Land cover differentiation using digital image processing of Landsat-7 ETM+ samples

This study presents the implementation of image processing techniques in satellite bands of the visible and infrared spectrum for the differentiation of land cover in the Colombian Andes, which is dominated by high mountain systems, to do this, we extracted features through texture analysis methods, principal component analysis, discrete cosine transform, combination of color spaces and vegetation indices to be combined together, we have obtained independent zones that guaranteed the separability of 9 classes and correct discrimination in step classification. In the classification module we used two approaches: a neural network and an expert system, with which it was possible to obtain a success rate of 77.38% and 86.36% respectively. tember 12 of 2008 and January 21 of 2010, obtained from United States Geological Survey (http://glovis.usgs.gov), in order to perform the correction to the SLC-off problem. Nine cover types or classes were selected and for each one of them were took the coordinates of 12 insitu points, these data were supplied by the Grupo de Estudios Ambientales (GEA) and taken on 2011. The classes studied were: natural forest, planted forest, type 1 crop, type 2 crop, grass, city, pántano, water and rock.