Automatic Cartographic Feature
                Attribution Using Panchromatic and
                     Hyperspectral Imagery

                  Digital Mapping Laboratory
	          Computer Science Department
                  Carnegie Mellon University

                     PIs: Dave McKeown
                         Chris McGlone

		   Period of Performance:  
                   April 1997 - March 1998

                ==================================


The goal of our program of research is to demonstrate feasibility of
210 channel airborne hyperspectral imagery (HYDICE) to construct surface 
material maps of cartographic features.  Traditional remote sensing 
applications do not perform geometric analysis of scene content and have 
a tolerance for positional error that is generally measured in hundreds of 
meters.  The goal in cartographic feature extraction is high geometric
accuracy as well as production of semantic attributions of surface
material, both natural and man-made.  Using hyperspectral imagery 
at a spatial resolution comparable to that of panchromatic imagery
our research goal is to fuse material maps with building or road
hypotheses, or with stereo elevation information derived from panchromatic 
imagery. In order to accomplish this task, accurate geopositioning 
of HYDICE imagery is crucial.  Additionally, radiometric effects must 
be accounted for to consistently classify surface materials across multiple 
HYDICE flightlines.

This slide gives an overview of these issues and shows some sample
image data and processing results.  The left image is from a panchromatic 
aerial frame image with a ground sample distance (GSD) of 0.3 meter. 
The right image shows a coarse surface material map derived from HYDICE 
hyperspectral (210 spectral bands) imagery with a GSD of 2.0 
meters over a barracks area in Fort Hood, Texas.  The panchromatic imagery 
shown is representative of imagery used in our cartographic feature analysis 
systems.   One goal is to use coarse and fine surface material maps such
as shown here to provide attributions for the building, road, and natural 
features extracted from the panchromatic imagery.  Analysis of these
material maps can also lead to object descriptions or as a cue for further
image analysis.