2020/08/17 by Marc Bosch, Christian W. Conroy, Bosch, Marc +5
Earth and Planetary Sciences · Environmental Science · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Flood Risk Assessment and Management #Image and Video Processing (eess.IV) #Tropical and Extratropical Cyclones Research #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2008.07418
openalex publication_date 2020/08/17 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We have developed a framework for crisis response and management that\nincorporates the latest technologies in computer vision (CV), inland flood\nprediction, damage assessment and data visualization. The framework uses data\ncollected before, during, and after the crisis to enable rapid and informed\ndecision making during all phases of disaster response. Our computer-vision\nmodel analyzes spaceborne and airborne imagery to detect relevant features\nduring and after a natural disaster and creates metadata that is transformed\ninto actionable information through web-accessible mapping tools. In\nparticular, we have designed an ensemble of models to identify features\nincluding water, roads, buildings, and vegetation from the imagery. We have\ninvestigated techniques to bootstrap and reduce dependency on large data\nannotation efforts by adding use of open source labels including OpenStreetMaps\nand adding complementary data sources including Height Above Nearest Drainage\n(HAND) as a side channel to the network's input to encourage it to learn other\nfeatures orthogonal to visual characteristics. Modeling efforts include\nmodification of connected U-Nets for (1) semantic segmentation, (2) flood line\ndetection, and (3) for damage assessment. In particular for the case of damage\nassessment, we added a second encoder to U-Net so that it could learn pre-event\nand post-event image features simultaneously. Through this method, the network\nis able to learn the difference between the pre- and post-disaster images, and\ntherefore more effectively classify the level of damage. We have validated our\napproaches using publicly available data from the National Oceanic and\nAtmospheric Administration (NOAA)'s Remote Sensing Division, which displays the\ncity and street-level details as mosaic tile images as well as data released as\npart of the Xview2 challenge.\n