ISAR-iNeRF: Inverting Neural Radiance Fields from ISAR Images for Attitude Estimation

The attitude of space targets is essential information for space security. Estimating the attitude of targets from inverse synthetic aperture radar (ISAR) images has attracted widespread attention. Most existing literature has proved the effectiveness of estimating attitude according to ISAR image features extraction. However, they either require complete information about the target or are limited by feature extraction methods, resulting in relatively low speed and accuracy in attitude estimation. In this paper, we propose a novel attitude estimation method, ISAR-iNeRF, which can quickly and accurately estimate the attitude of a space target. We model neural radiation fields from multi-view ISAR images, which implicitly build up the feature database of targets. A differentiable estimation pipeline is introduced to iteratively update the target's attitude via gradient-based optimization. The experiments demonstrate that the proposed method can achieve fast and accurate attitude estimation.