Graph-Weighted Contrastive Learning for Semi-Supervised Hyperspectral Image Classification

• arXiv

Most existing graph-based semi-supervised hyperspectral image classification methods rely on superpixel partitioning techniques. However, they suffer from misclassification of certain pixels due to inaccuracies in superpixel boundaries, i.e., the initial inaccuracies in superpixel partitioning limit overall classification performance. In this paper, we propose a novel graph-weighted contrastive learning approach that avoids the use of superpixel partitioning and directly employs neural networks to learn hyperspectral image representation. Furthermore, while many approaches require all graph nodes to be available during training, our approach supports mini-batch training by processing only a subset of nodes at a time, reducing computational complexity and improving generalization to unseen nodes. Experimental results on three widely-used datasets demonstrate the effectiveness of the proposed approach compared to baselines relying on superpixel partitioning.

Experiment

Data

# Navigate to the DR_HSI directory
# cd DR_HSI
matlab
# Run ARM_a2b.m to obtain the datasets
# This will generate 'IndiaP_input.mat', 'Salinas_input', and 'PaviaU_input'
# Place the generated files into the ../HSI_GWCL/data directory

Train

# Navigate to the HSI_GWCL directory
# cd HSI_GWCL

# Train on the IndiaP dataset
python train.py --dataset 'IndiaP' --k 10 --lr_train 0.001 --lr_trans 0.001 --sigma1 0.04 --sigma2 0.001

# Alternatively, train on the PaviaU dataset
# python train.py --dataset 'PaviaU' --k 50 --lr_train 0.005 --lr_trans 0.01 --sigma1 1 --sigma2 0.4

# Or train on the Salinas dataset
# python train.py --dataset 'Salinas' --k 10 --lr_train 0.001 --lr_trans 0.001 --sigma1 0.04 --sigma2 0.04

Citation

We appreciate it if you cite the following paper:

@Article{YuqingJEI2025,
  author  = {Yuqing Zhang and Qi Han and Ligeng Wang and Kai Cheng and Bo Wang and Kun Zhan},
  journal = {Journal of Electronic Imaging},
  title   = {Graph-weighted contrastive learning for semi-supervised hyperspectral image classification},
  year    = {2025},
}

Contact

https://kunzhan.github.io/

For any inquiries, feel free to email me at ice.echo#gmail.com.