Flood detection using remote sensing and deep learning approaches

Abstract

A flood is an overflow of water that covers dry land. Floods can happen for a number of reasons, such as prolonged periods of severe rain, melting snow, or the collapse of levees or dams. There are various adverse effects of flood including loss of life, agricultural damage, economic losses etc. Early flood detection is important to reduce these effects of flood. Flood detection can be done in a number of ways, including the use of water level sensors, flood warning systems, remote sensing methods, machine learning and deep learning methods, etc. Our proposed approach is combination of remote sensing and deep learning to detect flood. We took sen-12 floods related dataset of sentinel-2 images in which each folder contains 12 bands images with labels flood (1) or no flood (0). We performed various preprocessing techniques on each folder bands images to reduce the impact of shadow and clouds in flood detection process. After this step, we trained various CNN models on preprocessed images and examined results.