Characterization and Classification of Low-Resolution Satellites with Electro-Optical Fiducial Markers

In this paper, we demonstrate the efficacy of identifying low-Earth-orbit and geosynchronous-Earth-orbit satellites with an electro-optical fiducial marker in a complex mission environment. Our focus is on space-based observations, where the observing satellite sensors produce low-resolution or unresolved images. To demonstrate this novel approach, we will use the digital mission engineering software tool Ansys Systems Tool Kit and its Electro-Optical Infrared capability to model missions, vary sensor properties, modify electro-optical fiducial markers on the satellite of interest, and generate synthetic sensor imagery data to train and evaluate support vector machine and convolutional neural network classifiers. We will also investigate how feature selection and machine-learning model performance is impacted using low-resolution/unresolved images and how well our models can distinguish the satellite with the electro-optical fiducial markers from the bloblike shapes. The approach discussed here will provide a generalized framework for configuring systems and for object identification and characterization. The primary application in this work is for space situational awareness and space domain awareness; however, the workflows can also be applied to object identification in other domains.