Detection of urban housing development by fusing multisensor satellite data and performing spatial feature post-classification

Abstract

A large number of change detection techniques have been developed, but little has been done to detect detailed changes, such as urban housing development, using satellite data. In this study a new approach is presented. By fusing multispectral satellite data, e.g. Landsat Thematic Mapper (TM) or SPOT XS, with panchromatic satellite data, e.g. SPOT pan, big buildings (10-20 m in width) in urban areas can be extracted. By performing the spatial feature post-classification, e.g. (1) co-occurrence matrix-based filtering for separating buildings from noise, (2) axis-oriented linking and segmentation for a complete extraction of urban water areas, and (3) mathematical morphology operations for improving the classified green areas, the accuracy of the extracted classes is significantly increased and a detailed urban housing map can be generated. By overlaying this map with the built-up areas extracted from data of an earlier date, newly developed built-up areas can be detected and big buildings can individually be highlighted. This new approach was tested in the urban area of Shanghai, China, using Landsat TM and SPOT pan data. More detailed change detection than with conventional methods resulted in an average accuracy of the building extraction of 86%.