ARSAR-Net: Intelligent SAR Imaging with Adaptive Regularization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Fu, Shiping, Chen, Yufan, Zhang, Zhe, Qiu, Xiaolan, Ye, Qixiang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909660542402560
author Fu, Shiping
Chen, Yufan
Zhang, Zhe
Qiu, Xiaolan
Ye, Qixiang
author_facet Fu, Shiping
Chen, Yufan
Zhang, Zhe
Qiu, Xiaolan
Ye, Qixiang
contents Deep unfolding networks have recently emerged as a promising approach for synthetic aperture radar (SAR) imaging. However, baseline unfolding networks, typically derived from iterative reconstruction algorithms such as the alternating direction method of multipliers (ADMM), lack generalization capability across scenes, primarily because their regularizers are empirically designed rather than learned from data. In this study, we introduce a learnable regularizer into the unfolding network and propose a SAR imaging network with adaptive regularization (ARSAR-Net), which aims to generalize across heterogeneous scenes including offshore ships, islands, urban areas, and mountainous terrain. Furthermore, two variants of ARSAR-Net are developed, targeting improved imaging efficiency and reconstruction quality, respectively. Extensive validation through simulated and real-data experiments demonstrates three key advantages of ARSAR-Net: (1) a 50% increase in imaging speed over existing unfolding networks, (2) a PSNR gain of up to 2.0 dB in imaging quality, and (3) enhanced adaptability to complex scenes. These advancements establish a new paradigm for computationally efficient and generalizable SAR imaging systems.
format Preprint
id arxiv_https___arxiv_org_abs_2506_18324
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ARSAR-Net: Intelligent SAR Imaging with Adaptive Regularization
Fu, Shiping
Chen, Yufan
Zhang, Zhe
Qiu, Xiaolan
Ye, Qixiang
Signal Processing
Deep unfolding networks have recently emerged as a promising approach for synthetic aperture radar (SAR) imaging. However, baseline unfolding networks, typically derived from iterative reconstruction algorithms such as the alternating direction method of multipliers (ADMM), lack generalization capability across scenes, primarily because their regularizers are empirically designed rather than learned from data. In this study, we introduce a learnable regularizer into the unfolding network and propose a SAR imaging network with adaptive regularization (ARSAR-Net), which aims to generalize across heterogeneous scenes including offshore ships, islands, urban areas, and mountainous terrain. Furthermore, two variants of ARSAR-Net are developed, targeting improved imaging efficiency and reconstruction quality, respectively. Extensive validation through simulated and real-data experiments demonstrates three key advantages of ARSAR-Net: (1) a 50% increase in imaging speed over existing unfolding networks, (2) a PSNR gain of up to 2.0 dB in imaging quality, and (3) enhanced adaptability to complex scenes. These advancements establish a new paradigm for computationally efficient and generalizable SAR imaging systems.
title ARSAR-Net: Intelligent SAR Imaging with Adaptive Regularization
topic Signal Processing
url https://arxiv.org/abs/2506.18324