AN EXPERT-SYSTEM FOR LAND-COVER CLASSIFICATION

The analysis of remotely sensed imagery for various applications have evolved over the past three decades from manual (visual) interpretatio n to computer-based approaches, Sophisticated digital techniques have been developed to tackle specific image processing tasks in the analys is, and the need for efficient reference to collateral data like maps have led to special handling systems called Geographic Information Sys tems, However, the automation of the aspect of human expert's interact ion is still in its adolescence, and only recently has there been some interest in finding methods or models for representing the expert's k nowledge and inference mechanism so as to build expert systems for rem ote sensing image analysis. In this paper a framework to represent a b road class of problems in the analysis of remote sensing imagery is pr oposed, and an inference engine to tackle such problems is derived, A simple model for spectral knowledge representation is used along with a method for quantification of knowledge through an evidential approac h, An automatic knowledge extraction technique is also proposed to gat her knowledge from training samples. The techniques of knowledge extra ction representation and inferencing have been used to do a land cover analysis on two data sets, and the results are compared with contempo rary digital techniques, It is found that the proposed approach has th e advantages of avoiding commission errors, and can incorporate non-sp ectral and collateral knowledge, while its accuracy using only spectra l knowledge is comparable with standard digital methods,