Seed Innovations and subcontractor BIT Systems, a division of CACI International, apply our experience in machine learning, data analytics andimage processing to accomplish the research for the SBIR topic: Suppression of false alarms in Automated Target Recognizers (ATR) that useMachine Learning. With the amount of available imagery data increasing and adversaries vehicles and tactics becoming more sophisticated,falsely identifying a target becomes costly in terms of the warfighters productivity. Through the research on this project, Seed Innovationsdesigns and prototypes a system, Boost, to drive down false alarm rates while not suppressing real alarms. Boost demonstrates the feasibilityof enhancing the output of existing ATR systems by leveraging a targets contextual data; e.g. observed locations relative to a specified area.Boost transforms the contextual data into heatmaps and trains a deep neural network on these heatmaps to determine the probability thatthe output of the ATR is a false alarm. It is important to note that Boost does not attempt to develop image recognition neural networks tooutperform those in current ATR systems, but instead uses the output of the ATR and contextual heatmaps to lower false alarm rates.
