Deep Learning-Based Modeling of High-Resolution Lunar Topography Using Orbiter Imagery

Abstract

Fine topographic modeling of the lunar surface is vital for exploration missions and scientific applications. While there has been rapid development in Deep Learning (DL) methods that leverage single-view orbiter images to generate high-resolution Digital Terrain Models (DTMs), these approaches encounter challenges in accurately retrieving fine-scale terrains. Here, we propose a transformer-based DL method for the reconstruction of multi-scale terrain features. Moreover, the improved approach takes into account lunar elevation characteristics, with the goal of improving prediction accuracy, particularly for subtle features. Input images for evaluating the performance of the proposed method are obtained from the Narrow Angle Camera (NAC) onboard Lunar Reconnaissance Orbiter (LRO). We demonstrate that our method achieves enhanced accuracy and effectively captures fine-scale terrain features, such as small-scale craters and ejecta rays, exhibiting performance comparable to the Shape-From-Shading (SFS) method. These results suggest that our method is well-suited for many practical applications.