Object-oriented Analysis of High-resolution Satellite Images for Intra-urban Land Cover Classification: Case Study in São José Dos Campos, São Paulo State, Brazil

The detailed analysis of urban regions is among those areas which most benefited from the availability of high-resolution satellite data such as e.g. IKONOS and QUICKBIRD. These data offer as well high spatial, radiometric and temporal resolution, competing with aerial photographs for several applications. Merging these characteristics allows the detection of intra-urban targets and consequently proves to be suitable for mapping urban and intra-urban land cover using automatic classifiers. Taking into account the huge volume of data at each scene from these sensor systems (11 bits, 2048 gray levels) the conventional pixel-by-pixel classifiers, considering only spectral characteristics, show clear limitations for classification tasks. An alternative to this shortcoming is the incorporation of other types of attributes to the classification process, such as shape, size, color and contextual information. Being so, we used an object-oriented classifier from software package eCognition 4.0 which is an effective option, since it uses both topologic (neighborhood, context) and geometric information (shape and size). In this frame, an image classification experiment was conducted for test-site São José dos Campos, São Paulo State (Brazil), where a classification scheme was conceived and applied using both IKONOS and QUICKBIRD data. The classification results were compared and evaluated in order to assess which sensor allows best classification performance in such a highly complex and heterogeneous environment.