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

Synthetic aperture radar (SAR) images are highly sensitive to observation configurations and exhibit significant variations across different viewing angles, making it challenging to represent and learn their anisotropic features. As a result, deep learning methods often generalize poorly across different view angles. Inspired by the concept of neural radiance field (NeRF), this study combines SAR imaging mechanisms with neural networks to propose a novel NeRF model for SAR image generation. Following the mapping and projection principles, a set of SAR images are modeled implicitly as a function of attenuation coefficients and scattering intensities in the 3-D imaging space through a differentiable rendering equation. SAR-NeRF is then constructed to learn the distribution of attenuation coefficients and scattering intensities of voxels, where the vectorized form of the 3-D voxel SAR rendering equation and the sampling relationship between the 3-D space voxels and the 2-D view ray grids are analytically derived. Through quantitative experiments on various datasets, we thoroughly assess the multiview representation and generalization capabilities of SAR-NeRF. In addition, this article includes few-shot classification performance improvement as a metric for generation performance. The study found that using 12 images per class resulted in an accuracy improvement of nearly 10% for the classification algorithm.