Target-adaptive CNN-based pansharpening is an advanced version of pansharpening method PNN with residual learning, different loss and a target-adaptive phase.

This the python version of the code, Go to Matlab version for Matlab.

Giuseppe Scarpa ([email protected]); Sergio Vitale (contact person, [email protected]); Davide Cozzolino ([email protected]).

Copyright (c) 2018 Image Processing Research Group of University Federico II of Naples ('GRIP-UNINA').

All rights reserved. This work should only be used for nonprofit purposes.

By downloading and/or using any of these files, you implicitly agree to all the terms of the license, as specified in the document LICENSE.txt (included in this directory)

This code is written for Python2.7 and uses Theano library. The list of all requirements is in requirements.txt.

The command to install the requirements is:

cat requirements.txt | xargs -n 1 -L 1 pip2 install

Optional requirements for using gpu:

• cuda = 8
• cudnn = 5

• set all paramaters in config_testing_<SENSOR>.xml:

  • sensor: GeoEye1, IKONOS, WV2
  • mode: full to high resolution pansharpening (PAN scale); reduce to low resolution pansharening (MS scale).
  • fine tuning: set epochs 0 to not fine tune; set epochs n to do n epochs of fine tuning.
  • paths: path of pretrained network; path of image to test; path of output image; path of fine tuned network.

• run test without gpu:

python PNN_testing.py -s <SENSOR>

• run test with gpu:

PATH=<CUDAPATH>:$PATH python PNN_testing.py -g -s <SENSOR>

• Output:
  • result: ./outputs/<SENSOR>/output.mat;
  • (eventual) fine-tuned network: ./outputs/<SENSOR>/FT_network/PNN_model.mat.

Where <SENSOR> can be: 'GE1','IK','WV2'; and <CUDAPATH> is your bin local path of cuda.