Autonomous Sensing of Defense Tactical Targets by LEO Imaging Systems

Toyon is performing advanced algorithm development of an autonomous analytics architecture for generating moving target information from image data collected from LEO satellites leveraging machine-learning-based methodologies. The effort includes data simulation and curation for algorithm development and model training, algorithm performance analysis, software development of validated algorithms, an on-orbit hardware implementation capabilities assessment, and a development plan for transitioning the technology to ground-based processing centers and potential future proliferated LEO systems. The algorithms leverage neural network models for: calculating dense optical flow parameters for registering imagery prior to clutter suppression, image segmentation for point-source target detection, and fully-resolved target detection. Two new clutter suppression methods for mitigating the effects of parallax commonly seen in LEO data are also being matured. The validated algorithms will be implemented in software that can be onboarded to FORGE MDPAF to expedite the transition of this technology to government ground-processing centers.