Attentive decoder network for flood analysis using sentinel 1 images

Automatic satellite based flood identification is an essential activity. Indices and threshold based flood detection can be enhanced using learning driven approaches. In this work, we proposed a multi-scale attentive decoder based network named ADNet to identify the flood events using sentinel-1 images. We validated our proposed approach on the Sen1floods11 benchmark dataset, and it achieved better results than recently published deep learning and threshold based works. We presented detailed experimentation on several dataset settings. Our proposed work is able to effectively delineate permanent water, flood water, and all water pixels using VV and VH inputs of the Sentinel 1 images. We achieved the intersection over union score values of 0.405, 0.404 using handlabeled set on two test sets of Sen1floods11 dataset.