Light Curves for GEO object characterisation

Characterisation of space objects in Earth orbit is an essential task, particularly with the increase in space traffic and the advent of space traffic management. Proper understanding of an objects shape, size and attitude are vital in predicting its future behaviour. Light curves are increasingly being used to characterise objects, with methods ranging from simple regression analysis through to complex AI solutions. The method presented and demonstrated herein is a machine learning algorithm based on convolutional neural networks, capable of characterising object parameters such as geometry, attitude and materials of an object. The method is intended to be a flexible classification method, which could be extended to all orbits and any type of object, including debris. Herein, intermediate results of the ongoing study are presented, demonstrating the use of a multiclassification and multi-branch classification model. The results demonstrate that the method can successfully, with greater than 80% accuracy classify the shape, size, attitude and main material of an object in geosynchronous orbit from a single full night light curve.