In order to apply the methodology of unsupervised change detection on the complete city extent of Dongying, building geometries were delineated area-wide. The city comprises a total of around 30.000 buildings in 2013, whereas the number of changed buildings since 2006 was determined based on multi-temporal QuickBird and WorldView imagery by means of change detection.
After preprocessing of the imagery, clustering was applied to the multi-dimensional feature space of the object-based image difference vector in order to classify changed buildings according to Leichtle et al. (2017). A comprehensive number of around 7000 buildings were classified visually to serve as a control data set. In terms of accuracy, we achieved an overall accuracy of 87% and kappa statistics of 0,73.
Fig. 1 shows an excerpt of the classification result for the Eastern part of Dongying while Fig. 2 shows part of Western Dongying. Overall, the change detection workflow detected around 12.500 changed and 17.500 unchanged buildings within the city.
Fig.1: Result of change detection in the Eastern part of Dongying
Fig. 2: Result of change detection in the Western part of Dongying
Reference: Leichtle, T., Geiß, C., Wurm, M., Lakes, T., Taubenboeck, H. (2017). Unsupervised change detection in VHR remote sensing imagery – an object-based clustering approach in a dynamic urban environment. International Journal of Applied Earth Observation and Geoinformation. 54. pp.15-27.