EUVIMEDEUVIMED entdecken
Crossref

HDP-BR-YOLO11n: A Lightweight Framework for Weak-Boundary Fluid Target Segmentation

Yue Xie, Chengtian Song, Keyu Xu, Hong Su, Tie Li

IEEE Signal Processing Letters · 2026 · 33 · 4151-4155

DOI: 10.1109/lsp.2026.3740635 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Yue Xie und 4 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Instance segmentation of targets with weak boundaries and severe gradient sparsity remains challenging, particularly under small-sample and highly homogeneous background constraints, where deep networks are prone to background-dependent shortcut learning and suffer from significant boundary localization degradation.This letter proposes HDP-BR-YOLO11n, a lightweight and task-adaptive framework for chaff cloud segmentation. A Heuristic Domain Perturbation (HDP) strategy expands the effective training distribution to mitigate inductive biases induced by fixed sensor viewpoints, thereby suppressing shortcut learning and enhancing cross-scenario robustness. Furthermore, a Boundary Refinement (BR) module integrating Retina Mask Reconstruction and Boundary Binary Cross-Entropy (BCE) supervision is proposed to impose explicit geometric constraints on sparse, fragmented edges, while introducing no additional inference overhead. Experiments on a challenging field-acquired chaff cloud dataset demonstrate that the proposed framework achieves a Mask mAP$_{50\text{-}95}$ of $0.7408\pm 0.0059$ with a compact footprint of 2.84 M parameters and 15.2 GFLOPs, outperforming the strongest competitor YOLO11 s by 2.7 points with merely 30% of its parameters, delivering an exceptional accuracy–efficiency balance for edge deployment.

Zitieren

APA

Xie, Y., Song, C., Xu, K., Su, H., & Li, T. (2026). HDP-BR-YOLO11n: A Lightweight Framework for Weak-Boundary Fluid Target Segmentation. IEEE Signal Processing Letters, 33, 4151-4155. https://doi.org/10.1109/lsp.2026.3740635

Vancouver

Xie Y, Song C, Xu K, Su H, Li T. HDP-BR-YOLO11n: A Lightweight Framework for Weak-Boundary Fluid Target Segmentation. IEEE Signal Processing Letters. 2026;33:4151-4155. doi:10.1109/lsp.2026.3740635

BibTeX
@article{Xie2026,
  author = {Xie, Yue and Song, Chengtian and Xu, Keyu and Su, Hong and Li, Tie},
  title = {HDP-BR-YOLO11n: A Lightweight Framework for Weak-Boundary Fluid Target Segmentation},
  journal = {IEEE Signal Processing Letters},
  year = {2026},
  volume = {33},
  pages = {4151--4155},
  doi = {10.1109/lsp.2026.3740635}
}
Kostenlos starten →