Deep learning based water feature mapping using Sentinel-2 satellite image

Abstract

As a foremost resource, water has a wide-ranging effect on numerous economic activities including agriculture, manufacturing, households etc. and services such as hydroelectricity generation, recreation, and amenity. Water is also very significant for the urban ecological community. Precise and systematic location of water resources is important for urban administration. Extraction of water resources and its management is extremely vital in a variety of remote sensing applications such as estimation of availability of water, flood, and drought prediction etc. This research work implements and investigates the applicability of the popular deep learning architectures such as U-Net and Encoder-Decoder for extraction of water features from sentinel-2 satellite Images of Delhi-NCR region, India. The performance of both the models have also been evaluated by calculating F1-score as 82.70%, and 89.41% respectively. In general, the proposed deep learning approach is relevant to precise water planning with improved spatial goals and exactness, which conceivably encourages effective management of water resources.