The development of a method for semi-automatic classification of built-up areas from aerial imagery

The identification of built-up areas from aerial imagery is essential for map updating and detecting changes for a national mapping organisation. At the Chief Directorate of National Geospatial Information (CD: NGI), South Africa’s national mapping agency, this process relies on operators to manually digitise features of interest, which is very time consuming and labour intensive. It is therefore necessary to explore methods that will assist in automatic classification of such areas to speed up the process of updating topographic databases. This pilot project attempts to automate the classification of urban built-up areas, which are important as they can grow and change rapidly, and can indicate where other landscape changes may have occurred. The method of classification should be suitable in multiple South African landscapes. This research evaluates various image classification methods, and illustrates the development of the proposed methodology for urban scene land cover classification in South Africa. Three of the methods evaluated were tested and compared, and the results achieved indicate that object-based classifiers are more suitable than pixel-based classifiers in detecting urban built-up areas. The importance of generating suitable image objects in an object-based classification is indicated in this study, and the significance of texture measures used in classifying urban built-up areas is highlighted.