Optical Matched Filters for Autonomous Infrared Seekers

摘要

The purpose of this study was to determine the capability of optical matched filtering to detect and classify infrared tactical targets. The imagery on which this determination was made was provided by the Environmental Research Institute of Michigan. The dataset consisted of 14 target types, in 370 infrared scenes of varying contrasts and intensities, from ranges of 250 m to over 12,000 m. The approach was to use a computer simulation of the optical matched filter developed by the Grumman Corporate Research Center. The Grumman approach to the problems of scale and rotational variability is to use multiple versions of the target in the reference memory. The details of this approach and the ways in which the results of the multiple reference memory are combined in the correlation plane are discussed. Matched filters were derived from the imagery as well as synthetically generated, and results for both are discussed in detail for a variety of filter-dependent and scene-dependent situations. Results are expressed in terms of detection probabilities and false alarm rates and where possible operating characteristic curves are given for various ranges and threshold values. Finally, the results point strongly to features of the approach which require improvement, and recommendations for future studies which could achieve such improvements are made.