Fisher Vectors for PolSAR Image Classification

Code used in paper:

Javier Redolfi, Jorge Sánchez, and Ana Georgina Flesia
"Fisher Vectors for PolSAR Image Classification"
IEEE Geoscience and Remote Sensing Letters, 99, Septiembre 2017. (link)

Features

The library consist basically of the following python modules:

• PolSAR image manipulation: pseudocolor image generation, force SPD.
• PolSAR datasets: San Francisco Bay, Flevoland and Foulum. Train/test splits generation, manual selection of training points. Accuracy and Confusion Matrix computation.
• Smoothing algorithms: Potts, Graph Cut and Maximun.
• Unary potential generation from Complex Wishart, Real Wishart Mixture and FV (RWM based).

Dependencies

• vrl -> http://www.famaf.unc.edu.ar/~jsanchez/efv/

• matplotlib
• numpy

• pygco -> http://vision.csd.uwo.ca/code/gco-v3.0.zip

• pystruct -> https://pystruct.github.io/

• scikit-image -> http://scikit-image.org/

• scikit-learn -> scikit-learn.org/

• scipy

Installation on Debian/Ubuntu

(use sudo for Ubuntu)

> \# apt-get install python2.7 python-matplotlib python-numpy python-skimage python-sklearn python-scipy

• Install vrl using the README provided by the library.

> \# pip install pystruct

• Install pygco using the gco_python/README.md provided by the library.

Optionally, you might also need virtualenv:

> \# apt-get install python-virtualenv

Installing the module

• Just uncompress it.

Example: Comparison with paper "Pol-SAR Classification Based on Generalized Polar Decomposition of Mueller Matrix"

• Download the dataset and uncompress it: -> https://www.dropbox.com/s/xulk8t5pizoceh5/dataset.tar.gz?dl=0

• Run the script "src/bin/comparison_mueller.py --dataset=path_to_dataset"