Awesome

deepdespeckling

Speckle noise seriously limits the interpretability of synthetic aperture radar (SAR) images. This package provides <strong>deep learning based despeckling methods</strong> to denoise SAR images.

<strong>deepdespeckling</strong> gives access to two different despeckling methods:

• MERLIN
• SAR2SAR

Installation

Before installing deepdespeckling, make sure to install gdal dependancies, it can be done using conda with the following command :

conda install -c conda-forge gdal

Then, deepdespeckling can be installed simply using pip :

pip install deepdespeckling

Inference with MERLIN

To despeckle SAR images using MERLIN, images need to be in .cos or .npy format.

In order to get the right model, the model_name has to be specified when building a MerlinDenoiser.

This model_name can either be :

• "spotlight" for SAR images retrieved with spotlight mode
• "stripmap" for SAR images retrieved with stripmap mode
• "Sentinel-TOPS" for SAR images retrieved with TOPS mode

During the preprocessing steps of the noisy image for MERLIN, the real and the imaginary parts are <strong>"symetrised"</strong> (to match the theoretical assumptions of MERLIN). To skip this step, you can set the symetrise parameter to False

Despeckle one image with MERLIN

from deepdespeckling.utils.load_cosar import cos2mat
from deepdespeckling.utils.constants import PATCH_SIZE, STRIDE_SIZE
from deepdespeckling.merlin.merlin_denoiser import MerlinDenoiser

# Path to one image (cos or npy file)
image_path="path/to/image"
# Model name, can be "spotlight", "stripmap" or "Sentinel-TOPS"
model_name = "spotlight"
symetrise = True

image = cos2mat(image_path).astype(np.float32)
# As an alternative, you can load your image by being sure to put in the following format:
# The shape of image is [img_height,img_width,2], where image[:,:,0] contains the real part and image[:,:,1] contains the imaginary part of the SLC SAR image.

denoiser = MerlinDenoiser(model_name=model_name, symetrise=symetrise)
denoised_image = denoiser.denoise_image(image, patch_size=PATCH_SIZE, stride_size=STRIDE_SIZE)

Despeckle a set of images using MERLIN

For each of this method, you can choose between 3 different functions to despeckle a set of SAR images contained in a folder :

• despeckle() to despeckle full size images
• despeckle_from_coordinates() to despeckle a sub-part of the images defined by some coordinates
• despeckle_from_crop() to despeckle a sub-part of the images defined using a crop tool

Despeckle fullsize images

from deepdespeckling.despeckling import despeckle

# Path to a folder of several images (cos or npy files)
image_path="path/to/image"
# Folder where results are stored
destination_directory="path/where/to/save/results"

despeckle(image_path, destination_directory, model_name="spotlight", symetrise=True)

Noisy image	Denoised image

Despeckle parts of images using custom coordinates

from deepdespeckling.despeckling import despeckle_from_coordinates

# Path to a folder of several images image (cos or npy files)
image_path="path/to/cosar/image"
# Folder where results are stored
destination_directory="path/where/to/save/results"
# Coordinates of the subparts of the images to be despeckled
coordinates_dictionnary = {'x_start':2600,'y_start':1000,'x_end':3000,'y_end':1200}

despeckle_from_coordinates(image_path, coordinates_dict, destination_directory, model_name="spotlight", symetrise=True)

Noisy image	Denoised image

Despeckle parts of images using a crop tool

from deepdespeckling.merlin.inference.despeckling import despeckle_from_crop

# Path to a folder of several images image (cos or npy files)
image_path="path/to/cosar/image"
# Folder where results are stored
destination_directory="path/where/to/save/results"
fixed = True "(it will crop a 256*256 image from the position of your click)" or False "(you will draw free-handly the area of your interest)"

despeckle_from_crop(image_path, destination_directory, model_name="spotlight", fixed=False, symetrise=True)

• The cropping tool: Just select an area and press "q" when you are satisfied with the crop !

• The results:

Inference with SAR2SAR for Sentinel-1 Ground Range Detected (GRD) images

To despeckle Sentinel-1 GRD SAR images using SAR2SAR, images need to be in .tiff or .npy format.

Despeckle one image with SAR2SAR

from deepdespeckling.utils.load_cosar import cos2mat
from deepdespeckling.utils.constants import PATCH_SIZE, STRIDE_SIZE
from deepdespeckling.sar2sar.sar2sar_denoiser import Sar2SarDenoiser

# Load you image
image = ..
# NB: image must be in AMPLITUDE format and have the following shape: [img_height,img_width]

# Denoise the image with SAR2SAR
denoiser = Sar2SarDenoiser()
denoised_image = denoiser.denoise_image(
                image, patch_size=PATCH_SIZE, stride_size=STRIDE_SIZE)

• Example of result with SAR2SAR :

Noisy image	Denoised image

Despeckle a set of images using SAR2SAR

The despeckling functions (despeckle, despeckle_from_coordinates, despeckle_from_crop) work the same as with MERLIN. To use SAR2SAR, the model_name parameter has to be set to "sar2sar"

For example, to despeckle a set of fullsize images:

from deepdespeckling.despeckling import despeckle

# Path to a folder of several images (tiff or npy files)
image_path="path/to/image"
# Folder where results are stored
destination_directory="path/where/to/save/results"

despeckle(image_path, destination_directory, model_name="sar2sar")

Documentation

The documentation of the package is available here

Authors

• Emanuele Dalsasso (Researcher at ECEO, EPFL)
• Hadrien Mariaccia (Hi! PARIS Research Engineer)

Former contributors

• Youcef Kemiche (Former Hi! PARIS Research Engineer)
• Pierre Blanchard (Former Hi! PARIS Research Engineer)

References

[1] DALSASSO, Emanuele, DENIS, Loïc, et TUPIN, Florence. As if by magic: self-supervised training of deep despeckling networks with MERLIN. IEEE Transactions on Geoscience and Remote Sensing, 2021, vol. 60, p. 1-13.

[2] DALSASSO, Emanuele, DENIS, Loïc, et TUPIN, Florence. SAR2SAR: a semi-supervised despeckling algorithm for SAR images. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (Early Access), 2020