Abstract

Optical correlators have successfully been integrated with missile guidance electronics to track and intercept cooperative targets in benign environments. Subsequent development of optical correlators stresses the use of realistic targets in cluttered backgrounds. This requirement results in the use of a large filter space representative of a wide class of targets. Therefore, initial selection of a correlation filter by the tracker is computationally expensive because of the large search space involved. A hybrid paradigm employing expert systems technology and neural networks can efficiently locate likely target regions in the missile's field of view and provide tentative identification of target type for use by the tracker. This presentation will introduce expert network architectures currently under consideration and compare their performance against test images with varying degrees of target and background complexity. Hardware utilization and processing latency will be evaluated for each network architecture in an attempt to present some of the design trade-offs inherent in the construction of these systems.

© 1992 Optical Society of America