Flood Extent Mapping: An Integrated Method Using Deep Learning and Region Growing Using UAV Optical Data

• Published in: IEEE journal of selected topics in applied earth observations and remote sensing Vol. 14; pp. 2127 - 2135
• Main Authors: Hashemi-Beni, Leila, Gebrehiwot, Asmamaw A.
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial neural networks; Classification; Classifiers; Convolutional neural network (CNN); Data augmentation; Deep learning; Digital Elevation Models; Digital imaging; Emergency preparedness; Emergency response; Flood mapping; Flooded areas; Flooding; Floods; Hydrologic data; Image classification; Image enhancement; Imagery; LiDAR; Mapping; Neural networks; Optical imaging; Optical sensors; Plant cover; region growing (RG); Remote sensing; Training; Vegetation; Vegetation mapping
• ISSN: 1939-1404; 2151-1535
• DOI: 10.1109/JSTARS.2021.3051873

Flooding occurs frequently and causes loss of lives, and extensive damages to infrastructure and the environment. Accurate and timely mapping of flood extent to ascertain damages is critical and essential for relief activities. Recently, deep-learning-based approaches, including convolutional neural network (CNN) has shown promising results for flood extent mapping. However, these methods cannot extract floods underneath vegetation canopy using the optical imagery. This article attempts to address this problem by introducing an integrated CNN and region growing (RG) method for the mapping of both visible and underneath vegetation flooded areas. The CNN-based classifier is used to extract flooded areas from the optical images, whereas, the RG method is applied to estimate the extent of floods underneath vegetation that are not visible from imagery using the digital elevation model. A data augmentation technique is applied for training the CNN-based classifier to improve the classification results. The results show that the data augmentation can enhance the accuracy of image classification and the proposed integrated method efficiently detects floods in both the visible and the areas covered by vegetation, which is essential to supporting effective flood emergency response and recovery activities.