Document Type : Original Article
School of Surveying and Geospatial Engineering, College of Engineering, University of Tehran, Tehran, Iran
University of Tehran
As one of the natural hazards that occur in some parts of the world, floods have devastating effects on people, the environment, and infrastructure. The management of this crisis can be effective in reducing severe financial and life losses. A critical aspect of managing this natural disaster is the accurate identification of flooded areas and their trends. The article presents a method for the identification and segmentation of flooded areas using the UNET++ neural network and Sentinel-1 satellite images in c-band and single-polarized (HH and VV) and double-polarized (HH+HV and VV+VH) forms. These images were provided by NASA from Nebraska, Alabama, Bangladesh, Red River North, and Florence. The labeling process for all these images was done by the NASA implementation team and the IEEE GRSS Earth Science Informatics Technical Committee. In this network, EfficientNet-B7 is used as an encoder for feature extraction. Based on evaluation criteria such as IoU, F1-score, Recall, Accuracy, and Precision, the efficiency of the model has been evaluated. This model has demonstrated a high potential for detecting and segmenting flooded areas. Using this method, 84.77% IoU is obtained, which is higher than other methods such as UNet and FPN neural networks which participated in the ETCI (the maximum IoU obtained by these methods is 76.81%).