A fuzzy rule-based system to classify multispectral Ikonos imagery

This paper describes a fuzzy rule-based system for supervised classification of multispectral high-resolution images. The new method uses as input variables combinations of bands, like NDVI, that detaching certain terrestrial features. The membership functions of the fuzzy input variables are defined starting from obtained thresholds of samples selected previously. These variables are used to generate the fuzzy rules in way to detect in the Ikonos image the following land cover types: low vegetation, arboreal vegetation, forest, water, buildings and road. The classified image by the proposed method is compared with the results generated by the maximum likelihood method, and the system with only five fuzzy rules generated better results in the classification.