2023-JRS-DCLN-LQY

高光谱遥感影像分类

Deep Contrastive Learning Network for Small-Sample Hyperspectral Image Classification

The source of our paper: [https://spj.science.org/doi/epdf/10.34133/remotesensing.0025]

Description

The DCLN is method for small-sample HSI classification. It can realize effective spatial–spectral feature extraction, pseudo-label learning, and classification in the case of limited training samples.

##Model

Prerequisites

[Anaconda 3]
[Pytorch 1.7]
[CUDA 10.1]
[sklearn 0.23.2]

dataset

You can download the hyperspectral datasets in mat format at: http://www.ehu.eus/ccwintco/index.php/Hyperspectral_Remote_Sensing_Scenes, and move the files to ./data folder.

An example dataset folder has the following structure:

data
├── IP
│   ├── indian_pines_corrected.mat
│   ├── indian_pines_gt.mat
├── salinas
│   ├── salinas_corrected.mat
│   └── salinas_gt.mat
├── hou
│   ├── houston.mat
│   └── houston_gt.mat
└── paviaU
    ├── paviaU_gt.mat
    └── paviaU.mat

Usage:

Take DCLN method on the UP dataset as an example:

Download the required data set and move to folder./data.
Install the requirements : conda env create -f environment.yml.
Taking 5 labeled samples per class as an example, run train.py to train the model.
run test.py and get the results.

##Citation

lf you use DCLN code in your research, we would appreciate a citation to the original paper:

“Liu Q, Peng J, Zhang G, Sun W, Du Q. Deep Contrastive Learning Network for Small-Sample Hyperspectral Image Classification. J. Remote Sens. 2023;3:Article 0025. https://doi.org/10.34133/remotesensing.0025”