Discriminant Superpixel Graph (dSPG) for Hyperspectral Image Classification.

Matlab Implimentation of dSPG

Files:

demo_dSPG.m - Matlab demo over real AVIRIS Indian Pines dataset

Within_Superpixel_Graph.m - Matlab function for within-superpixel graph

Between_Superpixel_Graph.m - Matlab function for between-superpixel graph

Description

The dSPG is the combination of within-superpixel graph and between-superpixel graph.

The newly proposed within-superpixel graph is aimed at disconnecting pixels belonging to different classes in a superpixel, so as to decrease inter-class connection weights.

The between-superpixel graph attempts to connect spectral adjacent superpixels to increase the intra-class subset connections.

Results

Indian Pines (IP) dataset

Fig.1 The Indian Pines dataset classification result (Overall Accuracy 85.48%) of dSPG using 5 labeled samples per class. (a) False color composition. (b) Ground truth. (c) Classification map.

ZY1-02D-HongHu dataset

ZY1-02D-HongHu数据集下载链接

Fig.2 The ZY1-02D-HongHu dataset classification result (Overall Accuracy 99.06%) of dSPG using 5 labeled samples per class. (a) False color composition. (b) Ground truth. (c) Classification map.

Citation

If you use dSPG code in your research, we would appreciate your citation to the following paper:

@ARTICLE{yu2024dSPG,
  author={Yu, Long and Li, Jun and He, Lin and Plaza, Antonio and Wang, Lizhe and Tang, Zhonghui and Zhuo, Li and Yuan, Yuchen},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={dSPG: A New Discriminant Superpixel Graph Regularizer and Convolutional Network for Hyperspectral Image Classification}, 
  year={2024},
  volume={62},
  number={},
  pages={1-18},
  doi={10.1109/TGRS.2024.3439434}}