Post-Flood Disaster Mapping Using Deep Neural Network

This work proposes an efficient approach for post-flood disaster change detection in the Derna region, Libya, utilizing a pair of Sentinel-2 images captured before and after the flood disaster. A dataset was generated by pairing these two temporally distinct images of the same scene, each composed of four spectral bands, with a corresponding label image indicating change or no change for each region of interest. This dataset was then fed into a custom-designed deep neural network (DNN) for change classification. The DNN was trained on this data to automatically identify and map areas that had undergone changes following the disaster. This method effectively combines the potential of Sentinel-2 images and deep learning to facilitate precise and automated post-disaster change detection, thereby contributing to emergency assessment and management.