Official implementation of DCSN

[Paper] [Project Website]

DCSN: Deep Compressed Sensing Network for Efficient Hyperspectral Data Transmission of Miniaturized Satellite

Fig.1. Visualized results of the reconstructed hyperspectral image by the proposed DCSN under 1% compression ratio.

Prerequisites

Create a conda environment for DCSN. Tested under Python 3.7 and CUDA 10.0 under Ubuntu 16.10/18.04. Windows OS does not test yet.

conda create -n dcsn python=3.* -y
conda activate dcsn

Install pytorch.

conda install pytorch torchvision -c pytorch
pip install opencv-python

and install all dependencies

pip install -r requirements.txt

Data prepration

To generate datasets, please read README.md in folder 'data_preprocessing/'. Matlab is required in generating the dataset.

Preparing a file list for training and testing samples like train.txt and valiation.txt for training and inference produces (see example in train_4fig.txt).

Training

Make sure you have right setting for the hyper-parameters in the train_sr.py, then

python train_sr.py

Testing

Make sure you have indicated a correct checkpoint and setting in the testing.py, then

python testing.py

Pretarined model for 1% compression

The checkpoint can be found here please put the pth file to ckpt directory.

Citation

If you find our work useful in your research or publication, please cite our work:

@ARTICLE{dcsn_cchsu,
  author={C. -C. {Hsu} and C. -H. {Lin} and C. -H. {Kao} and Y. -C. {Lin}},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={{DCSN}: Deep Compressed Sensing Network for Efficient Hyperspectral Data Transmission of Miniaturized Satellite}, 
  year={2020},
  volume={},
  number={},
  pages={1-17},
  doi={10.1109/TGRS.2020.3034414}}