Entire implementation is present in the notebook as well as splitted in scripts.
The paper will be be published in IEEE-XPlore after it is presented in IEEE-IGARSS conference, Belgium-2021. Please reach out to us via email for further details about the project.
In case you have any questions, please reach out to us at email: Neeraj or email: Akshita
Synthetic Aperture Radar (SAR) images most often get corrupted by multiplicative granular noise known as ‘speckle’ during the remote acquisition. In this paper, we propose an unsupervised deep learning approach for SAR despeckling without the need of clean SAR ref- erences. To this end, we employ Stein’s unbiased risk estimator (SURE) that facilitates estimation of the true mean-squared error between output of despeckling net- work and the ground truth, however, using only noisy input SAR images. We utilize thus unsupervised trained U-Net architecture, coupled with the proposed SURE- based loss function. Experiments are conducted on syn- thetic as well as real Sentinel-1 SAR images. The results are compared both qualitatively and quantitatively with popular methods in SAR despeckling.