NeRF-enabled Analysis-Through-Synthesis for ISAR Imaging of Small Everyday Objects with Sparse and Noisy UWB Radar Data

TL;DR

This work tackles high-resolution ISAR imaging of small objects using sparse, noisy UWB radar with a portable setup. It introduces an Analysis-through-Synthesis framework built on Neural Radiance Fields, employing a differentiable radar forward model and spherical sampling to infer scene scatterers and synthesize measurements for training. Across simulated and real experiments, ATS outperforms traditional backprojection, including in non-line-of-sight and low-view scenarios, while reducing hardware complexity. The method enables practical, cost-effective ISAR imaging for robotics and mobile sensing, with future opportunities for acceleration and 3D extensions.

Abstract

Inverse Synthetic Aperture Radar (ISAR) imaging presents a formidable challenge when it comes to small everyday objects due to their limited Radar Cross-Section (RCS) and the inherent resolution constraints of radar systems. Existing ISAR reconstruction methods including backprojection (BP) often require complex setups and controlled environments, rendering them impractical for many real-world noisy scenarios. In this paper, we propose a novel Analysis-through-Synthesis (ATS) framework enabled by Neural Radiance Fields (NeRF) for high-resolution coherent ISAR imaging of small objects using sparse and noisy Ultra-Wideband (UWB) radar data with an inexpensive and portable setup. Our end-to-end framework integrates ultra-wideband radar wave propagation, reflection characteristics, and scene priors, enabling efficient 2D scene reconstruction without the need for costly anechoic chambers or complex measurement test beds. With qualitative and quantitative comparisons, we demonstrate that the proposed method outperforms traditional techniques and generates ISAR images of complex scenes with multiple targets and complex structures in Non-Line-of-Sight (NLOS) and noisy scenarios, particularly with limited number of views and sparse UWB radar scans. This work represents a significant step towards practical, cost-effective ISAR imaging of small everyday objects, with broad implications for robotics and mobile sensing applications.

Figures (9)

• Figure 1: Hardware and Measurement Setup: (a) Front view of Time Domain P440 radar module with absorber behind the antenna (b) Top view of the radar (c) JAYEGT JAYEGT electronically motorized turn-table (d) Electronically enhanced Floor Cleaning Robot Roomba_smarter
• Figure 2: illustrates our proposed analysis-through-synthesis (ATS) pipeline for radar image reconstruction. First, scene coordinates are sampled using our spherical sampling scheme and encoded with a multi-resolution hash encoding. Next, the encoded coordinates are fed into the NeRF network to predict scene scattering functions, which are then used to generate radar measurements through a differentiable forward model. Training the network involves minimizing the loss between estimated and true radar measurements for each scan of the sinogram.
• Figure 3: Illustrates our forward model geometry and sampling strategy. A transmitted ray (blue) is emitted towards the scene, propagating to a scene $\chi$ weighted by the directivity function $b_T(x)$ and transmission probability $T(o_T, x)$. All transmitted rays within the antenna beamwidth (orange lines) are then sampled at the intersection point of the sphere (green) defined by range samples. TX and RX are collocated.
• Figure 4: Simulated data in noise-free and noisy conditions. The rows show the Sinogram, BP, and ATS reconstructions respectively for single, double, triple, and quad point targets.
• Figure 5: Comparison of PSNR and LPIPS metrics for BP and ATS with different numbers of simulated reflectors at various skip angles