EUVIMEDEUVIMED entdecken
Crossref

Background-Robust Knowledge Distillation for Forest Fire Detection in UAV Remote Sensing

Wentao Zhou, Deqiang He, Zhenzhen Jin, Qinzhan Wu, Shuqiong Dai

IEEE Geoscience and Remote Sensing Letters · 2026 · 23 · 2504205-2504205

DOI: 10.1109/lgrs.2026.3705482 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Geoscience and Remote Sensing Letters, 2026. Verfasst von Wentao Zhou und 4 weiteren Autor:innen. Die Arbeit nennt 16 Literaturangaben. Gefördert von: Major Science and Technology Project of Guangxi Province of China, Innovation Project of Guangxi Graduate Education and Major Talent Project in Guangxi Zhuang Autonomous Region.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Uncrewed aerial vehicle (UAV)-based remote sensing object detection is crucial for automated forest fire monitoring and Earth observation. However, existing algorithms for processing high-resolution aerial imagery are constrained by high computational complexity and generalization bottlenecks, making them difficult to deploy on edge devices. To address this, this letter proposes a background-robust knowledge distillation (BRKD) mechanism tailored for aerial scenarios. By introducing a cross-scale weighted soft foreground mask at the feature level and employing a teacher-confidence-based GIoU and encoding-difference-weighted distillation at the localization level, BRKD effectively suppresses the complex background clutter typical in remote sensing imagery and enhances object alignment. Systematic evaluations are conducted utilizing aerial remote sensing data collected by a custom-built UAV platform. Experimental results demonstrate that BRKD, incorporating model pruning, outperforms mainstream methods, improving average precision by 1.4% while reducing computational complexity by 33.7% and increasing inference speed by 43.5%. Real-world UAV deployment further verifies that the proposed model features low computational overhead and rapid inference, fulfilling the requirements for long-term, stable remote sensing observation on resource-constrained devices.

Förderung

  • Major Science and Technology Project of Guangxi Province of China
  • Innovation Project of Guangxi Graduate Education and Major Talent Project in Guangxi Zhuang Autonomous Region

Zitieren

APA

Zhou, W., He, D., Jin, Z., Wu, Q., & Dai, S. (2026). Background-Robust Knowledge Distillation for Forest Fire Detection in UAV Remote Sensing. IEEE Geoscience and Remote Sensing Letters, 23, 2504205-2504205. https://doi.org/10.1109/lgrs.2026.3705482

Vancouver

Zhou W, He D, Jin Z, Wu Q, Dai S. Background-Robust Knowledge Distillation for Forest Fire Detection in UAV Remote Sensing. IEEE Geoscience and Remote Sensing Letters. 2026;23:2504205-2504205. doi:10.1109/lgrs.2026.3705482

BibTeX
@article{Zhou2026,
  author = {Zhou, Wentao and He, Deqiang and Jin, Zhenzhen and Wu, Qinzhan and Dai, Shuqiong},
  title = {Background-Robust Knowledge Distillation for Forest Fire Detection in UAV Remote Sensing},
  journal = {IEEE Geoscience and Remote Sensing Letters},
  year = {2026},
  volume = {23},
  pages = {2504205--2504205},
  doi = {10.1109/lgrs.2026.3705482}
}
Kostenlos starten →