Spectral Partitioning Residual Network with Spatial Attention Mechanism for Hyperspectral Image Classification

This repository is the implementation of our paper: Spectral Partitioning Residual Network with Spatial Attention Mechanism for Hyperspectral Image Classification.

If you find this work helpful, please cite our paper:

@ARTICLE{9454961,  
author={Zhang, Xiangrong and Shang, Shouwang and Tang, Xu and Feng, Jie and Jiao, Licheng},  
journal={IEEE Transactions on Geoscience and Remote Sensing},   
title={Spectral Partitioning Residual Network With Spatial Attention Mechanism for Hyperspectral Image Classification},   
year={2021},  
volume={},  number={},  
pages={1-14},  
doi={10.1109/TGRS.2021.3074196}}

Requirements

Only Python3 is supported. We recommend you to create a Python virtual environment and then run the following command to install dependencies.

pip install -r requirement.txt

CUDA and cuDNN are optional

Datasets

You can download hyperspectral image datasets at http://www.ehu.eus/ccwintco/index.php?title=Hyperspectral_Remote_Sensing_Scenes, and move the files to ./Datasets folder

Usage

To train a model, simply run main.py, for example:

python main.py --dataset PaviaU --model HPDM-SPRN --runs 10  --patch_size 7 --percentage 0.01 --data_aug

To get colored results, run eval.py. The colored results can be found in the results folder. For example:

python eval.py --dataset PaviaU --model HPDM-SPRN --patch_size 7 --weights (saved model path)

Models

DFFN (HPDM-DFFN)
DHCNet (HPDM-DHCNet)
BASSNet (HPDM-BASSNet)
SSRN (HPDM-SSRN)
FDSSC (HPDM-FDSSC)
SPRN (HPDM-SPRN)

Acknowledgement