Satellite image classification with artificial intelligence methods

Abstract

The scope of this article is to acquaint Slovene geologic public with rudiments of image Processing of satellite data and especially with classification issue. These methods are not useful only with remotely sensed data but can be also used with any other kind of spatial data (geophysical, geochemical O). Standard classification methods based on statistical principles do not always give satisfactory results. Therefore a variety of new approaches are being tested in order to achieve better accuracy. One of the most promising fields is artificial intelligence where artificial neural networks (ANN) have proven to be useful. In this article two methods have been tested, unsupervised learning with Self Organising Maps and supervised learning with backpropagation network. In comparison with standard approach better results were gained especially in more complicated cases where classes are not linearly separable. One of the advantages of ANN is that X and Y coordinates can be incorporated in learning process. In this way much better accuracy is achieved. This kind of model has ability to favour certain classes according to spatial position of input data, but has disadvantage of not being the general model. It is strictly aplicable only to examined area.