jinnh / ReSSS-ConvSet

[TPAMI 2024] Deep Diversity-Enhanced Feature Representation of Hyperspectral Images

ReSSS-ConvSet

This is an implementation of Deep Diversity-Enhanced Feature Representation of Hyperspectral Images. [arXiv]

Requirement

• python 3.7, pytorch 1.7.0, and cuda 11.0

• Matlab

HS image denoising

Dataset

You can refer to the following links to download the dataset, ICVL. Following QRNN3D, we generated the noisy images for training and testing. You can run the matlab programs in the folder 'datasets' to get the pre-processed training and testing data.

Training the model for Gaussian noise

Enter the HSID folder and run

bash train_gaussian.sh

Training the model for Complex noise

Enter the HSID folder and run

bash train_complex.sh

Testing

python test.py --cuda --gpu "0" --dataset "ICVL" --noiseType "gaussian" --model_name "res3net" --checkpoint checkpoints/ICVL/res3net_gaussian_epoch_25.pth

python test.py --cuda --gpu "0" --dataset "ICVL" --noiseType "complex" --model_name "res3net" --checkpoint checkpoints/ICVL/res3net_complex_epoch_25.pth

HS image super-resolution

Dataset

You can refer to the following links to download the dataset, CAVE. And run the matlab programs in the folder 'datasets' to get the pre-processed training and testing data.

Training

Enter the HSISR folder and run

bash train.sh

Testing

python test.py --cuda --gpu "0" --dataset "CAVE" --model_name "res3net" --upscale_factor 4 --checkpoint checkpoints/CAVE_x4/res3net_4_epoch_50.pth'

HS image classification

This codebase borrows from Spectralformer and 3D-CNN.

Dataset

You can refer to the following link to download the datasets, IndianPine and Pavia.

Training

Enter the HSIC folder and run

python main.py --dataset="Indian" --method="res3net" --epoch=1000 --patches=7 --weight_decay=1e-2 --learning_rate=1e-3 --gpu_id=0 --loss_weight=1e-4

python main.py --dataset="Pavia" --method="res3net" --epoch=160 --patches=7 --weight_decay=1e-3 --learning_rate=1e-3 --gpu_id=0 --loss_weight=3e-4

Testing

python main.py --flag_test="test" --dataset="Indian" --method="res3net" --model_name="checkpoints/res3net_Indian/res3net_best.pt" --patches=7

python main.py --flag_test="test" --dataset="Pavia" --method="res3net" --model_name="checkpoints/res3net_Pavia/res3net_best.pt" --patches=7

Citation

Please kindly cite our work if you find it helpful.

@article{hou23deep,
	title={Deep Diversity-Enhanced Feature Representation of Hyperspectral Images},
	author={Hou, Jinhui and Zhu, Zhiyu and Hou, Junhui and Liu, Hui and Zeng, Huanqiang and Meng, Deyu},
  	journal={arXiv preprint arXiv:2301.06132}
	year={2023}
}