mhaut / hsi-sdecnn

Source code of "A Single Model CNN for Hyperspectral Image Denoising"

A Single Model CNN for Hyperspectral Image Denoising

The Code for "A Single Model CNN for Hyperspectral Image Denoising". [https://ieeexplore.ieee.org/abstract/document/8913713]

Alessandro Maffei, Juan M. Haut, Mercedes E. Paoletti, Javier Plaza, Lorenzo Bruzone and Antonio Plaza.
A Single Model CNN for Hyperspectral Image Denoising.
IEEE Transactions on Geoscience and Remote Sensing.
DOI: 10.1109/TGRS.2019.2952062
vol. 58, no. 4, pp. 2516-2529, April 2020.

Example of use

Dataset

Go to trainset folder and run "split_file.py"

Some tips

In the "trainset" folder are reported the 4 part of the Washigton DC mall HSI used for the training of the network, while in the "testsets" directory the HSI on which the method has been tested.

To train the network run "Demo_training_HSI_SDeCNN.mat". The best model will be saved in the "data" folder.

To apply the denoising process with the model trained in the paper (the one in the "BestModel" directory) run the demo-test scripts. The results will be saved in the "Results" folder.

NB! if you want to change the number of considered channel in the training process you need to change the "nch" variable in the "model_train.mat" and in the "model_init_HSI_SDeCNN.mat" scripts.

Reference code

https://github.com/cszn/FFDNet

hsi-sdecnn's Issues

Washington DC Mall test image

Hello,
Thank you for sharing your code, very nice project ! I have a question regarding the Washington DC Mall dataset : how did you obtain the test image from the original one ? I understand that you cropped the original one selecting x=[50:250] and y=[600:800], but did you apply any kind of normalization / preprocessing / clipping ?
Thank you in advance :)