Hybrid Framework for High-Fidelity Lunar DEMs

The Máni mission (Frydenvang et al. 2025) aims to create 20cm/pixel Digital Elevation Maps (DEMs) of a few surface areas of interest using a multi-image approach. However, achieving this goal requires overcoming two fundamental challenges: The success of the mission depends on the verifiable physical accuracy of its DEMs. Raw images from push-broom scanners are contaminated with systematic noise, such as vertical striping, which creates unreliable terrain information, compromising the safety of future missions and the scientific value of the resulting DEMs (Gómez-Chova et al. 2008). Our method offers considerable expected improvement over using solely the Fernandes-Mosegaard inversion (Fernandes & Mosegaard 2022). The volume of high-resolution, multi-angular data makes purely physics-based processing computationally intensive. This poses a significant challenge to delivering data products in a timely manner (Ma et al. 2015). We mitigated this issue using a Deep-Learning Multi-View Stereo (DL-MVS) model, which reduces the load on the computationally most intensive part of the process. The expected quantitative results seem promising and worthy of further consideration, 18% better model fit, 8db reduction in artifact power, and significantly faster convergence then using solely a physics based inversion. https://www.space2025.aau.dk/ https://www.space2025.aau.dk/program