Methods of Machine Learning for Space Object Pattern Classification

Space superiority requires space protection and space situational awareness (SSA), which rely on rapid and accurate space object behavioral and operational intent discovery. The analytics required for space object detection, tracking, and pattern classification include machine learning. The development of deep learning (DL) methods shows promise in many areas. In this paper, a convolutional neural network (CNN) classifies the behaviors of space objects for evasive satellite behaviors detection. Additionally, within the Adaptive Markov Inference Game Optimization (AMIGO) engine, a game theoretic approach describes the situation versus a control problem. Using data-level fusion, stochastic modeling/propagation, and DL pattern classification. Numerical simulations demonstrate the advantage of DL methods for space object pattern classification improving from 34% to 98% with training.