Target-driven extraction of built-up land changes from high-resolution imagery

Abstract

Information on land conversion to modern urban use is needed for many studies such as the impact of urbanization on environmental quality. Although extensive remote sensing research has been undertaken to detect conversion of nonurban to urban lands, little effort has been directed at assessing modernization of existing built-up land. Detection and quantification of this class of urban growth present significant challenges since the difference between radiometric signatures before and after “land modernization” is much more subtle and complicated than the case of conversion from typical rural to impervious urban land surfaces. A target-driven approach is presented for an efficient extraction of built-up land change distribution that provides superior results to those based on the traditional data-driven land cover approaches. The extraction strategy, integrating pixel- and object-based methodologies, is comprised of three components: delineation of the baseline built-up areas, detection of the areas that have undergone change, and integration of targeted change features to generate a final built-up land change map. A case study was carried out using RapidEye and SPOT5 images over suburban Beijing, China. The overall accuracy of built-up change mapping is about 91% and exceeds accuracies achievable by pixel or segment processing used in isolation.