Abstract:Multi-baseline SAR 3D imaging faces significant challenges due to data sparsity. In recent years, deep learning techniques have achieved notable success in enhancing the quality of sparse SAR 3D imaging. However, previous work typically rely on full-aperture high-resolution radar images to supervise the training of deep neural networks (DNNs), utilizing only single-modal information from radar data. Consequently, imaging performance is limited, and acquiring full-aperture data for multi-baseline SAR is costly and sometimes impractical in real-world applications. In this paper, we propose a Cross-Modal Reconstruction Network (CMR-Net), which integrates differentiable render and cross-modal supervision with optical images to reconstruct highly sparse multi-baseline SAR 3D images of vehicle targets into visually structured and high-resolution images. We meticulously designed the network architecture and training strategies to enhance network generalization capability. Remarkably, CMR-Net, trained solely on simulated data, demonstrates high-resolution reconstruction capabilities on both publicly available simulation datasets and real measured datasets, outperforming traditional sparse reconstruction algorithms based on compressed sensing and other learning-based methods. Additionally, using optical images as supervision provides a cost-effective way to build training datasets, reducing the difficulty of method dissemination. Our work showcases the broad prospects of deep learning in multi-baseline SAR 3D imaging and offers a novel path for researching radar imaging based on cross-modal learning theory.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper primarily aims to address the issue of data sparsity in three-dimensional imaging using multi-baseline Synthetic Aperture Radar (SAR). Specifically, the research team proposes a Cross-Modal Reconstruction Network (CMR-Net) that leverages differentiable rendering techniques and optical image supervision to enhance 3D imaging results obtained from very sparse multi-baseline SAR data. #### Main Issues: 1. **Image Enhancement Limitation**: Existing deep learning-based methods train neural networks using low-resolution to high-resolution SAR image pairs, but this approach is limited by the electromagnetic imaging mechanism, making it difficult to further improve resolution. 2. **Data Quality Constraint**: Current algorithms require full-aperture data to generate high-resolution SAR images, which is often inefficient and expensive in practical applications. 3. **Observation Sensitivity and Noise Interference**: SAR imaging results are highly sensitive to observation geometry and noise interference. The feature information of SAR images input into neural networks is unstable, leading to poor generalization ability of deep learning methods. #### Solutions: 1. **Cross-Modal Supervision**: Utilize 2D optical images to supervise the 3D reconstruction process, guiding the network to generate high-resolution 3D images with coherent structures and prominent features. 2. **Cost-Effectiveness**: Using optical image data as a supervisory means can achieve more cost-effective high-resolution SAR images compared to processing full-aperture electromagnetic data. 3. **Data Augmentation and Projection-Reprojection Module**: A unique data augmentation scheme is designed, and a Projection-Reprojection Module (PRP) is integrated into the network to enhance its robustness and generalization ability. 4. **Simulated Data Training**: The network is trained only on simulated data and validated on real measurement data without any fine-tuning. Through the above methods, the paper demonstrates the superiority of the proposed CMR-Net in 3D reconstruction performance under low signal-to-noise ratio and extremely sparse measurement conditions.

Sparse Multi-baseline SAR Cross-modal 3D Reconstruction of Vehicle Targets

Single Target SAR 3D Reconstruction Based on Deep Learning

Learning-Based Sparse Recovery Algorithm for 3D SAR Imaging

SAR Non-Sparse Scene Reconstruction Network Via Image Feature Representation Learning

DeepRED Based Sparse SAR Imaging

3-D SAR Autofocusing with Learned Sparsity

SAR Nonsparse Scene Reconstruction Network via Image Feature Representation Learning

LRSR-ADMM-Net: A Joint Low-Rank and Sparse Recovery Network for SAR Imaging

3-D SAR Data-Driven Imaging via Learned Low-Rank and Sparse Priors

Fast Super-resolution 3D SAR Imaging Using an Unfolded Deep Network

SAR-NeRF: Neural Radiance Fields for Synthetic Aperture Radar Multi-View Representation

Multistatic Synthetic Aperture Radar Baseline Design for 3-D Imaging

Nonsparse SAR Scene Imaging Network Based on Sparse Representation and Approximate Observations

SAR Image Reconstruction of Non-Sparse Scene via Deep NSR-Net.

Lightweight Pixel2mesh for 3D Target Reconstruction from a Single SAR Image

An Iterative Multidimensional Feature Reconstruction Network for Tomographic SAR Imaging

ISAR-NeRF: Neural Radiance Fields for 3D Imaging of Space Target from Multi-view ISAR Images

3D SAR Imaging Method Based on Learned Sparse Prior

STLS-LADMM-Net: A Deep Network for SAR Autofocus Imaging

End-to-End SAR Deep Learning Imaging Method Based on Sparse Optimization

SAR Parametric Super-Resolution Image Reconstruction Methods Based on ADMM and Deep Neural Network