Waterbody information extraction from remote-sensing images after disasters based on spectral information and characteristic knowledge

This article proposes a post-disaster waterbody information extraction method based on spectral information from remote-sensing images and characteristic knowledge that can resist interference from factors such as changes in water quality, waves caused by accelerated water flow, and varying water levels. The method first analyses the display characteristics of waterbodies from remote-sensing images (their spectral characteristics, geometric features, and textural features), forming a decision tree of rules that represent characteristic knowledge for waterbody information extraction. This rule set is added to the various processing stages of waterbody information extraction after disasters to construct a waterbody information extraction model. Second, an object-oriented method is used for image segmentation. A rough initial waterbody information extraction is performed based on spectral information, and then refined based on the characteristic knowledge. Third, noise is eliminated and holes are filled in the images of the refined waterbody information extraction results. Finally, the accuracy of this new waterbody information extraction method is evaluated from both qualitative and quantitative aspects. Accuracy assessments of the experimental results obtained using remote-sensing images from the Wenchuan earthquake and a 2010 flood in Pakistan show that the proposed method is both efficient and accurate at extracting post-disaster waterbody information even when the background is complex.