Satellite model determination from 2D images

Space situational awareness requires the ability to extract satellite shape information and infer intent from space-based or ground-based imagery. For resident space objects in MEO or GEO orbit, or SmallSats in low-LEO orbit, imagery with only a few resolution elements may be available; in these cases, the limited and uncertain information content hinders decision making.Spectral Sciences Inc. (SSI) proposes to develop a new satellite identification tool, Cognitive Image Recovery Code (CIRC), to evaluate low resolution imagery, including views from multiple locations and temporal light curves generated from unresolved imagery, through comparisons to potential 3D satellite models.The result will be a predicted 3D model for the satellite and an associated uncertainty. SSIs first-principles QUID simulation code will be used to build training imagery for a deep learning algorithm that matches the data to potential satellite models.QUIDs exceptional combination of fidelity and speed will insure that the simulated imagery is both comprehensive and radiometrically accurate, and make it possible to generate images on-the-fly as the retrievals are refined.Phase I will prototype the basic approach, including generating model assignments and uncertainties, and development of a system-level architecture for the Phase II software.