HyperSL

Official implementation of "HyperSL: A Spectral Foundation Model for Hyperspectral Image Interpretation".

🔗 Pretrained Weights

You can download the pretrained weights from Baidu Netdisk:

• Link: https://pan.baidu.com/s/11uuzhKs-dtFExlnph1IYTQ
• Extraction Code: 27mh

You can download the training dataset from Baidu Netdisk:

• Link: https://pan.baidu.com/s/1YP_OMkAf4LytjyowQBq9JQ
• Extraction Code: tw7g

You can also accsess the training dataset and pretrained weights from Huggingface:

• Link: https://huggingface.co/WeilKon/HyperSL/tree/main

📖 Citation

If you find our work useful in your research, please consider citing:

@ARTICLE{10981753,
  author={Kong, Weili and Liu, Baisen and Bi, Xiaojun and Yu, Changdong and Li, Xinyao and Chen, Yushi},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={HyperSL: A Spectral Foundation Model for Hyperspectral Image Interpretation}, 
  year={2025},
  volume={63},
  number={5513119},
  pages={1-19},
  keywords={Hyperspectral imaging;Data models;Sensors;Adaptation models;Training;Representation learning;Data mining;Autoencoders;Vectors;Transformers;Cross-scenario;foundation model;hyperspectral image;knowledge transfer},
  doi={10.1109/TGRS.2025.3566205}}

📎 Overview

HyperSL is designed to:

• Learn transferable spectral representations across heterogeneous hyperspectral sensors
• Accept arbitrary spectral input dimensions (bands, range)
• Generalize without retraining and modifying network structure

🛠️ Features

• Masked self-supervised learning strategy tailored for spectral data
• Unified spectral token representation
• Wavelength-aware positional encoding
• Trained on over 300 million spectra from multiple sensors