Modular and Scalable Machine Learning Framework for Infrared Search and Track

Infrared Search and Track (IRST) systems play a critical role in detecting and monitoring airborne targets by passively sensing their thermal emissions in the infrared spectrum. These systems offer critical advantages over radar-based counterparts, including a low probability of intercept by enemy sensors and the ability to track targets based on the heat generated by engines or other aircraft components. However, despite their strategic value, IRST systems face enduring challenges in performing robustly across diverse conditions, limited availability of high-quality data, and the significant computational complexity of real-time processing, all within the constraints of Size, Weight, and Power (SWaP) requirements for airborne platforms. To overcome these challenges, we propose to develop a data-driven framework that facilitates practical implementation and robust operation, in the face of limited data, while running quickly on low-SWaP hardware. By integrating advanced artificial intelligence and machine learning (AI/ML) algorithms into key stages of the IRST pipeline, our framework will address computational inefficiencies, enhance robustness under varying environmental conditions, and leverage novel data curation processes to generate high-quality datasets for training and refinement. The salient aspects of the proposed framework are: (1) an IRST Data Generation module, (2) a Target Detection module, (3) a Target Tracking module, and (4) a complete ML Operations pipeline. In Phase I, core detection, tracking, and data generation components were developed and demonstrated. In Phase II, focus will be placed on enhancing the fidelity of data generation; investigating new ML architectures and training processes for detection and tracking; and streamlining model training iterations as datasets are curated. Technology validation will be carried out on relevant, low-SWaP hardware, and essential steps will be taken toward integrating the framework into Air Force workflows.