Machine Learning Algorithms for Infrared Search and Track Applications

Toyon Research Corp. proposes development, validation, and delivery of an efficient and highly effective AI/ML-based IRST detection and tracking prototype. The resulting automated processing technology will enable early warning of long-range airborne threats using commercial off-the-shelf (COTS) low cost, size, weight, and power (C-SWaP) sensor and processor technology. Using efficient AI/ML architectures will provide significant improvements in computational efficiency while also improving on the accuracy of dim target detection and tracking provided by classical IRST signal processing algorithms. In Phase I, we demonstrated the feasibility and evaluated the performance of AI/ML-based IRST in comparison with representative baseline IRST algorithms. A mix of real and synthetic IRST data was used to train AI/ML models, and evaluations were performed using sequestered real data provided by the government from a low cost uncooled long-wave infrared sensor. In Phase II, the AI/ML architectures and end-to-end software will be optimized. Large-scale datasets will be curated from a variety of real and semi-real/semi-synthetic sources. Updated AI/ML models will be trained and validated. Measures of performance will be defined, evaluated, and reported to enable prediction of operational utility in a variety of real-world conditions. The end-to-end software solution will be demonstrated using COTS hardware and delivered to the government along with documentation.