Abstract
This paper presents an enhanced convolutional neural network model, AgriFloodNet, to map flood-affected agricultural lands on remote sensing images. It consists of a dual patch convolutional neural network to effectively learn the permanent and flood water features from the bi-temporal Synthetic Aperture Radar (SAR) images of Sentinel-1 and the agricultural land features from the pre-flood cloud-free multispectral images (MS) of Sentinel-2. A decision-level fusion of the processing results of multisensory images is applied to create a change map to depict flooding in agricultural lands. The proposed AgriFloodNet shows superior performance in flood mapping on the images of agricultural lands from the SEN12-FLOOD dataset with an accuracy of 98.75%. In comparison, the single-sensor solutions give accuracies of 96.88% and 91.11% for SAR and MS images, respectively. Additionally, the model has been evaluated for a new flood event in Patna, India. It is estimated that 75% of the test site was covered with agricultural lands when the flood occurred, of which 61% was affected.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call