Hezhongchen-1998 / ASSMN

The code implementation of our paper "Adaptive Spectral-Spatial Multiscale Contextual Feature Extraction for Hyperspectral Image Classification" (IEEE TGRS, 2020).

Adaptive Spectral-Spatial Multiscale Network for Hyperspectral Image Classification

Pytorch implementation of our method for hyperspectral image classification.

Usage

1. Install Pytorch 1.1 with Python 3.5.

2. Clone this repo.

git clone https://github.com/DotWang/ASSMN.git

3. Training and evaluation with trainval.py.

   For example, for Indian Pines dataset, if SeMN and SaMN are all employed:

CUDA_VISIBLE_DEVICES=0 python trainval.py \
	--dataset 'indian' \
	--dr-num 4 --dr-method 'pca' \
	--mi -1 --ma 1 \
	--half-size 13 --rsz 27 \
	--experiment-num 10 \
	--lr 1e-2 --epochs 200 --batch-size 16 \
	--scheme 2 --strategy 's2' \
	--spec-time-steps 2 \
	--group 'alternate' --seq 'cascade' \
	--npi-num 2

    Then the assessment results are recorded in the corresponding *.mat file and the generated model is saved.

4. Predicting with the previous stored model through infer.py

CUDA_VISIBLE_DEVICES=0 python infer.py \
      --dataset 'indian' \
      --mi -1 --ma 1 \
      --half-size 13 --rsz 27 \
      --bz 50000 \
      --scheme 2 --strategy 's2' 

    and then produce the final classification map.

Paper and Citation

If this repo is useful for your research, please cite our paper.

@ARTICLE{2020ASSMN,
  author={D. {Wang} and B. {Du} and L. {Zhang} and Y. {Xu}},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Adaptive Spectral-Spatial Multiscale Contextual Feature Extraction for Hyperspectral Image Classification}, 
  year={2020},
  pages={1-17},
  doi={10.1109/TGRS.2020.2999957},
  url={https://doi.org/10.1109/TGRS.2020.2999957},
}

Acknowledgement

Thanks Andrea Palazzi for providing the Pytorch implementation of Convolutional LSTM!