AN EXPERT SYSTEM FOR FOREST AND LANDUSE CLASSIFICATION OF REMOTELY SENSED IMAGES

In the field of forestry, computer image procesing has long been used as a subsidiary means to support human interpretation, instead of as a means to process the images independently.One of the reasons is that the conventional methods of computer classification lack the abilities of making effective use of ancillary data and human knowledge during the processing.A PC-based experimental expert system for forest and landuse classification of remotely sensed images is built to remedy this shortcoming.This paper describes the overall architecture, knowledge base and evidential reasoning strategy of the system.The explanation of a new method of pre-processing of digital images is also included.The system was tested with a TM image on a mountain area in North China.Both the reults from the system and a conventional method were checked by human interpretation and the comparison is reported.