[GRASS-user] Q&A on image classification, segmentation, groups and subgroups

Sophie Crommelinck sophie.crommelinck at gmail.com
Thu Sep 1 02:07:02 PDT 2016


Hello,

I came across some questions (and answers) while working on image analysis
with GRASS, which might be interesting to other users. I will post the
questions and answers (partly from Moritz Lennert and from me) here, so
others might profit from them as well.

*Q1:* How can I run *v.class.mlR* as it is not a recognized module in GRASS
7.0.4? How does this module differ from the SVM implementation in Orfeo?

*A1:* *v.class.mlR* is an extension that needs to be installed with
*g.extension
v.class.mlR*. The SVM classification in Orfeo is pixel-based, while the one
in GRASS is object-based, which is also indicated by the name (v.class).

*Q2:* What is the purpose and difference of groups and subgroups in GRASS?
They are often required and if I export an image containing 3 bands to a
group (i.group), it contains 5 bands.

*A2:* It is not possible to "export" to a group. You define a group as a
collection of maps. And subgroups are subcollections within a collection.
In the traditional workflow, you would get imagery that is not
georeferenced. You could then georeference all bands of a scene by putting
them into one group, identifying GCPs on one band and launching the
georeferencing for the entire group at once. After that, you could proceed
to specific analysis using only some of the bands by grouping them into
subgroups. This is the logic you find in the i.gensig/i.maxlik workflow.
More generally said, groups thus allow you to treat some maps collectively,
and subgroups allow you to treat only a subselection. However, some
developers are questioning the need and utility of subgroups. This is why,
for example, in i.segment there is only a 'group' parameter, but not a
subgroup.

When importing a raster, GRASS creates one raster for each raster band.
GRASS treats each band as a separate raster map. Each band can be
visualized separately, or, if desired, a composite can be created. To
perform an image classification, the raster maps (bands) must be combined
in to a group and subgroup. A group is a collection of raster maps. A
subgroup is a subset of the group’s raster maps.

*Q3:* Is there a possibility to apply Simple Linear Iterative Clustering
(SLIC) image segmentation in GRASS?

*A3:* Currently not, but let me know, if anyone is planning to change this
=)



Best,

Sophie Crommelinck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/grass-user/attachments/20160901/b3a79a76/attachment-0001.html>


More information about the grass-user mailing list