DEEP FOCUS: USING DEEP LEARNING TO DISCERN TARGETS IN CLUTTERED RADAR

Deep Focus applies deep learning neural networks to Apache Fire Control Radar (FCR) targeting and target identification, with applicability to related systems. Recent innovations in deep learning theory and implementation have enabled neural nets to achieve what was once unthinkable: beat humans at complex image recognition skills, safely pilot cars over chaotic road systems, and overwhelm Grandmaster Lee Sedol in the game of Go, a challenge previously thought immune to AI because of the game’s near-infinite complexity. Deep Focus builds upon Phase I results to more accurately identify vehicular targets despite high noise. While training a deep neural net is computationally intensive and requires specialized hardware, execution is computationally inexpensive and can be implemented with very modest CPU and memory requirements. Phase 1 of the project proved the feasibility of the theory by training a deep learning neural net to analyze radar images and output an accurate target identification. The success metrics included false negative and false positive rates for radar targets. Phase II builds upon the highly successful Phase I results and augments Deep Focus's classification accuracy and flexibility by training on a wide range of new datasets. Computer vision and deep learning are core CLOSTRA competences.