tulilin / S2HM2

S2HM2: A Spectral-Spatial Hierarchical Masked Modeling Framework for Self-Supervised Feature Learning and Classification of Large-Scale Hyperspectral Images

S²HM²: A Spectral-Spatial Hierarchical Masked Modeling Framework for Self-Supervised Feature Learning and Classification of Large-Scale Hyperspectral Images

The Pytorch implementation of the paper "S²HM²: A Spectral-Spatial Hierarchical Masked Modeling Framework for Self-Supervised Feature Learning and Classification of Large-Scale Hyperspectral Images" [Paper Link].

WHU-OHS dataset

The WHU-OHS large-scale hyperspectral dataset is used for the network pre-training and fine-tuning. Download the dataset and crop the images into sub-images with the size of 256*256 as network inputs.

Dataset download: http://irsip.whu.edu.cn/resources/WHU_OHS_show.php

Pytorch toolbox for WHU-OHS dataset: https://github.com/zjjerica/WHU-OHS-Pytorch

Paper: https://www.sciencedirect.com/science/article/pii/S1569843222002102

Example of usage

Step 1: Run pretraining.py to pre-train the proposed S²HM² framework in a self-supervised manner, using the images in the source domain of WHU-OHS dataset.

Step 2: Run finetuning.py to fine-tuning the pre-trained network using the images and labels in the target domain of WHU-OHS dataset.

Step 3: Run testing.py or predicting.py to evaluate the performance of S²HM² quantitatively or qualitatively.

Paper

If our work is useful to you, please cite our following papers:

[1] L. Tu et al., “S2HM2: A spectral-spatial hierarchical masked modeling framework for self-supervised feature learning and classification of large-scale hyperspectral images,” IEEE Trans. Geosci. Remote Sens., 2024, doi: 10.1109/TGRS.2024.3392962.

[2] J. Li, X. Huang, and L. Tu, “WHU-OHS: A benchmark dataset for large-scale hersepctral image classification,” Int. J. Appl. Earth Obs. Geoinf., vol. 113, 2022, doi: 10.1016/j.jag.2022.103022.