Automated Cartographic Feature Attribution Using Panchromatic and Hyperspectral Imagery

Our focused research effort in the area of Automatic Population of Geospatial Databases (APGD) is toward automating cartographic feature attribution using high spatial resolution hyperspectral imagery, in combination with our existing cartographic feature extraction (CFE) systems running on panchromatic imagery. We believe that the use of such high-resolution hyperspectral imagery will enable more detailed and accurate surface material attributions for simulation databases, especially in complex urban areas. Fusion of this spectral information and derived surface materials with existing building and road extraction systems will greatly improve both the performance of such systems, by enabling hypothesis verification based on material type, and the cartographic utility of their output, by the addition of semantic attributions such as material type.

The goal of this basic research project is to investigate the automated extraction of semantic attribution information for manmade and natural features by the fusion of hyperspectral and panchromatic imagery. While our ongoing research on panchromatic imagery has focused on the geometric aspects of cartographic feature extraction, the generation of detailed surface material maps as well as the attribution of the composition of man-made objects has not until now been the subject of detailed analysis.

The low-to-moderate spatial resolution of multispectral data available until now, such as SPOT's 20 meter pixel size, has limited its usefulness to generating coarse descriptions of surface materials in fairly large, homogeneous areas. We believe that the high spatial resolution hyperspectral data now available will allow us to generate surface material maps with dramatically better detail and fidelity, especially in urban areas. In addition, the higher spatial resolution of the surface material map will make it applicable to cartographic feature extraction---we will be able to see and classify parts of individual buildings or roads, thereby greatly improving our hypothesis verification capabilities. We have done preliminary work in using multispectral data with moderate spatial resolution in conjunction with high resolution panchromatic imagery; this has raised important issues in multisensor registration, cross-sensor information fusion, spatial-temporal differences, and in new techniques for automated material classification, as well as verifying the power of the fusion of such data.

APGD: Automatic Population of Geospatial Databases: MAPSLab PI Overview Slide

APGD February Kickoff Meeting:

Research Project Overview

HYDICE Acquisition over Fort Hood, Texas

DARPA PI Meeting, Santa Fe, NM; February 2-3, 1998

Evaluation

(TBD)

Overview Slides

Simulared Daedalus ATM Shortwave Infrared

Fine Surface Material Clasification

Coarse Surface Material Classification

Detail of Barracks Area -- Ft. Hood

Barracks Area

HYDICE sensor information (Naval Research Laboratory)

Image Understanding Program Projects (DARPA ISO)

National Imagery and Mapping Agency (NIMA)

Topographic Engineering Center (TEC)

CMU Digital Mapping Laboratory (MAPSLab)