SAR-GS: Gaussian Splatting based SAR Images Rendering and Target Reconstruction

TL;DR

This work introduces SAR-GS, a differentiable Gaussian-splat framework for SAR target reconstruction that unifies forward SAR rendering with inverse optimization. By integrating Mapping and Projection Algorithm (MPA) with Gaussian splatting and a CUDA-based backward gradient flow, the method learns both geometric structure and scattering properties of targets from multi-view SAR imagery. Extensive experiments on rendered and real MSTAR data demonstrate improved image fidelity and accurate 3D reconstructions, with faster training than SAR-NeRF and more explicit target representations than some traditional DSR approaches. The results highlight the potential of Gaussian-point-cloud representations for SAR 3D reconstruction while underscoring challenges related to data quality, noise, and inter-point coherence, guiding future work on correlated scene modeling and robust noise handling.

Abstract

Three-dimensional target reconstruction from synthetic aperture radar (SAR) imagery is crucial for interpreting complex scattering information in SAR data. However, the intricate electromagnetic scattering mechanisms inherent to SAR imaging pose significant reconstruction challenges. Inspired by the remarkable success of 3D Gaussian Splatting (3D-GS) in optical domain reconstruction, this paper presents a novel SAR Differentiable Gaussian Splatting Rasterizer (SDGR) specifically designed for SAR target reconstruction. Our approach combines Gaussian splatting with the Mapping and Projection Algorithm to compute scattering intensities of Gaussian primitives and generate simulated SAR images through SDGR. Subsequently, the loss function between the rendered image and the ground truth image is computed to optimize the Gaussian primitive parameters representing the scene, while a custom CUDA gradient flow is employed to replace automatic differentiation for accelerated gradient computation. Through experiments involving the rendering of simplified architectural targets and SAR images of multiple vehicle targets, we validate the imaging rationality of SDGR on simulated SAR imagery. Furthermore, the effectiveness of our method for target reconstruction is demonstrated on both simulated and real-world datasets containing multiple vehicle targets, with quantitative evaluations conducted to assess its reconstruction performance. Experimental results indicate that our approach can effectively reconstruct the geometric structures and scattering properties of targets, thereby providing a novel solution for 3D reconstruction in the field of SAR imaging.

Figures (16)

• Figure 1: The general workflow of a target reconstruction algorithm based on SAR Gaussian Splatting.
• Figure 2: Mapping and Projection Algorithm.The airborne radar travels along the $O_\text{r}X_\text{r}$ direction (azimuth direction) and emits radar beams along the $O_\text{r}Z_\text{r}$ direction (range direction).
• Figure 3: Mapping Projection Algorithm (Single Azimuth Direction)
• Figure 4: Schematic Diagram of the Transformation between Radar Coordinate System and World Coordinate System, where $O_\text{w}-X_\text{w}Y_\text{w}Z_\text{w}$ represents the World Coordinate System and $O_\text{r}-X_\text{r}Y_\text{r}Z_\text{r}$ denotes the Radar Coordinate System. $\phi$ is the radar azimuth angle, and $\theta$ is the radar elevation angle.
• Figure 5: Gradient Flow Diagram