GMM vs SVM

[Gmm]

Hi, 

I'm working on my thesis comparing differents classifiers (in my case vs GMM SVM) on satellite images and i want a recommendation (if possible).

What is the best way to compare GMM vs SVM images (Landsat), I'm following the workflow OBIA (1-smooth, 2-segmentation, 3-Merging, 4-vectorization and 5-classification (here I compare different classifiers )).

Cheers

[Jake Vanderplas]

Hello,

I'm not sure I fully understand your question – GMM is an unsupervised algorithm suitable for density estimation or clustering, while SVM is a supervised algorithm suitable for classification or regression. They have different goals, and thus their outputs are not directly comparable.

Hope that helps,

   Jake

--
You received this message because you are subscribed to the Google Groups "astroML-general" group.
To unsubscribe from this group and stop receiving emails from it, send an email to astroml-gener...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Gmm]

Hi professor Jake, thank you very much for the help,

First of all, your work (Blog, notebooks, videos) has helped me a lot.

The idea is to evaluate which one is better, if applied GMM to a vectorizacion, then do Dissolve and analyze or Mapping (by hand), then apply SVM, and Dissolve and compare. (Already, the problem is that I need an image with a class).

So far the results of GMM were very good; while SVM need additional work which coincidentally is what I evaluated (Mapping Class / Target): Worth the work of mapping at hand, Does this approach is more objective??.

I found some papers using both supervised and unsupervised methods to find specific features in images:

http://www.prasa.org/proceedings/2003/prasa03-02.pdf

https://www.researchgate.net/publication/260369359_An_object-based_workflow_for_integrating_spatial_scale_and_semantics_to_derive_landforms_from_digital_elevation_models_DEMs

I would like to do something different especially with GMM; and I'm open to suggestions. I know that the validation of landforms, wetlands, etc, is a complex task.

regards,