The latest advances in the spatial resolution of orbital sensor systems have effectively increased our capacity to discriminate Earth surface targets. One of the application fields mostly favoured by this new type of sensor data is the remote sensing of urban areas. Although urban remote sensing already disposed of information sources with high spatial resolution (aerial photos), this application field could not rely so far on a data type that offered high spatial resolution and at the same time high radiometric and temporal resolutions. The merging of these characteristics enables the detection of intra-urban targets, and hence, proves to be suitable for mapping urban and intra-urban land cover with the aid of automatic classifiers. At purpose, the application of automatic classification routines to high spatial resolution images has been facing many challenges, for such images present remarkable noise as well as high intra- and inter-classes spectral variability. In this sense, pixel-per-pixel classifiers have shown to be limited for this kind of classification, since they can basically handle spectral information, what is widely acknowledged as insufficient to distinguish features on the intra-urban scale. An alternative to this shortcoming is the incorporation of other types of attributes to the classification process, like shape, size and contextual information. In this way, the object-oriented classifiers arise as an effective option to conventional pixel-per-pixel classifiers, once they make use of topologic information (neighbourhood, context) and geometric information (shape and size) as well. This scientific research is committed to explore the object-oriented approach in the intra-urban land cover classification of high spatial resolution images (IKONOS II and QuickBird) for the municipality of São José dos Campos – SP, Brazil. With this end, two experiments have been conducted: (a) Experiment I, carried out for a complex intra-urban setting; and, (b) Experiment II, accomplished for a smaller intra-urban area. In the first experiment, a classification scheme has been conceived and further applied to the whole study area, using both sensor images. The classification results have undergone comparison and evaluation analyses, aiming to assess which sensor presents the best performance in such a highly complex and heterogeneous environment. The goal of Experiment II was to evaluate the influence of urban occupation on the performance of land cover classification. For that, five districts of Sao José dos Campos with different spatial patterns were selected.

Master´s Thesis document