Unsupervised Impervious Surface Change Detection Using a Coarse-to-Fine Hierarchical Approach

Detecting changes in impervious surface from remote sensing images helps evaluate urbanization and environmental impacts and provides an important basis for urban planning and ecological protection. Most existing impervious surface change detection (CD) methods are supervised, which makes them time-consuming and laborious in practical applications. In this letter, a coarse-to-fine hierarchical approach, RIDKDC, which refines initial land cover CD using a domain knowledge driven constraint (DKDC) method, was proposed for unsupervised impervious surface CD. First, the initial land cover CD map containing all types of changes is generated by iteratively reweighted multivariate alteration detection (IRMAD) method. Then, DKDC is applied to mask the nonimpervious surface changes and pseudo changes, retaining only the changes in impervious surface. Finally, the desired result is obtained by mathematical morphology processing. Experimental results show that the proposed RIDKDC outperformed the compared methods.