Intelligent interpretation of SPOT data for extraction of a forest road network

Abstract

This paper describes the development of a (semi)automated process for detection of a road network in a forest area near Perth, Western Australia using SPOT satellite images. Initially, the imagery was processed with specially designed local spatial operators, where the primal sketch was extracted. The data were then linked and vectorised, and C programs used to extract low-level image features from the spatial domain. The method also utilised morphological operators to give a geometric description of the linear features. Subsequently, a prototype expert system was implemented to improve the results. Knowledge was encoded in the form of rules, incorporated into a backward chaining expert system shell, VP-Expert. Based on the attributes of the low level image features, the knowledge base defines the road network by eliminating extraneous elements from the image. The system utilises only data derived from the image, without ancillary information from other sources, keeping the process ‘autonomous’. Results s...