Importance of Disjoint Sampling

This demo is associated with Unveiling the True Potential: Disjoint Sampling for Rigorous Evaluation for Land Cover Classification Submitted for possible publication.

@misc{ahmad2024importance, title={Unveiling the True Potential: Disjoint Sampling for Rigorous Evaluation for Land Cover Classification}, author={Muhammad Ahmad and Manuel Mazzara and Salvatore Distifano}, year={2024}, eprint={2404.14944}, archivePrefix={arXiv}, primaryClass={cs.CV} }

Models implemented in this work:

2D CNN

3D CNN

Hybrid CNN

2D Inception Net

3D Inception Net

Hybrid Inception Net

Attention Graph CNN

Spatial-Spectral Transformer

Requirements

This tool is compatible with Python 2.7 and Python 3.5+ and executed over Colab.

Hyperspectral datasets

Several public hyperspectral datasets are available on the EHU. Users can download those beforehand.

An example dataset folder has the following structure:

Datasets
├── University of Houston
│   ├── UH.mat
│   └── UG_gt.mat
├── IndianPines
│   ├── Indian_pines_corrected.mat
│   ├── Indian_pines_gt.mat
├── Pavia University
│   ├── PU.mat
│   └── PU_gt.mat
├── Botswana
│   ├── BS.mat
│   └── BS_gt.mat
├── Salinas
│   ├── SA.mat
│   └── SA_gt.mat