Invited to present
``Unmanned Aerial Vehicle-Based Remote Sensing 
  for Detecting Minefields''
at the Eastern North American Region (ENAR) 
of the International Biometric Society annual meeting,
Charlotte, NC, March 2001.


Session Organizer: Ron McRoberts


AUTHOR: Carey E. Priebe & Daniel Q. Naiman

CONTACT: Carey E. Priebe
         Johns Hopkins University
         Department of Mathematical Sciences
         Baltimore, MD 21218-2682
         tel: (410) 516-7200
         e-mail: cep@jhu.edu

TITLE: Unmanned Aerial Vehicle-Based Remote Sensing for Detecting Minefields

ABSTRACT: 
  Minefield reconnaissance via unmanned aerial vehicle is an
important problem currently receiving much attention in the
engineering and scientific literature.  Multispectral imagery of an
area of interest is processed.  First, potential mines are located
with a mine detection algorithm.  The detector produces a binary
detection map such that D(x)=1 for all points in the image domain
subset at which a mine or minelike object is detected.  Categorizing
the candidate detections into `true targets' (mines) and `false
targets' (minelike objects, debris, noise, etc.) and considering an
operational imperative imposed on the mine detector to find (nearly)
all true targets, it can be expected that the number of false targets
in the detection map will be relatively high.
  Among the most promising approaches to the minefield detection
problem are statistical methods which consider the map of candidate
detections to be a realization of a spatial point process.  These
methods proceed by analyzing the detection map for clustering or
regularity to determine if it represents a minefield point pattern
buried in noise, or noise alone.
  It is common to consider the detection of clustering in point
processes as a formal hypothesis test, with the null hypothesis
represented by homogeneity, or complete spatial randomness.  For
the minefield detection application, in which minefields and false
detections are modeled as point processes, the null hypothesis is
`no minefield.' While there are cases for which regularity patterns in
the observed point process can be used as a key to minefield
detection, other applications do not allow for this restriction of the
alternative hypothesis of `minefield present'.  It is these latter
situations, in which minefield detection becomes a test of homogeneity
against a more general alternative of nonhomogeneity, that we address
here.