Blending Ground View and Overhead Models

We propose to build ARGON, the ARet Ground-to-Overhead Network. The network will ingest analyst-supplied ground-level imagery ofobjects and retrieve instances of those objects in overhead collections, providing tips back to the analysts. A proprietary method of trainingthe network, leveraging in-house capabilities, data sources, and tools, will be critical to its success. During Phase I, we will prototype thealgorithm, validate its performance, and iteratively improve it by modifying network architecture and training data as necessary. Phase II willcontinue this iterative improvement, incorporating relevant classified data if available. Once acceptable probability of detection and falsealarm rates are achieved, the prototype will be transitioned into operational software, leveraging Arets experience deploying IC products.