ISAR-NeRF: Neural Radiance Fields for 3-D Imaging of Space Target From Multiview ISAR Images

With the rapid development of aerospace technology, especially the rise of the commercial aerospace industry, the number of space targets in orbit has increased rapidly. As aerospace activities are unprecedentedly active, developing space situation awareness technology has become an urgent need for space security. High-resolution imaging of space targets to obtain space target information is critical in space situational awareness. Compared to traditional 2-D imaging, 3-D imaging can provide complete information about a space target, which makes it a key technology for space situational awareness. However, due to the limitations of observation setups and imaging algorithms, all current 3-D imaging methods tend to construct sparse 3-D geometries and cannot infer 2-D novel view images with geometry consistency. In this article, we will focus on far-field imaging and propose ISAR-NeRF to address these problems, which encodes the 3-D physical geometry of the target into an implicit neural network (INN). According to the physical mechanisms of inverse synthetic aperture radar (ISAR) imaging, we can effectively determine sampling points from ISAR images and feed these sampling points into an INN to output densities and backscattering coefficients. We propose a novel differentiable radar rendering algorithm that exploits densities and backscattering coefficients to render multiview consistent images and construct 3-D geometries without any explicit association algorithms. The comparative experiments qualitatively and quantitatively show that ISAR-NeRF outperforms any other methods in both 3-D reconstruction and novel view synthesis. In addition, the experiments of low signal-to-noise ratio (SNR) and noisy projection vectors demonstrate the robustness of our methods.