Characterization and Typing of Hard-to-Acquire Targets using Advanced Machine Learning Methods on WFOV Staring Data

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy; Space Technology The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and import of defense-related material and services, including export of sensitive technical data, or the Export Administration Regulation (EAR), 15 CFR Parts 730-774, which controls dual use items. Offerors must disclose any proposed use of foreign nationals (FNs), their country(ies) of origin, the type of visa or work permit possessed, and the statement of work (SOW) tasks intended for accomplishment by the FN(s) in accordance with the Announcement. Offerors are advised foreign nationals proposed to perform on this topic may be restricted due to the technical data under US Export Control Laws. OBJECTIVE: Develop a prototype software algorithm to perform detection, identification, characterization and typing of hard-to-acquire targets (air-/space-based) using advanced machine learning methods on uncued surveillance data from ground-based, wide field of view (WFOV), electro-optical (EO), staring sensors. DESCRIPTION: USSPACECOM surveillance architectures have difficulty maintaining track and custody of today's Starlink constellation, despite SpaceX's cooperative diligent self-reporting. The U.S. Government will likely be challenged as foreign entities, like China, begin to deploy their own, potentially uncooperative, mega constellations. Recent public reports from China include a new 12,000+ satellite communications constellation. Imagining that a sufficient global space surveillance capability is available, coalescing and sifting through continuous streams of big data will be nearly impossible using current concepts of operations (CONOPS) and will stress the existing 18th and 19th Space Defense Squadrons' services. The key enabler is being able to efficiently use this vast amount of data in multiple different ways to extract useful information and insights. Employing artificial intelligence/machine learning (AI/ML) algorithms to detect, ID, characterize and type hard-to-acquire targets in large amounts of data from ground-based, WFOV, EO, staring systems is highly desirable and will lead to better security and broader surveillance overall. PHASE I: Applicant must have developed a concept for a workable prototype or design to address at a minimum the basic capabilities of the stated objective. The documentation provided must substantiate that the offeror has developed a preliminary understanding of the technology to be applied in their Phase II proposal to meet the objectives of this topic. Documentation should include all relevant information including, but not limited to, technical reports, test/real data, prototype designs/models, and performance goals/results. GFE will not be provided. Applicant is expected to procure its own WFOV, EO, staring data on which to test and evaluate their algorithms. Signed Letters of Support from customers and/or end-users are encouraged. PHASE II: Expand upon the initial software algorithms to meet the needs of the DAF customer leveraging previous Ph I-type feasibility study results and data. Develop a pristine labeled dataset of WFOV sky images with aircraft, satellites, and other visible objects labeled in metadata for the purposes of testing characterization and typing algorithms, training new AI/ML-based algorithms, and fine-tuning existing algorithms. Evaluate algorithms with common classification metrics including, but not limited to, precision-recall (PR) curves for each algorithm with area under curve (AUC), recall at max or desired precision, precision at max or desired recall, and max f1 score (harmonic mean of precision and recall); receiver operating characteristics (ROC) curves for each algorithm and AUC metrics; and additional metrics including performance speed and throughput, memory requirements, etc. Demonstrate how results can be ingested and displayed by existing, operational tools to reach real-world users without costly refactoring or licensing for a new tool. GFE will not be provided. PHASE III DUAL USE APPLICATIONS: Mature prototype software into a commercial product for commercial Space Domain Awareness; identify government and commercial organizations for transition; and generate the technical and training documentation required for third party integration. The commercial product should provide a dual-use capability to enable real-time Space Domain Awareness (SDA) and situational understanding to aid the U.S. warfighter with actionable tactical intelligence as well as condition monitoring for the commercial sector and, in general, to ensure spaceflight safety. Provide services to the government to maximize the utility of the algorithm's results to operations; update software prototype for different applications as necessary; offer pipeline solutions for model training, compression, runtime inference optimization, and deployment to edge devices of accurate, secure, and duplicatable AI/ML models. REFERENCES: 1. United States Space Force, White Paper on Competitive Endurance: A Proposed Theory of Success for the U.S. Space Force , Office of the Chief of Space Operations, Strategic Initiatives Group, 11 January 2024. https://www.spaceforce.mil/Portals/2/Documents/White_Paper_Summary_of_Competitive_Endurance.pdf.; 2. Bartusiak, Emily R. Machine Learning for Speech Forensics and Hypersonic Vehicle Applications , Purdue University. Thesis, 2020. https://doi.org/10.25394/PGS.21678095.v1.; 3. G. Martin, J. Wetterer, J. Lau, J. Case, N. Toner, C. Chow, and P. Dao. Cislunar Periodic Orbit Family Classification from Astrometric and Photometric Observations Using Machine Learning , 21st Advanced Maui Optical and Space Surveillance Technologies (AMOS) Conference, 2020.; 4. Dinsley, Ralph, and Newman, Christopher. LEO Space Surveillance and Tracking Through a Non-Traditional Lens , Proc. 2nd NEO and Debris Detection Conference, Darmstadt, Germany, 24-26 January 2023.; KEYWORDS: Operational surprise; uncued surveillance; target characterization and typing; uncorrelated track; real-time data processing; wide field-of-view; information exploitation; space domain awareness; artificial intelligence/machine learning; hypersonics