A neural network topographic correction method for the Lunar Reconnaissance Orbiter Diviner Christiansen feature map

Abstract

Topographic correction for lunar optical and thermal infrared remote sensing images is a challenging task. Although geometrical, photometrical or albedo-dependent correction methods have been used in an attempt to eliminate topographic effects in the initial lunar reflectance and emissivity images, these methods did not perform well on lunar steep slopes, large crater interiors or walls. In this paper, a novel neural network topographic correction (NNTC) model that was trained based on the relationships between flat (< 0.1° slope angle) and rugged pixels with similar Kaguya FeO abundance and optical maturity parameters (OMATs) (relative ratio < 0.1%) was applied to the Lunar Reconnaissance Orbiter (LRO) Diviner standard Christiansen feature (CF) image. The topographic correction accuracy is 0.03 μm, and no outliers are produced. Visual comparisons with the previous topographic corrected reflectance and emissivity images indicate that the topographic effects on various slope and aspect pixels were effectively corrected. The excavated anorthosite materials in the highland crater (such as the Jackson, Giordano Bruno and Tycho) interiors, surroundings, ejecta, and ray deposits have very similar NNTC CF values. The crater structures in maria or cryptomaria are identified more clearly. In addition, the NNTC method has the potential to be a universal topographic correction method for lunar optical and thermal infrared images, and it provides reliable data sources and a new method for space weathering correction.