An “exclusion-inclusion” framework for extracting human settlements in rapidly developing regions of China from Landsat images

Abstract

© 2016 Elsevier Inc. Satellite based human settlement extraction at medium resolution (30 m) with supervised classification has been widely carried out. However, adequate training sample collection and mapping accuracy are two hindering factors over large regions. Here we propose a new framework for efficient human settlement extraction from Landsat images over large areas. First, an inventory-based training set is adopted to obtain some statistical parameters required to build a non-settlement mask. The mask can not only reduce unnecessary computation but also reduce the impact of background noise. Thereafter, for the un-masked areas we calculate the similarity of each image pixel to pre-collected sample points, and only those within certain threshold are treated as the settlement class. This approach is very fast and has been applied to three rapidly developing regions in China. Accuracy assessment indicates that the mean overall accuracies are 87%, 89% and 89% for Jing-Jin-Ji region, Yangtze River Delta and Pearl River Delta, respectively. This work may be applied to human settlement extraction at even broader spatial scales.