HyperMamba: A Spectral-Spatial Adaptive Mamba for Hyperspectral Image Classification

Qiang Liu, Jun Yue, Yi Fang, Shaobo Xia, and Leyuan Fang, Senior Member, IEEE

Getting Started

step1: Environment Setup:

To get started, we recommend setting up a conda environment and installing dependencies via pip. Use the following commands to set up your environment

conda create -n hypermamba python==3.10
conda activate hypermamba
pip install -r requirements.txt

install mmcv

pip install -U openmim
mim install mmcv==2.1.0

download vmamba dependencies at https://github.com/MzeroMiko/VMamba/archive/refs/tags/%2320240220.tar.gz

unzip and run:

# Install selective_scan and its dependencies
cd selective_scan && pip install .

step2: Model Training and Inference:

Our work is evaluated on three pulic hyperspectral dataset

To train Hypermamba for classification on those datasets, you should changeinclude_path for different dataset in code fileworkflow.py use the following commands for model training.

python workflow.py

the reults are saved in res folder and are saved at ckpt folder

Moreover

if you want to change the data path or model settings, please go to params_use folder.

Citation

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

Q. Liu, J. Yue, Y. Fang, S. Xia and L. Fang, "HyperMamba: A Spectral-Spatial Adaptive Mamba for Hyperspectral Image Classification," in IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1-14, 2024, Art no. 5536514, doi: 10.1109/TGRS.2024.3482473.

@ARTICLE{10720896,
  author={Liu, Qiang and Yue, Jun and Fang, Yi and Xia, Shaobo and Fang, Leyuan},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={HyperMamba: A Spectral-Spatial Adaptive Mamba for Hyperspectral Image Classification}, 
  year={2024},
  volume={62},
  number={},
  pages={1-14},
  keywords={Computational modeling;Adaptation models;Transformers;Training;Feature extraction;Accuracy;Quaternions;Context modeling;Hyperspectral imaging;Convolutional neural networks;Deep neural network;hyperspectral image (HSI) classification;Mamba},
  doi={10.1109/TGRS.2024.3482473}}

Acknowledgment

This code is mainly built upon SQSFormer and VMamba repositories.