Deep Learning Stationary Target Detector and Classifier

To address the Army’s need for an algorithm for enhanced detection and classification of stationary ground targets for fire control radar of Apache attack helicopters, Physical Optics Corporation (POC) proposes to mature, in Phase II, software for the new Deep Learning Stationary Target Detector and Classifier (DEESTAC) technology based on innovations in deep learning neural networks for enhancing radar resolution and classifying objects. During Phase I, POC successfully trained and tested neural networks with multiple radar datasets. We showed improved performance by using polarization data, evaluated effects of azimuth on classification accuracy, and demonstrated DEESTAC’s potential to detect and classify stationary ground targets as a function of signal-to-noise ratio. During Phase II, POC plans to further mature this technology by bringing the DEESTAC prototype to technology readiness level-6 with comprehensive testing. We will quantify computing power required to process the data in real time and deliver a low-size, weight, and power working prototype with an electronics test bed to facilitate independent testing by the Army. POC believes that the proposed DEESTAC technology, when fully developed, will be the most advanced and practical airborne stationary target detector and classifier, ideal for use in both the military and civilian commercial sectors.