Synthetic scene generation for AI/ML training that accounts for system aberrations and environmental effects

Training a machine learning or artificial intelligence system requires large data sets, of an array of mission conditions and scenarios. Accumulating these data sets for all scenarios from actual missions is difficult and creating synthetic data sets that are idealized and not close enough to the real world conditions is dangerous as the ML/AI system may fail under certain circumstances. Creating synthetic data sets which account for real world physics and can accurately create a scene that a sensor in the field would detect is crucial for training a system accurately. This work seeks to illustrate how the complexities of the environment, the optical system, the sensor and the mission scenario all contribute to the creation of a realistic scene that a system can expect to see when deployed.