Neural Networks to Solve the Closely-Spaced Objects Problem

Aggessive training algorithms will allow neural networks to separate closely-spaced objects in target identification for missile defense. Resulting neural networks will show an order-of-magnitude improvement in extracting individual target information from sensor data. This focal plane sensor data processing, performing target identification and decoy discrimination functions as the "eyes and ears" of missile defense systems.The project objective is to demonstrate that neural networks can be focused to accurately extract individual target information from detector outputs of infrared sensors observing multiple point-targets. The technical approach provides accuracy, numerical stability and reduces computation cost by addressing function complexity, network capacity and training algorithm aggressiveness. Reduced computational cost subsequently reduces system power, weight and size. This technology can be further applied to robotics and image processing.