LNEM：月球神经高程模型
LNEM: Lunar Neural Elevation Model
把推扫式（pushbroom）成像过程显式嵌入体渲染，并联合学习阴影模型，为卫星/行星遥感影像提供几何一致的三维重建方案；做 NeRF 类重建时若输入是线阵推扫影像，可直接借鉴其相机建模与光照感知渲染思路。
原文依据 a volumetric reconstruction method that explicitly incorporates the pushbroom imaging process
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论月球表面高分辨率、高精度的数字高程模型（DEM）对着陆点选址与地质研究至关重要。然而，传统立体匹配对三维场景的表示能力有限，且难以处理无纹理区域与极端光照变化。近期的月球神经渲染方法同样不适用于三维重建，因为它们依赖简单的针孔近似来建模推扫式传感器。卫星影像处理流程中的几何失配、分布偏差以及繁重的人工预处理进一步加剧了这些困难。为解决上述问题，我们提出月球神经高程模型（LNEM），一种显式纳入推扫式成像过程的体素重建方法。本方法的核心组件之一是 Lunar Studio，这是一个多轨道数据集与处理流程，基于严格传感器模型（RSM）构建，可从月球勘测轨道飞行器相机（LROC）窄角相机（NAC）与韩国探路者月球轨道器（KPLO）月球地形成像仪（LUTI）生成几何一致的观测。LNEM 将该推扫式相机公式与学习到的阴影建模相结合，从而在具有挑战性的月球光照条件下实现几何有据且光照感知的体渲染。大量实验表明，LNEM 在多样化的视角与光照条件下实现了跨多个传感器的几何一致重建，为传统 DEM 流程提供了可扩展的补充。为支持可复现性与未来月球研究，我们公开 Lunar Studio 多轨道数据集与 LNEM 重建流程。
英文原文摘要 1723 字符
High-resolution and high-precision digital elevation models (DEMs) of the lunar surface are essential for landing site selection and geological research. However, traditional stereo matching provides a limited representation of the 3D scene and struggling with non-textured regions and extreme illumination variations. Recent lunar neural rendering methods are also ill-suited for 3D reconstruction due to their reliance on simple pinhole approximations for pushbroom sensors. These challenges are further compounded by geometric misalignment, distributional bias, and labor-intensive handcrafted preprocessing in satellite image pipelines. To address these issues, we introduce the Lunar Neural Elevation Model (LNEM), a volumetric reconstruction method that explicitly incorporates the pushbroom imaging process. A core component of our approach is Lunar Studio, a multi-orbit dataset and pipeline constructed using Rigorous Sensor Models (RSMs) to produce geometrically consistent observations from the Lunar Reconnaissance Orbiter Camera (LROC) Narrow Angle Camera (NAC) and the Korea Pathfinder Lunar Orbiter (KPLO) Lunar Terrain Imager (LUTI). LNEM integrates this pushbroom camera formulation with learned shadow modeling, enabling geometrically grounded and illumination-aware volumetric rendering under challenging lunar lighting conditions. Extensive experiments demonstrate that LNEM achieves geometrically consistent reconstruction across multiple sensors under diverse viewing and illumination conditions, providing a scalable complement to conventional DEM pipelines. To support reproducibility and future lunar research, we release Lunar Studio, the multi-orbit dataset, and the LNEM reconstruction pipeline.