Aurelio Kaluthantrige, University of Strathclyde; Jinglang Feng, University of Strathclyde; Jesús Gil-Fernández, ESA/ESTEC
Keywords: Image Processing, Machine Learning, Sun Phase angle, Pseudorange, HERA, Didymos
Abstract:
The Asteroid Impact and Deflection Assessment (AIDA) is a joint mission between the European Space Agency (ESA) and the National Aeronautics and Space Administration (NASA) aimed to investigate the binary asteroid system (65803) Didymos and to demonstrate asteroid impact hazard mitigation by means of a kinetic impactor. NASA developed and launched in November 2021 the Double Asteroid Redirection Test (DART) spacecraft that will impact the secondary of the binary asteroid system, Dimorphos, in September 2022. ESA contributes to AIDA with HERA, an asteroid rendezvous mission that focuses on the investigation of Didymos and to observe the outcome of the kinetic impactor test.
The design of the proximity operations of HERAs spacecraft around the target includes an autonomous optical navigation system to collect enough on-board information to estimate the relative position with respect to the asteroid. The core component of this navigation method is the Image Processing (IP) algorithm that extracts optical observables from images captured by the spacecrafts on board Asteroid Framing Camera (AFC). The Early Characterization Phase (ECP) is a proximity operation that has the objective of conducting physical and dynamical characterizations of Didymos. In this phase, the IP algorithm is designed to estimate the position of the Center of Mass (COM) of the primary to enable Line of Sight (LOS) navigation. However, the performance of standard IP algorithms is affected by illumination conditions, hence they introduce correction terms depending on the Sun phase angle (Sun-asteroid-spacecraft) to improve the accuracy of the estimation of the COM. To measure the range with the asteroid, HERA uses the Planet ALTimeter (PALT), a LIDAR experiment with an accuracy of 0.5 meters. Nevertheless, PALT is not operating during the ECP as it requires closer distances with the asteroid.
Within this context, this paper aims to measure the pseudorange with Didymos and to estimate the Sun phase angle by developing a Convolutional Neural Networks (CNN)-based IP algorithm and applying it to the images captured during the ECP of the HERA mission. For the first aim, the proposed algorithm regresses a set of keypoints on the border of the primary in the image. Knowing the intrinsic characteristics of the AFC, the algorithm measures the pseudorange with the primary by evaluating its apparent radius and comparing it with its real radius. For the second aim, the algorithm estimates the pixel position of the centroid of the primary and the pixel position of the susbsolar point of the primary. With these two keypoints, the CNN-based IP algorithm measures the phase angle.
The choice of the CNNs over standard IP algorithms is based on three main reasons. Firstly, CNNs have the main advantage to be robust over adverse illumination conditions. Secondly, standard algorithms would require to analyze each pixel of the image to measure the apparent radius of the asteroid, which is time consuming and computationally expensive. Finally, CNNs are robust over the irregular shape of the asteroid, which is an important characteristic for the estimation of the pixel position of keypoints.
The training, validation and testing datasets are generated with the software Planet and Asteroid Natural scene Generation Utility (PANGU) at different epochs of the ECP trajectory. The High-Resolution Network (HRNet) is used as CNN architecture as it represents the state-of-the-art technology in keypoint detection. The performance of the HRNet is assessed in terms of Root Mean Squared Error (RMSE) between the pixel coordinates of the estimated and the Ground Truth (GT) keypoints in the image. The performances of the HRNet-based IP algorithm are evaluated in terms of absolute error between the true range and the measured pseudorange and between the true and the estimated value of the Sun phase angle. The HRNet-based IP algorithm measures the pseudorange and estimates the position of the subsolar point with high accuracy. The Sun phase angle estimation is accurate with limited dependence on the shape of the asteroid.
Date of Conference: September 27-20, 2022
Track: Machine Learning for SSA Applications