SBIR Phase I: Deep Learning Framework for Mapping Flood Extent from Unstructured Photos

The broader/ commercial impact of this Small Business Innovation Research (SBIR) Phase I project is in helping communities handle floods better by turning public photos and videos into near-real-time useful data. Currently, it is difficult to get a clear picture during emergencies because sensors and satellites are often too far away or field surveys are too expensive. This system uses smartphone images to create real-time maps that help rescuers and city planners see exactly where the water is. This technology makes flood response faster and more accurate, which helps save lives and reduce property damage. For businesses, this tool provides a scalable service for emergency responders, infrastructure planners, utilities, and insurance companies, creating new jobs in the tech industry. Overall, this project creates a more affordable and effective way for the nation to manage flood risks using the photos people already take on their phones. The core technical innovation of this project is a high-risk, difficult-to-replicate method for automatically converting individual flood photographs into precise, georeferenced flood extent maps using artificial intelligence. This effort utilizes advanced computer vision to identify and match ground control points between unstructured images and geographic maps, a process that has historically required time-consuming manual intervention. This innovation is particularly challenging because it requires identifying stable reference features in unstructured images that may be noisy or partially obscured by floodwater. The Phase I research focuses on overcoming the primary technical barrier: automatically identifying reference features in unstructured images and matching them to existing geospatial data. The research scope includes automating image segmentation for flooded areas, developing and testing deep learning models that detect visual features and align them with mapped reference points, and leveraging monoplotting techniques to project observed flood boundaries onto elevation maps. The resulting localized flood extents will then be combined to form continuous maps of peak inundation at the storm level. The work will demonstrate technical feasibility, quantify accuracy, and reduce uncertainty in an automated workflow suitable for operational use. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.