Toward an expert system for terrain analysis.

摘要

Terrain analysis is the systematic study of image patterns relating to the origin, and composition of distinct terrain units called landforms. It takes into account and provides information about physical site factors which are used by civil engineers for evaluating the suitability of a site for a terrain related engineering application. Terrain analysis is a time consuming labor intensive process and requires a significant degree of expertise. In this dissertation, an expert system paradigm has been adopted, for developing a computational approach to terrain analysis problem solving. A methodology was developed for the representation and management of uncertain terrain knowledge. The "vagueness" that is inherent in the descriptions of terrain analysis terms was represented using fuzzy models. The Dempster-Shafer theory of evidence was adopted to establish hypotheses about the type of terrain based on observed evidences. A goal directed backward form of reasoning was employed for evaluating the suitability of a site for a terrain related engineering application. The reasoning strategy was formalized in production rules, and the fuzzy models of terrain terms were formalized in frames. Procedural computations were formalized in LISP code. The methodology was implemented in the Terrain Analysis eXpert (TAX) system. TAX was developed by employing the expert system shell KEE (Knowledge Engineering Environment) and the image processing package ELAS (Earth resources Laboratory Application Software). TAX was tested with a real data set consisting of a digitized color infra-red photograph and digital elevation data. The conclusions arrived at by TAX compared favorably to those reached by an expert who analyzed the same site using traditional photointerpretation techniques.