A PCA-based hybrid approach for built-up area extraction from Landsat 5, 7 and 8 datasets

Urban expansion monitoring and organization can be performed through space-based observation thanks to the revisit time and level of details guaranteed by satellite remote sensing. In particular, the Landsat mission products are the most used thanks to the long time coverage and open access policy. This paper proposed a hybrid method - combination of pixel- and object-based analysis - in order to automatically extract built-up areas from Landsat imagery. Segments are delineated from spectral indices computed in order to increase the spectral distance among the different land cover classes. The principal component analysis is applied to the original bands and constitutes the pixel-based side of the method. Segments and PCA are then combined and classified using an unsupervised approach. Results of the method were quite satisfying with an average Kappa value over 0.5 in both case studies.