TARGET LOCATION BY A MULTIAPERTURE VISION SYSTEM COUPLED TO A BACKPROPAGATION NEURAL-NETWORK

A software simulation of a multiaperture vision system (MAVS) that loc ates a point-source target by means of a neural network processor is p resented. Models are developed and tested for one-, three-, and nine-e yelet systems. The advantages and limitations of each MAVS system are discussed. A back-propagation neural network is developed and trained as a target-location processor for each system. The accuracy of the ne ural network in determining the spatial position of a point-source tar get placed inside the field of view is then determined. The performanc e of each of the three systems is compared, and the accuracy of the sy stem is shown to increase with the addition of more eyelets.