2020/04/21 by Youngjin Park, Park, Young-Jin, Han‐Lim Choi +1 · 1 citation
Computer Science · Engineering · Physics and Astronomy · #Applications (stat.AP) #FOS: Computer and information sciences #FOS: Electrical engineering #Gaussian Processes and Bayesian Inference #Image and Video Processing (eess.IV) #Planetary Science and Exploration #Robotics and Sensor-Based Localization #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2004.09791
openalex publication_date 2020/04/21 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
With the advent of NASA's lunar reconnaissance orbiter (LRO), a large amount\nof high-resolution digital elevation maps (DEMs) have been constructed by using\nnarrow-angle cameras (NACs) to characterize the Moon's surface. However, NAC\nDEMs commonly contain no-data gaps (voids), which makes the map less reliable.\nTo resolve the issue, this paper provides a deep-learning-based framework for\nthe probabilistic reconstruction of no-data gaps in NAC DEMs. The framework is\nbuilt upon a state of the art stochastic process model, attentive neural\nprocesses (ANP), and predicts the conditional distribution of elevation on the\ntarget coordinates (latitude and longitude) conditioned on the observed\nelevation data in nearby regions. Furthermore, this paper proposes sparse\nattentive neural processes (SANPs) that not only reduces the linear\ncomputational complexity of the ANP O(N) to the constant complexity O(K) but\nenhance the reconstruction performance by preventing overfitting and\nover-smoothing problems. The proposed method is evaluated on the Apollo 17\nlanding site (20.0\degN and 30.4\degE), demonstrating that the suggested\napproach successfully reconstructs no-data gaps with uncertainty analysis while\npreserving the high resolution of original NAC DEMs.\n