A neural process approach for probabilistic reconstruction of no-data gaps in lunar digital elevation maps

With the advent of NASA's lunar reconnaissance orbiter (LRO), a large amount high-resolution digital elevation maps (DEMs) have been constructed by using narrow-angle cameras (NACs) to characterize Moon's surface. However, NAC DEMs commonly contain no-data gaps (voids), which makes map less reliable. To resolve issue, this paper provides deep-learning-based framework for probabilistic reconstruction in DEMs. The is built upon state-of-the-art stochastic process model, attentive neural processes (ANP), and predicts conditional distribution on target coordinates (latitude longitude) conditioned observed data nearby regions. Furthermore, proposes sparse (SANPs) that not only reduce linear computational complexity ANP O(N) constant O(K) but enhance performance preventing over-fitting over-smoothing problems. proposed method evaluated three different with distinct geographical features, including anthropogenic site, graben, craterlet, demonstrating suggested approach successfully reconstructs an uncertainty analysis while preserving original