dzetsaka

is very fast and easy to use but also a powerful classification plugin for Qgis. Based on Gaussian Mixture Model classifier developped by Mathieu Fauvel, this plugin is a more generalist tool than Historical Map which was dedicated to classify forests from old maps. This plugin has by developped by Nicolaï Van Lennepkade (aka Nicolas Karasiak).

A quick tutorial is available online (dzetsaka : how to make your first classification in qgis ?), or you can just download samples to test the plugin on your own.

What does dzetsaka mean ?

As this tool was developped during my work in the Guiana Amazonian Park to classify different kind of vegetation, I gave an Teko name (a native-american language from a nation which lives in french Guiana) which represent the objects we use to see the world through, such as satellites, microscope, camera...

Discover the magic of dzetsaka

dzetsaka : Classification tool runs with scipy library. You can download package like Spider by Anaconda for a very easy setup.

Then, as this plugin is very simple, you will just need two things for making a good classification :

• A raster
• A shapefile which contains your ROI (Region Of Interest)

The shapefile must have a column which contains your classification numbers *(1,3,4...)*. Otherwise if you use text or anything else it certainly won't work.

Installation of scikit-learn

On Linux simply open terminal and type : pip install scikit-learn

On Windows, you have few more steps to do. Open Windows menu, and search for OsGeo Shell, then type :
curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py

After get-pip.py has been downloaded write :
python get-pip.py

Now use pip in OsGeo Shell like on Linux. Just type :
pip install scikit-learn

You can now use Random Forest, SVM, or KNN !

Tips

• If your raster is spot6scene.tif, you can create your mask under the name spot6scene_mask.tif and the script will detect it automatically.
• If you want to keep your spectral ROI model from an image, you can save your model to use it on another image.

Todo

• Implement best progress bar for classifier like Random Forest / K-Nearest Neighbors.

Thanks to...

I would like to thank the Guiana Amazonian Park for their trust in my work, and the Master 2 Geomatics Sigma for their excellent lessons in geomatics.