Generation of Infrared Synthetic Data for System Training and Testing

Areté, in collaboration with the University of Arizona (UA), is developing RealSim-IR, a novel tool designed to generate high-fidelity, physics-based long-wave infrared (LWIR) imagery of maritime targets for advanced AI/ML applications. Leveraging Unreal Engine (UE5) for first-order physics-based rendering and a custom Physics Informed Neural Network (PINN) for second-order effects, RealSim-IR enables rapid, accurate, and cost-effective generation of synthetic LWIR data sets under diverse environmental conditions. This approach overcomes current limitations in data availability for training and validating automated target recognition (ATR) algorithms, particularly in challenging ocean environments where acquiring real-world data can be expensive and time-consuming.By combining UE5’s real-time 3D rendering capabilities with PINN-enabled realism enhancements, RealSim-IR accurately models critical LWIR phenomenology such as object temperature distributions, emissivity, atmospheric transmission, and dynamic surface interactions. The tool’s user interface provides intuitive control over scene content and environmental factors, outputting both imagery and metadata essential for AI/ML algorithm development. This groundbreaking simulation framework addresses the critical need for robust synthetic training data, expediting the fielding of reliable AI-driven ATR solutions in both defense and commercial arenas.