Generative Augmentation Process (GAP)

To support the DoDÆs need to produce synthetic multi-spectral sensor data for the purpose of training low-shot machine learning models, ARiA has successfully demonstrated the feasibility of the Generative Augmentation Process (GAP). A novel, conditionally generative artificial intelligence (AI) method was used to synthesize infrared (IR) from visible data over multiple target classes using measured data coming from multiple satellites with images sampled globally over multiple seasons and at different look angles. This model was trained in an unsupervised manner, making it amenable to both large and small amounts of unlabeled, unstructured data. Compared to using just the visible spectrum, including this additional synthesized IR data was proven to improve performance of automatic target recognition (ATR) models over all targets of interest. To continue the development of GAP, the Phase II effort will: (1) expand the generative models to different image modalities and wavelengths, (2) include additional ATR and adversarial robustness studies, (3) create a prototype advanced programming interface (API) of the GAP methods for future deployment on edge and enterprise systems.á