S2DL: A Superpixel-based and Spatially-regularized Diffusion Learning Method for Unsupervised Hyperspectral Image Clustering

This code is an implementation of Superpixel-based and Spatiallyregularized Diffusion Learning proposed in "Superpixel-based and Spatially-regularized Diffusion Learning Method for Unsupervised Hyperspectral Image Clustering", see Link. S2DL can be used as a clustering method for remote sensing datasets.

S2DL uses several Matlab Toolboxes, such as Entropy Rate Superpixel, and Diffusion Learning.

• Contact: kangnicui2@gmail.com

If you find it useful or use it in any publications, please cite the following papers:

References

Cui, K., Li, R., Polk, S.L., Lin, Y., Zhang, H., Murphy, J.M., & Plemmons, R. J., Chan, R. H.. "Superpixel-based and Spatially-regularized Diffusion Learning Method for Unsupervised Hyperspectral Image Clustering". in Transactions on Geoscience and Remote Sensing, 62, 1-18, IEEE, 2024. Link.

Polk, S.L., Cui, K., Chan, A. H., Coomes, D. A., Plemmons, R. J., & Murphy, J.M.. "Unsupervised Diffusion and Volume Maximization-Based Clustering of Hyperspectral Images". in Remote Sensing, 15(4), 1053, MDPI, 2023. Link.

Using Benchmark Datasets with S2DL

To facilitate the use of benchmark datasets with the S2DL framework, follow these simple steps. This will ensure that you can seamlessly run S2DL with widely recognized datasets for unsupervised hyperspectral image clustering.

Step 1: Downloading the Data

1. Visit the following URL to access a collection of benchmark hyperspectral datasets: RS Lab Data Repository.
2. Select the dataset you wish to use with the S2DL framework. Download options for various datasets are provided on the website.

Step 2: Preparing the Dataset

1. Once downloaded, save the dataset into a folder within the working directory of MATLAB. This is crucial for the S2DL code to access and process the dataset correctly.
   • For example, if your MATLAB working directory is C:\MATLAB\Projects\S2DL, you could create a new folder within it named Datasets and save your downloaded dataset there, resulting in a path like C:\MATLAB\Projects\S2DL\Datasets.

Step 3: Running the Framework

1. Open MATLAB and navigate to the S2DL project's working directory.
2. Launch the S2DL_main.m script. The script will automatically detect datasets placed within the appropriate directory and proceed with the unsupervised hyperspectral image clustering process. The best parameters for datasets listed in the paper are located at backEnd/Misc/Preprocessing/loadHyperparameters.m.

By following these steps, you'll be able to utilize benchmark datasets, such as Indian Pines, Salinas and Salinas A, to evaluate the performance of the S2DL framework effectively.

Using Custom Datasets with S2DL

To adapt the S2DL codebase for your own hyperspectral dataset, follow these steps:

Step 1: Preparing Your Data

1. Access the data loading script at: loadHSI.m.

2. Add your dataset by inserting the following code within the if-elseif chain:

   elseif strcmp(HSIName, 'Your_Data_Name')
       HSI = % Load your Hyperspectral data here, with shape m*n*p, where:
             % m and n are the spatial dimensions, and p denotes the number of spectral bands.
       GT = % Load your ground truth for the Hyperspectral data here, with shape m*n.
   

   Replace 'Your_Data_Name' with your dataset's identifier and provide the appropriate loading commands for HSI and GT.

Step 2: Modifying the Main Script

1. Modify S2DL_main.m to include your dataset in the selection prompt. Update lines 9-17 as follows:

   prompt = 'Which dataset? \n 1) Indian Pines (Corrected) \n 2) Salinas (Corrected) \n 3) Salinas A (Corrected) \n 4) Your Data \n';
   DataSelected = input(prompt);
   if DataSelected > 4 || DataSelected < 1
       disp('Incorrect prompt input. Please enter a valid number [1-4].')
   end
   
   datasets = {'IndianPines', 'Salinas', 'SalinasA', 'Your_Data_Name'};
   

   Ensure 'Your_Data_Name' matches the identifier used previously.

Step 3: Running Your Dataset

Now, you're set to run S2DL with your dataset. Simply choose your data when prompted by S2DL_main.m.

By following these steps, you'll be able to integrate and utilize your own hyperspectral dataset within the S2DL framework, expanding its application to new and unique datasets.