Artificial neural system approach to IR target identification

摘要

Abstract: A system has been developed which successfully blendsthe strengths of digital image processing with theadvantages of artificial neural systems (ANS) toachieve target identification in non-optimumenvironments. Infrared (IR) video images of fivedifferent military vehicles were acquired to provide adata base of targets. These images were digitized andpreprocessed using conventional digital imageprocessing techniques. A novel ANS model was developedto analyze certain features of the images and performclassification of the targets. Tests on real IR targetimages have shown that the system was capable ofcorrectly identifying any of the five targetsconsidered to an accuracy of 86% independent of targetorientation and location within the scene.!