Automated synthetic infrared image generation for AI applications

AI-powered systems have demonstrated state-of-the-art target recognition performance, however, they require large training sets of images to ensure decent accuracy.
In the case of electro-optical infrared (EO/IR) remote sensing, acquiring sufficient measured imagery can be difficult, in particular for confidential and high-value targets.  Highly accurate EO/IR scene simulation is a great alternative to the testing which is expensive and, in most cases, impossible & confidential. In this presentation, we are proposing an automated process to generate such highly accurate IR images through our physics-based simulation software, MuSES, and Process flow Software, Cotherm. The MuSES inputs include environmental factors, global location, date and time, vehicle engine state, target perspective, sensor resolution, waveband, etc. The output of this process is a MuSES-generated EO/IR sensor radiance image dataset with the necessary variety to be suitable for algorithm development and training. This process enables users to build imagery of any target under any condition.