This repo contains the KERAS implementation of "AGSDNet: Attention and Gradient based SAR Denoising Network"
To test for SAR denoising using AGSDNet 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 AGSDNet 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.
To train the AGSDNet 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 AGSDNet model file for synthetic image denoising using:
python AGSDNet_Synthetic.py
This will save the 'AGSDNet_Synthetic.h5' file in the folder 'Pretrained_models/'.
Then run the AGSDNet model file for SAR image denoising using:
python AGSDNet_SAR.py
This will save the 'AGSDNet_SAR.h5' file in the folder 'Pretrained_models/'.
@article{thakur2022agsdnet, title={AGSDNet: Attention and Gradient-Based SAR Denoising Network}, author={Thakur, Ramesh Kumar and Maji, Suman Kumar}, journal={IEEE Geoscience and Remote Sensing Letters}, volume={19}, pages={1--5}, year={2022}, publisher={IEEE} }