Neural networks for the selection of payload data on an interplanetary nanosatellite mission

In recent years, artificial intelligence algorithms have began to spread in various sectors of the aerospace industry, covering applications such as failure detection, autonomous operations and design optimization. Different methodologies have been defined to date, and among these it is worth mentioning fuzzy logics and neural networks. One of the most interesting characteristics of these algorithms is that, if properly designed, they can be used to emulate the expert knowledge and therefore they can be applied in parallel, or even substitute, human workers in certain fields or during specific phases of a space mission. Micro- and nano-satellites are one of the most appealing platforms on which to deploy these algorithms, for several reasons: they usually have faster, more flexible development cycles, and are less constrained with respect to larger and expensive traditional spacecraft. In this sense, they can be identified as the ideal test bed to apply new technologies. On the other hand, nano-satellite missions (and especially interplanetary missions) are characterized by a very stringent limiting factor: the low data rates available. Therefore, any algorithm that aims at increasing the downlinked data would be beneficial to the missions. This paper presents the use of artificial neural networks to perform payload data selection for an interplanetary mission to a Near Earth Asteroid carried out by a nanosatellite system. An algorithm to evaluate and select images acquired during the science operations is designed and analysed. The algorithm is trained on images compatible with those that can be taken nearby the asteroid, and allows to determine autonomously on-board whether the images are to be discarded, or are deemed interesting enough to be downlinked. The algorithm brings many advantages, impacting different segments in the mission architecture. As far as the space segment is concerned, payload data downlink will benefit thanks to the fact that only the most promising images will be sent to Earth, reducing the quantity of downlinked data. In the ground segment, a reduced and improved data flow will allow for more agile resource allocations.