Feature Extraction for Hyperspectral Imagery: The Evolution from Shallow to Deep (Overview and Toolbox)
Behnood Rasti, Danfeng Hong, Renlong Hang, Pedram Ghamisi, Xudong Kang, Jocelyn Chanussot, Jon Atli Benediktsson
The code in this toolbox implements the "Feature Extraction for Hyperspectral Imagery: The Evolution from Shallow to Deep: Overview and Toolbox". More specifically, it is detailed as follow.
Please kindly cite the papers if this code is useful and helpful for your research.
B. Rasti, D. Hong, R. Hang, P. Ghamisi, X. Kang, J. Chanussot, J. Benediktsson. Feature Extraction for Hyperspectral Imagery: The Evolution from Shallow to Deep: Overview and Toolbox, IEEE Geosci. Remote Sens. Mag., 2020, 8(4): 60-88.
@article{rasti2020feature,
title = {Feature Extraction for Hyperspectral Imagery: The Evolution from Shallow to Deep: Overview and Toolbox},
author = {B. Rasti and D. Hong and R. Hang and P. Ghamisi and X. Kang and J. Chanussot and J. Benediktsson},
journal = {IEEE Geosci. Remote Sens. Mag.},
note = {DOI: 10.1109/MGRS.2020.2979764},
volume = {8},
number = {4},
pages = {60--88},
year = {2020},
publisher = {IEEE}
}
The paper provides a detailed and organized overview of hyperspectral feature extraction techniques, categorized into two general sections: shallow feature extraction techniques (further categorized into supervised and unsupervised) and deep feature extraction techniques. Each section provides a critical overview of the state-of-the-art that is mainly rooted in the signal and image processing, statistical inference, and machine (deep) learning fields. The toolbox also includes the Random Forest classifier plus training and test samples used for the Houston 2012 hyperspectral Dataset.
The hyperspectral data can be downloaded here (http://hyperspectral.ee.uh.edu/?page_id=459): Houston 2013 and (https://drive.google.com/file/d/1gN5yiPJ5PUZk67125WbS1jLdgQzBQbZk/view?usp=sharing): Houston 2018.
Moreover, the Indian Pine 2011 data can be also found in https://drive.google.com/file/d/1cWl6fzrx9doabyp-pYp13YoeK3bowY4A/view?usp=sharing.
The shallow and deep feature extraction techniques given in HyFTech is listed below:
Shallow Unsupervised Feature Extraction:
1- PCA: Principal Component Analysis
2- MSTV: Multi-scale Structural Total Variation
3- OTVCA: Orthogonal Total Variation Component Analysis
4- LPP: Locality Preserving Projection
Shallow Supervised Feature Extraction:
5- LDA: Linear Discriminant Analysis
6- CGDA: Collaborative Graph-based Discriminant Analysis
7- LSDR: Least-Squares Dimension Reduction
8- JPlay: Joint & Progressive Learning Strategy
Deep Feature Extraction:
9- SAE: Stacked Autoencoder
10- RNN: Recurrent Neural Network
11- CNN: Convolutional Neural Network
12- CAE: Convolutional Autoencoder
13- CRNN: Convolutional RNN
14- PCNN: PCA is applied prior to CNN
Danfeng Hong: hongdanfeng1989@gmail.com
Danfeng Hong is with the Univ. Grenoble Alpes, CNRS, Grenoble INP, GIPSA-lab, 38000 Grenoble, France.