The need for training and benchmark datasets for convolutional neural networks in flood applications

Abstract Flood-related image datasets from social media, smartphones, CCTV cameras, and unmanned aerial vehicles (UAVs) present valuable data for the management of flood risk, particularly application modern convolutional neural networks (CNNs) to specific flood-related problems such as extent detection depth estimation. This review discusses increasing role CNNs in research with a growing number published datasets, since 2018. We note lack open labelled need an open, benchmark library classification, object detection, segmentation relevant management. Such would provide advance CNN applications general serve resource, providing scientists community necessary model training validation.