We apply deep learning AI technology to the automated detection and identification of high value objects in overhead Synthetic Aperture Radar (SAR) imagery.  Unlike many other businesses operating in this area, EMSI has an extensive understanding of radar phenomenology.  We use our understanding of radar phenomenology to produce simulated training imagery for our AI systems.  This is significant because, unlike the non-governmental commercial and academic world, the US government has limited amounts of labeled data.  Accordingly, our ability to produce simulated AI training data allows us to excel and makes EMSI unique among the small number of companies operating in this radar AI technology area.

Deep Learning ATRs are context aware and hence need to be trained using data representative of the operational environment in order to be effective.  A significant portion of the ATR development funded by the US government uses pristine scripted US test range data for training and testing.  This data does not match the statistical characteristics of operational foreign imagery and, as a result, operational performance will suffer.

EMSI’s ATR development uses a combination of synthetic target imagery superimposed within operational SAR imagery and labeled collected imagery for training.  Our ATRs are tested and deployed using operational imagery of the relevant foreign sites.  Accordingly, we are able to achieve superior performance on operational imagery.