Home

SAR2SAR: a self-supervised despeckling algorithm for SAR images

Based on the work of Emanuele Dalsasso, Loïc Denis, Florence Tupin. Link to Repo

The code is made available under the GNU General Public License v3.0: Copyright 2020, Emanuele Dalsasso, Loïc Denis, Florence Tupin, of LTCI research lab - Télécom ParisTech, an Institut Mines Télécom school. All rights reserved.

Please note that the training set is only composed of Sentinel-1 SAR images, thus this testing code is specific to this data.

How to Use

Before we Dive in to explain how this model works, we first need to consider the type of data we are going to be working with. If we are looking at an image file, we will have to preprocess the file into a NumPy array and will be handled by 'preprocess.py' and you will have to use one of the various provided solutions to help in the conversion of the image to a NumPy array.

If the data is going to be a GDAL Image, you will have to dump the image as a NumPy array and will be handled by the Notebook called '00_GDAL.ipynb'.

1. Preprocessing: The first step would be to preprocess your data, into a NumPy array so that the model can accept that data. Make sure to take a look at Preprocessing Wiki Page to choose how to proceed with this step, according to the type of input you wish to operate on.

2. Processing: Now that we have data in a usable NumPy representation, we give the NumPy representation of the SAR Image as the model's input for despeckling. Make sure to take a look at Processing Wiki Page to choose how to proceed with this step and more information on the expected parameters.

3. Inpainting (optional): If the Images generated by the model have artifacts in them, they can be reduced in intensity by using Image Inpainting techniques. Please refer to Inpainting Wiki Page to choose how to proceed with this step.

Note: The inpainting step can be skipped if the generated image doesn't exhibit any artifacts (white dots scattered across the image). This step is only meant to fix the artifacts generated by the model and serves no other functional purpose.

Resources

• Paper (ArXiv) The material is made available under the GNU General Public License v3.0: Copyright 2020, Emanuele Dalsasso, Loïc Denis, Florence Tupin, of LTCI research lab - Télécom ParisTech, an Institut Mines Télécom school. All rights reserved.

To cite the article:

@article{dalsasso2020sar2sar,
    title={{SAR2SAR}: a self-supervised despeckling algorithm for {SAR} images},
    author={Emanuele Dalsasso and Loïc Denis and Florence Tupin},
    journal={arXiv preprint arXiv:2006.15037},
    year={2020}
}