2nd International Conference on GeoComputation

Eliciting Knowledge with Visualization - Instant Gratification for the Expert Image Classifier who wants to Show Rather than Tell

Paul Crowther and Jacky Hartnett

Artificial Intelligence and Spatial Systems Research Group
Department of Computing
University of Tasmania

Email: P.Crowther@utas.edu.au

Presented at the second annual conference of GeoComputation 97 & SIRC 97, University of Otago, New Zealand, 26-29 August 1997

Abstract

GIS systems the world over are awash with data that experts can classify visually. This process is time consuming and costly. Expert Systems have been built which attempt to at least pre-classify images and hence speed up the process. To build these systems it is necessary to elicit information from the human expert classifiers in order to assist the classification of these many hundreds of images. Traditionally this knowledge has been captured through interview and protocol analysis. However, this required either the expert classifier to describe verbally what they were seeing or the expert systems developer (knowledge engineer) to interpret what they were being shown.

To overcome this problem, a visual knowledge acquisition tool, KAGES (Knowledge Acquisition for Geographic Expert Systems), was developed. Impetus to the development of this tool was given to our group by the need to classify many remotely sensed images of Antarctica in order to provide information on global climate change and Southern Ocean currents.