Using artificial neural networks to improve photometric modeling in airless bodies

TL;DR

This paper presents a neural network-based photometric modeling method for airless bodies, improving accuracy and efficiency in interpreting surface properties from spacecraft images, validated on Bennu data.

Contribution

It introduces a supervised neural network approach that outperforms previous methods in modeling brightness dependence, reducing computation time and increasing precision.

Findings

Achieved up to 14.30% improvement in modeling accuracy

Reduced computational time significantly

Validated on Bennu OSIRIS-REx images

Abstract

Relevant information about physical properties of the surface of airless bodies such as porosity, particle size, or roughness can be inferred knowing the dependence of the brightness with illumination and observing geometry. Additionally, this knowledge is necessary to standardize or photometrically correct data acquired under different illumination conditions. In this work we develop a robust, automatic, and efficient photometric modeling methodology which is tested and validated using Bennu images acquired by the camera MapCam from the OSIRIS-REx spacecraft. It consists of a supervised machine learning algorithm through an artificial neural network. Our system provides a more precise modeling for all color filters than the previous procedures which are already published, offering an improvement over this classic approach of up to 14.30%, as well as a considerable reduction in computing time.