Spectralformer: Rethinking hyperspectral image classification with transformers

Danfeng Hong, Zhu Han, Jing Yao, Lianru Gao, Bing Zhang, Antonio Plaza, Jocelyn Chanussot

The code in this toolbox implements the "Spectralformer: Rethinking hyperspectral image classification with transformers". More specifically, it is detailed as follow.

Citation

Please kindly cite the papers if this code is useful and helpful for your research.

Danfeng Hong, Zhu Han, Jing Yao, Lianru Gao, Bing Zhang, Antonio Plaza, Jocelyn Chanussot. Spectralformer: Rethinking hyperspectral image classification with transformers. arXiv preprint, arXiv:2107.02988, 2021.

@article{hong2021spectralformer,
  title={Spectralformer: Rethinking hyperspectral image classification with transformers},
  author={Hong, Danfeng and Han, Zhu and Yao, Jing and Gao, Lianru and Zhang, Bing and Plaza, Antonio and Chanussot, Jocelyn},
  journal={arXiv preprint arXiv:2107.02988},
  year={2021}
}

System-specific notes

The data were generated by Matlab R2016a or higher versions, and the codes of networks were tested using PyTorch 1.6 version (CUDA 10.1) in Python 3.7 on Ubuntu system.

How to use it?

This toolbox consists of two proposed modules, i.e., group-wise spectral embedding (GSE: by setting band_patches larger than 1) and cross-layer adaptive fusion (CAF: by setting mode to CAF), that can be plug-and-played into both pixel-wise and patch-wise hyperspectral image classification. For more details, please refer to the paper.

Here an example experiment is given by using Indian Pines hyperspectral data. Directly run demo.py functions with different network parameter settings to produce the results. Please note that due to the randomness of the parameter initialization, the experimental results might have slightly different from those reported in the paper.

❗ You may need to manually download IndianPine.mat to your local in the folder under path Codes_SpectralFormer/data/, due to their too large file size, from the following links of google drive or baiduyun:

Google drive: https://drive.google.com/drive/folders/1nRphkwDZ74p-Al_O_X3feR24aRyEaJDY?usp=sharing

Baiduyun: https://pan.baidu.com/s/1rY9hj7Ku1Un4PPOjEFpEfQ (access code: 6dme)

If you want to run the code in your own data, you can accordingly change the input (e.g., data, labels) and tune the parameters.

If you encounter the bugs while using this code, please do not hesitate to contact us.

Licensing

Copyright (C) 2021 Danfeng Hong

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 3 of the License.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program.

Contact Information:

Danfeng Hong: [email protected]
Danfeng Hong is with the Univ. Grenoble Alpes, CNRS, Grenoble INP, GIPSA-lab, 38000 Grenoble, France.

If emergency, you can also add my QQ: 345088114.