EUVIMEDEUVIMED entdecken
Crossref

Kernel Adaptive Deformable Convolution for Remote Sensing Object Detection

Kai Hu, An Luo, Kai Jiang, Xieping Gao

IEEE Transactions on Pattern Analysis and Machine Intelligence · 2026 · 1-16

DOI: 10.1109/tpami.2026.3752695 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026. Verfasst von Kai Hu und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Modeling geometric variations, particularly the extreme fluctuations in scale and shape, stands as a persistent challenge in computer vision, which is significantly exacerbated in remote sensing object detection (RSOD). Existing methods are often constrained by the fixed geometric topology of standard convolutions or the insufficient representation power of linear feature aggregation, making it difficult to strike an ideal balance between adaptability and computational efficiency. To address these challenges, we propose a novel scale- and shape-adaptive method named Kernel Adaptive Deformable Convolution (KAD-Conv). Specifically, KAD-Conv employs a specialized Deformable Spatial Convolution (DSC) at its core to adaptively adjust sampling locations, while reformulating the mask generation process to fully leverage the mask's potential for discriminative feature modulation. In KAD-Conv, we further introduce a Scale-aware Kernel Adaptive mechanism, which enables the dynamic adjustment of sampling scales to effectively capture object representations across extreme scale variations without increasing kernel size. With KAD-Conv, we develop a novel object detection network based on Multi-task Guided Learning (MGL), called KAD-CNet. Extensive experiments conducted on widely used remote sensing benchmark datasets, including DOTA-v1.0, DIOR, and NWPU VHR-10, along with validation on natural scenes exhibiting extreme geometric variations, have validated the effectiveness of the proposed KAD-CNet. Moreover, comprehensive comparisons against various YOLO-series models further underscore the practical value and efficiency of the proposed approach.

Zitieren

APA

Hu, K., Luo, A., Jiang, K., & Gao, X. (2026). Kernel Adaptive Deformable Convolution for Remote Sensing Object Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1-16. https://doi.org/10.1109/tpami.2026.3752695

Vancouver

Hu K, Luo A, Jiang K, Gao X. Kernel Adaptive Deformable Convolution for Remote Sensing Object Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2026:1-16. doi:10.1109/tpami.2026.3752695

BibTeX
@article{Hu2026,
  author = {Hu, Kai and Luo, An and Jiang, Kai and Gao, Xieping},
  title = {Kernel Adaptive Deformable Convolution for Remote Sensing Object Detection},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2026},
  pages = {1--16},
  doi = {10.1109/tpami.2026.3752695}
}
Kostenlos starten →