Urban land cover mapping and building extraction from high resolution images

The availability of very high spatial and temporal resolution remote sensing data facilitates mapping highly complex and diverse urban environments. Such images capture the very fine details of urban land covers, which are of great benefits to urban management and decision-making institutions. However, mapping such detailed information necessitates the combined use of images and ancillary data and very effective image analysis techniques. In this study, an object-based image classification methodology is proposed for urban land cover classification, using very high resolution images, elevation data, city zoning maps, and address points data. Logically structured classification rules based on the spectral, spatial, and contextual features of the segmented objects are developed and tested over a small urban area. The same rule set is then transferred and tested on two similar images covering larger urban areas, as well as an image from a different sensor. The land cover classification results through the transferability of the rule set proves the effectiveness of the methodology and produces satisfactory classification results with an overall accuracy of 91% as compared to the 96% accuracy achieved over the small training area. The buildings class is then further processed using semantic information from zoning maps, and detected change using reference building data. The buildings detection accuracy measures is over 90% and the missing rate is about 5%. The classification methodology based on the integrated use of multiple data produces satisfactory land cover classification. Its transferability considerably reduces both the processing time and the analyst\u27s efforts