Application of the Jigsaw network to Hyperspectral Image (HSI) classification.

You need to donwload manually the datasets from: https://www.ehu.eus/ccwintco/index.php/Hyperspectral_Remote_Sensing_Scenes Copy the following files to your ./data directory:

• Indian Pines (Corrected Indian Pines and Indian Pines groundtruth)
• Pavia University (Pavia University and Pavia University groundtruth)
• Salinas scene (Corrected Salinas and Salinas groundtruth)

Create an Anaconda environment with Python 3.8 and the libraries from conda_requisites.txt

• You can run: "conda create --name <new_env> --file conda_requisites.txt", where <new_env> is the name of your new environment's name

The program makes use of GPU resources to train and use neural networks, therefore, you need a CUDA-compatible NVIDA GPU.

You need to run Jupyter in your local machine:

• Clone this repository to your computer
• Run anaconda and select your <new_env> environment
• Change directories to the cloned repository
• Run Jupyter in your local machine
• Follow the instructions to open Jupyter in your browser and connect to it
• Open the JigsawHSI.ipyb notebook and run the cells

There are two main ways to run variants of the JigsawHSI:

• Edit config.ini
• Select your preferred configuration from config.ini by changing the line "dataset = 'PU_100'"
• Edit the code to your heart's content!

You can read the early release of the paper in ArXiv or in this repository: JigsawHSI.pdf

Thanks to Gopal Krishna, for making the HyperSN code available in github (https://github.com/gokriznastic/HybridSN).