An hierarchical object based image analysis approach to extract impervious surfaces within the domestic garden

Creating detailed and easy to produce land cover maps of urban residential areas continue to be a challenge. In this research, we discuss and illustrate a methodology for VHR multispectral image segmentation and classification of impervious surfaces. On two levels we have firstly detected impervious and pervious land surfaces and secondly rooftops, pathways and gardens. By using a second image, we have tackled the problem of large shaded areas in the most recent multispectral Quickbird image, used as the main input. Our final map had an overall accuracy of 63,8% and a kappa value of 0,457. These accuracy values can be considered as good, yet may be improved by further fine-tuning the approach and post-processing.