SIFSDNet

This repo contains the KERAS implementation of "SIFSDNET: SHARP IMAGE FEATURE BASED SAR DENOISING NETWORK"

Run Experiments

To test for SAR denoising using SIFSDNet write:

python Test_SAR.py

The resultant images will be stored in 'Test_Results/SAR/'

Image wise ENL for the whole image database will also be displayed in the console as output.

To test for synthetic denoising using SIFSDNet write:

python Test_Synthetic.py

The resultant images will be stored in 'Test_Results/Synthetic/'

Image wise PSNR & SSIM as well as Average PSNR & Average SSIM for the whole image database will also be displayed in the console as output.

Train SIFSDNet denoising network

To train the SIFSDNet denoising network, first download the UC Merced Land Use data and copy the images into genData folder. Then generate the training data using:

python generateData.py

This will save the training patch 'img_clean_pats.npy' in the folder 'trainingPatch/'

Then run the SIFSDNet model file for synthetic image denoising using:

python SIFSDNet_Synthetic.py

This will save the 'SIFSDNet_Synthetic.h5' file in the folder 'Pretrained_models/'.

Then run the SIFSDNet model file for SAR image denoising using:

python SIFSDNet_SAR.py

This will save the 'SIFSDNet_SAR.h5' file in the folder 'Pretrained_models/'.

Citation

@inproceedings{thakur2022sifsdnet, title={Sifsdnet: Sharp image feature based sar denoising network}, author={Thakur, Ramesh Kumar and Maji, Suman Kumar}, booktitle={IGARSS 2022-2022 IEEE International Geoscience and Remote Sensing Symposium}, pages={3428--3431}, year={2022}, organization={IEEE} }