Clasificación orientada a objetos en fotografías aéreas digitales para la discriminación de usos del suelo

traditional classification techniques, basically pixel-based approaches, are limited. typically, they produce a characteristic "salt and pepper" effect, and are unable to extract objects of interest. these techniques have considerable difficulties in dealing with the rich information content of medium and high-resolution images. one alternative to these classification systems can be a previous segmentation of the image to be classified. in this way, object-based classification can be performed so that a significant increase on the information that can be extracted is obtained. in the present work, the aim is to obtain a digital classification of wilderness-urban interface areas that can be used by fire management services. to this end, a digital aerial image provided by the dmc sensor was segmented and classified using ecognition software, which allows homogeneous image object extraction. the meaningful image objects obtained were then used for the classification. segmentation before classification worked out as an efficient image analysis technique, overcoming traditional approaches limitations. digital aerial photographs and data of 350 plots in granada, spain, were used to validate the classifications obtained; the overall classification accuracy of 90% and an excellent kappa statistic (85%) for the object-based classification, proved the validity of this method.