Fusing Marginal Reducts for HRR Target Identification

Rough set theory has not been applied to automatic target recognition (ATR) problems because the problems of interest were too large. The determination of reducts (classifiers) was a problem whose solution grew in exponential time with the number of range bins. This paper introduces a method which allows the determination of reducts in quadratic time and a method of partitioning the problem (reducing the number of range bins being considered) so that ATR problems can be solved in a reasonable time. A method of fusing the individual classifier results, even though they may not have performed well on the training set (marginal reduct) is introduced. This fusion of marginal reducts yields a synergistic result that produces a well performing classifier.