Generalizable Unsupervised Array-SAR Restoration With Dual-Stage Refinement
IEEE Transactions on Aerospace and Electronic Systems · 2026 · 62 · 12329-12347
DOI: 10.1109/taes.2026.3699494 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Aerospace and Electronic Systems, 2026. Verfasst von Yangyang Wang und 5 weiteren Autor:innen. Die Arbeit nennt 61 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
Array synthetic aperture radar (array-SAR) systems are increasingly employed for scene sensing, such as security inspection and close-range restoration, due to their ability to exploit spatial diversity for enhanced resolution. However, the performance of traditional restoration methods is often limited by measurement noise, sparse sampling, and the reliance on large amounts of labeled datasets. Although regularization-based methods can improve synthetic aperture radar image quality to a certain extent, they suffer from difficulties in designing appropriate regularization functions for different features of targets and in tuning associated parameters. These constraints limit their applicability in practical scenarios. To address the above challenges, we propose a generalizable unsupervised restoration method for array-SAR with a two-stage coarse-to-fine refinement network driven by range-null space decomposition. The proposed framework initially extracts low-dimensional range components, followed by high-fidelity restoration using a cascaded encoder–decoder architecture. To enable unsupervised model learning, an unsupervised loss function is incorporated to enforce consistency with the measurement data. Furthermore, a recorruption-based strategy is integrated to generate virtual samples during training, effectively mimicking the advantages of supervised learning while eliminating the dependence on ground truth. Extensive simulations and real-world experiments demonstrate that the proposed method exhibits strong adaptability across diverse environments and target types, particularly in scenarios involving distributed or non-point-like targets.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Wang, Y., He, Z., Zhan, X., Zhang, C., Liu, Y., & Wei, S. (2026). Generalizable Unsupervised Array-SAR Restoration With Dual-Stage Refinement. IEEE Transactions on Aerospace and Electronic Systems, 62, 12329-12347. https://doi.org/10.1109/taes.2026.3699494
Vancouver
Wang Y, He Z, Zhan X, Zhang C, Liu Y, Wei S. Generalizable Unsupervised Array-SAR Restoration With Dual-Stage Refinement. IEEE Transactions on Aerospace and Electronic Systems. 2026;62:12329-12347. doi:10.1109/taes.2026.3699494
BibTeX
@article{Wang2026,
author = {Wang, Yangyang and He, Zhiming and Zhan, Xu and Zhang, Chudi and Liu, Yuxuan and Wei, Shunjun},
title = {Generalizable Unsupervised Array-SAR Restoration With Dual-Stage Refinement},
journal = {IEEE Transactions on Aerospace and Electronic Systems},
year = {2026},
volume = {62},
pages = {12329--12347},
doi = {10.1109/taes.2026.3699494}
}
EUVIMED