Rapid flood detection in global scene based on multimodal data collaboration

Flood detection in remote sensing images plays an important role in flood emergency monitoring. Compared with permanent water body detection, flood detection faces unique challenges. The scene in flood-stricken areas is often more complex. During the evolution of flood disasters, weather conditions often change dramatically. Only using Multispectral or Synthetic Aperture Radar (SAR) images cannot effectively and comprehensively obtain accurate flood disaster information. It is necessary to fully tap the potential of collaborative application of multimodal data such as Multispectral, SAR, and background data DEM in disaster area. However, the lack of globally distributed multimodal datasets limits the collaborative use of multimodal data for flood detection. Firstly, this study collects and compiles major flood disasters’ data. Based on those data, this study construct a global scene multimodal remote sensing flood dataset. Secondly, a multimodal fusion fast flood detection model based on multi branch structure is proposed for large-scale flood detection in multimodal data. This model achieves the highest F1-score of 94.71% for SAR images and 96.17% for Multispectral images in flood detection. Finally, this study constructed a multimodal data collaborative flood detection technology route, which can realize large-scale and large-area flood disaster emergency monitoring around the world.