Abstract
Producing vegetation maps is one of a myriad of uses that remotely sensed data is being used for. Low error rate classifers can be obtained from the training data generated from surveyed sites and expert knowledge. However, when these classifers are applied to an entire remotely sensed image to produce a map they contain at least many generalisations and at worst gross errors. This is, in part, due to the limited nature of spectral information and limited amounts of training data. In this paper we investigate a technique, called reinforcement classifcation, to generate more accurate classifcations of remotely sensed images. We demonstrate reinforcement classifcation using C4.5 although it is general enough to be applied to any domain and classifcation scheme.