Overview of statistical tests for unexploded ordnance detection

Abstract

In this chapter, we outline the statistical procedures that can be employed for the detection of unexploded ordnance (UXO). Phenomenological modeling is first developed to relate the collected data to a sensor's feature parameters, which in turn allow for physics-based signal processing. Starting with the Bayesian framework, we introduce minimax and robust detection that do not require prior probabilities and distributional information on the measurement uncertainty, respectively. Nonparametric tests that perform well for broad classes of distributions are also presented. Finally, the generalized likelihood ratio test is described as a joint estimation-detection method which first estimates the feature parameters and then tests for the presence-absence of the UXO. © 2009 Springer Netherlands.