Algorithms with attitude

Abstract

A new methodology has been developed for creating detection algorithms for the class of composite hypothesis testing problems. Rather than estimating unknown parameter values for each variate test value, as in the generalized likelihood ratio test, continuum fusion methods integrate an infinite number of optimal detectors, one for each parameter value. The final form of the algorithm depends on the type of threshold constraint enforced during the fusing procedure, and this choice defines an attitude in a design process. The attitude can be tailored to suppress outliers not well described by statistical models and yet common to realistic remote sensing problems.