BiKD: Bidomain Knowledge Distillation for SAR Object Detection
IEEE Geoscience and Remote Sensing Letters · 2026 · 23 · 1-5
DOI: 10.1109/lgrs.2026.3676449 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Geoscience and Remote Sensing Letters, 2026. Verfasst von Xueyi Wang und 5 weiteren Autor:innen. Die Arbeit nennt 24 Literaturangaben. Gefördert von: China National Funds.
Abstract (Originalsprache)
The synthetic aperture radar (SAR) imagery plays a pivotal role in remote sensing due to its all-weather and all-time imaging capabilities. Despite its advantages, deploying SAR object detection models on resource-limited platforms remains challenging. Existing knowledge distillation (KD) methods, primarily designed for natural images, fail to address the unique characteristics of SAR imagery, such as pronounced background clutter and pervasive speckle noise. To address these limitations, this letter proposes a novel bidomain KD (BiKD) framework tailored for SAR object detection. BiKD integrates frequency alignment distillation (FAD), which enhances discriminative structural representations by aligning high-frequency components between teacher and student detectors and spatial feature distillation (SFD), which improves core visual patterns learning by transferring channel-wise spatial attention distributions. Furthermore, a cross-head logit mimicking strategy is employed to alleviate target conflict during prediction supervision. Comprehensive experiments on the SSDD and SAR-AIRcraft-1.0 datasets demonstrate that BiKD consistently achieves superior performance, achieving significant improvements in detection accuracy while maintaining the computational efficiency.
Förderung
- China National Funds
Zitieren
APA
Wang, X., Xu, W., Li, G., Lin, K., Xu, Z., & Wu, Y. (2026). BiKD: Bidomain Knowledge Distillation for SAR Object Detection. IEEE Geoscience and Remote Sensing Letters, 23, 1-5. https://doi.org/10.1109/lgrs.2026.3676449
Vancouver
Wang X, Xu W, Li G, Lin K, Xu Z, Wu Y. BiKD: Bidomain Knowledge Distillation for SAR Object Detection. IEEE Geoscience and Remote Sensing Letters. 2026;23:1-5. doi:10.1109/lgrs.2026.3676449
BibTeX
@article{Wang2026,
author = {Wang, Xueyi and Xu, Wenjia and Li, Guangzuo and Lin, Kuan and Xu, Zhongqiu and Wu, Yirong},
title = {BiKD: Bidomain Knowledge Distillation for SAR Object Detection},
journal = {IEEE Geoscience and Remote Sensing Letters},
year = {2026},
volume = {23},
pages = {1--5},
doi = {10.1109/lgrs.2026.3676449}
}
EUVIMED