Thematic information extraction and data fusion, a knowledge engineering approach to the problem of integral use of data from diverse sources.

摘要

Multisource data integration is reformulated as a problem of defining a problem space with goals and constraints. Solving the problem of detecting objects, estimating parameters of geometric and radiometric models and classification of objects is described as searching for the goals by navigation through problem space which can be reduced to a tree search. An inference engine provides the mechanism for navigating through problem space. By defining the inference procedure as a backward chaining of rules it is possible to select only data which are relevant to current hypothesis evaluation. Backward chaining also allows the handling of cases of missing data. Information quality and error propagation are treated under the formalism of maximum likelihood f minimum cost decision making. Likelihood vectors are stored or regenerated for future use. Arguments are given for not using the Dempster, Schaefer method. The approach of defining a search space and use inference engines for navigating from initial state to goal state is contrasted with the usual approach of data merging by colour picture painting. The knowledge based approach is illustrated by hypothesis evaluation using both ordinal (remote sensing) data and nominal (GIS, attribute) data. Examples are provided of the integration of multispectral data with radar data, and on model based image interpretation applied to the recognition of buildings in airphotos.