Hi Julien,
Thanks for your answer. Here is the steps we used and possible problems
based on your answer and some problems identified.
*step 1*- otbcli_Segmentation or otbcli_LSMSSegmentation to produce a
labeled (id region polygon) raster segmentation
*step 2*- Build a concatenate images with otbgui_ConcatenateImages, with
desired N attributes (spectral channels, texture images from otb
applictations)
pb2.1 : there used to be an otb command to produce shape attributes and
statistics attributes based on otb filters (OBIA in software
documentation). I think this has not been put into an application. It could
be nice for object based to have these features such as the otb app for
texture
*step 3-* otbgui_ComputeImagesStatistics to produce an .xml with
mean,variance for the N channels of the Concatenate Images
*step 4- *otbgui_LSMSVectorization to produce a shapefile with 2*N features
(mean,variance) + id + nbpixels per region
*step 5- *Add a class feature to this shapefile and manually give a class
for training
pb5.1 : Ok with your comment i don't know any method to produce
automatically these attributes with vector data that is not consistent with
the segmentation but i am not familiar with vector analysis (join
attributes by location, qgis ?)
*step 6- *otbcli_TrainVectorClassifier to produce a model
pb6.1 : It is impossible to keep the variance features of the shapefile
for the model estimation since the statistics per band from
otbcli_ComputeImagesStatistics only give stats for the means. (N bands)...
difficult to explain ;-)
*step 7- *otbgui_OGRLayerClassifier that takes as input the previous
shapefile and the image statistics and add a new "predicted" attributes in
the shapefile.
pb7.1 : The OGRLayerClassifier only seem to accept SVM models. It
worked for a SVM classifier but failed for a Random Forest with error
messages related to itk SVM .... need to develop an otbapp VectorClassifier
?
(I did the tests on OTB 5.6.1, i'll have to to this test again with the 5.8
version)
Sébastien Giordano / Cécile Cazals
Le mardi 24 janvier 2017 16:07:04 UTC+1, Julien Michel a écrit :
>
> Hi,
>
> Please find my answers enclosed. Bottom line is that we have some
> components but we are still missing other ones to perform a complete
> OBIA scheme. This thread could be an opportunity to start a discussion
> on what is missing. We could then write a Request For Comments about that.
>
> Regards,
>
> Julien
>
> Le 24/01/2017 à 15:46, Sébastien G. a écrit :
> > Hi,
> >
> > Since recent OTB versions there is a very nice new framework for
> > classifictaion (PolygonStatistics, SampleSelection,
> > SampleExtraction,TrainVectorClassifier) that allow to train model on
> > vector data.
> >
> > Can anyone tell me an OTB recipee to use these new tools for
> > object-based classification, including the following steps :
> > 1- Derive features from a raster images on a segmentation (mean,
> > standard deviation of radiometric channels, or shape features)
> There is this step from the Large Scale MeanShift segmentation workflow :
>
>
> https://www.orfeo-toolbox.org/CookBook/recipes/improc.html#step-4-vectorization
>
>
> Will work with any pair of segmentation/input image as long as they
> share the same sampling grid (i.e. same size/orgin/spacing).
>
> There are also similar things in this remote module :
>
> https://github.com/sertit/SertitObject
>
> > 2- Train a vectorClassifier knowing a class attributes on a sample of
> > the (segmentation with features).
> This is covered by TrainVectorClassifier application, which can learn
> from points but also from polylines or polygons. The problem is to get
> the labels to train the polygons. Aside from manual editing, we are
> missing some tools to perform automatic labelling of polygons from a
> reference layer.
>
> > 3- Apply the model on the vector data (segmentation with features)
>
> This is covered by OGRLayerClassifier application;
> >
> >
> > Best regards,
> > Sébastien
> > --
> > --
> > Check the OTB FAQ at
> > http://www.orfeo-toolbox.org/FAQ.html
> >
> > You received this message because you are subscribed to the Google
> > Groups "otb-users" group.
> > To post to this group, send email to [email protected]
> <javascript:>
> > To unsubscribe from this group, send email to
> > [email protected] <javascript:>
> > For more options, visit this group at
> > http://groups.google.com/group/otb-users?hl=en
> > ---
> > You received this message because you are subscribed to the Google
> > Groups "otb-users" group.
> > To unsubscribe from this group and stop receiving emails from it, send
> > an email to [email protected] <javascript:>
> > <mailto:[email protected] <javascript:>>.
> > For more options, visit https://groups.google.com/d/optout.
>
>
> --
> Julien MICHEL
> CNES - DSO/SI/2A
>
>
--
--
Check the OTB FAQ at
http://www.orfeo-toolbox.org/FAQ.html
You received this message because you are subscribed to the Google
Groups "otb-users" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/otb-users?hl=en
---
You received this message because you are subscribed to the Google Groups
"otb-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
