Formulation for min-max clairvoyant fusion based on monotonic recalibration of statistics

Abstract

A formulation for min-max clairvoyant fusion (also known as continuum fusion) is developed that exploits the invariance of hypothesis testing statistics to monotonic transformations. In addition to generalizing an earlier formulation based on manipulated thresholds, the new formulation leads to efficient algorithms for two of the most widely advocated fusion flavors: one based on combining detectors with the same false alarm rate, and another based on constant detection rate. These algorithms are used to investigate and compare the performance of different detectors for a class of problems that arises from detecting small or weak targets in hyperspectral imagery. The experiments are performed on simulated data from well-defined distributions so as to isolate the effect of different flavors of fusion from the effects of model mismatch.