.png){kind=icon}
.png){kind=icon}
Return
Unsupervised VideoSAR Sparse Imaging Using Physically Interpretable Video Tensor Decomposition-Based Deep Prior
.png){kind=icon}
PRE
AI
DOI：10.1109/TGRS.2026.3670288.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Video synthetic aperture radar (VideoSAR) is an advanced microwave imaging technology that can be applied to the motion platforms such as autonomous aerial vehicles. It overcomes the limitations of traditional SAR imaging, enabling dynamic high-frame-rate observation. VideoSAR has significant advantages in applications such as traffic monitoring, disaster prediction, and environmental surveillance. However, the huge storage and computational overhead caused by high-frame-rate imaging pose challenges to the application of VideoSAR, particularly in small platforms like drones. Conventional VideoSAR imaging methods require Nyquist-rate sampling and consume a lot of resources. VideoSAR sparse imaging techniques, such as tensor recovery and deep learning, have been proposed to address this issue. However, existing sparse imaging methods exhibit limitations in computational efficiency, adaptability, and data dependency. To address these challenges, this article proposes an unsupervised VideoSAR sparse imaging method based on physically interpretable deep prior. First, a physically interpretable deep prior for VideoSAR data is presented, which can leverage the redundancy of VideoSAR data for sparse imaging. Second, an unfolded VideoSAR sparse imaging network is constructed. Finally, an unsupervised learning framework is developed to eliminate reliance on labeled data. By combining unsupervised learning with deep prior modeling, the proposed method can achieve efficient high-frame-rate imaging under sparse sampling conditions. Besides, the proposed method significantly enhances the robustness and adaptability of VideoSAR sparse imaging. Experimental results demonstrate superior performance compared with existing tensor-recovery-based and supervised deep-learning-based methods.
Keywords:
Sparse imaging
tensor sparse recovery
unsupervised imaging network
video deep decomposition
video synthetic aperture radar (VideoSAR)
Journal
IF:
8.6
Papers:
2.1W
Citations:
10.7W