Object-Based Approach and Tree-Based Ensemble Classifications for Mapping Building Changes

The aim of this paper is to efficiently detect and identify the building changes from newly registered very high spatial resolution (VHSR) image by comparing with outdated map. The whole process is performed mainly on four steps. First, the image was segmented to generate primitives, which are then represented by a feature vector composed from spectral, geometric, textural and contextual attributes. Thereafter, tree-based ensemble methods (Bagging, Random Forest and Extremely Randomized Trees) are used in a classification step. The final objects' prediction is deducted thanks to better classifier error rate. Last, a post classification change detection step allows to identify the segments which represent building changes. The data used in this research concerns the city of Rabat (Morocco). A Quickbird image has been used with an old map at the scale of 1:10,000. Regardless of the quality of the detected buildings' shape, the method achieves good rates of completeness (94%) and correctness (92%).