandong Province
Zitieren
APA
Huang, Z., Zou, G., Fu, G., Chen, X., Wang, W., & Qiu, H. (2026). Spatial prior and defect co-occurrence constraint fusion for vibration damper defect detection in power lines. Measurement Science and Technology, 37(44), 446105. https://doi.org/10.1088/1361-6501/aeb03c
Vancouver
Huang Z, Zou G, Fu G, Chen X, Wang W, Qiu H. Spatial prior and defect co-occurrence constraint fusion for vibration damper defect detection in power lines. Measurement Science and Technology. 2026;37(44):446105. doi:10.1088/1361-6501/aeb03c
BibTeX
@article{Huang2026,
author = {Huang, Zhiwei and Zou, Guofeng and Fu, Guixia and Chen, Xiangpeng and Wang, Weijian and Qiu, Hao},
title = {Spatial prior and defect co-occurrence constraint fusion for vibration damper defect detection in power lines},
journal = {Measurement Science and Technology},
year = {2026},
volume = {37},
number = {44},
pages = {446105},
doi = {10.1088/1361-6501/aeb03c}
}
EUVIMED