2013/11/27 Nikos Alexandris <nik@nikosalexandris.net>
Cesar Augusto Ramírez Franco wrote:
> > > I've been trying to do a supervised classification from landsat8
imagery
> > > using i.class and then i.maxlik, but I'm getting an error when
running
> > > i.maxlik saying it can't read the signature file.
Nikos Alexandris:
> > Just to double-check: verify that your "classes" don't overlay any NULL
> > cells. It might be completely irrelevant, I have, though, blurry
memories
> > of this being a blocker actually (with i.smap at least).
> There are no NULL cells on my classes, neither on my Landsat 8 images or
> the Landsat 7 in the North Carolina dataset
ok
> > > I tested my procedure on the NC sample dataset and I'm getting the
same
> > > error, I'm using Grass 6.4.2 on Debian Wheezy, installed from the
debian
> > > repo. Here's the command set:
> > > #Display RGB enhanced map
> > > i.landsat.rgb blue=lsat7_2000_10 green=lsat7_2000_20
red=lsat7_2000_30
> > > d.rgb blue=lsat7_2000_10 green=lsat7_2000_20 red=lsat7_2000_30
> > >
> > > #"Real color" map
> > > r.composite red=lsat7_2000_30 green=lsat7_2000_20 blue=lsat7_2000_10
> > > output=lsat7_2000_RGB
> > >
> > > #Group images to process
> > > i.group group=lsat7_2000_RGB subgroup=lsat7_2000_RGB
> > > input=lsat7_2000_10,lsat7_2000_20,lsat7_2000_30
> > >
> > > #Unsupervised classification
> > > i.cluster group=lsat7_2000_RGB subgroup=lsat7_2000_RGB
> > > sigfile=clustlsat7
> > > classes=20 report=rep_clust_lsat7.txt
> >
> > (I have never tried to feed a group that uses a single composite. I'll
try
> > it out.)
>
> The composite is for the supervised clasification, the group and subgroup
> are made with the 3 RGB bands (1,2,3 in Landsat7 or 2,3,4 in Landsat8)
Right, I didn't pay much attention. Apologies.
> > i.maxlik group=lsat7_2000_RGB subgroup=lsat7_2000_RGB
sigfile=clustlsat7
> > class=lsat7_classes reject=lsat7_classes_rej
> > Does the unsupervised classification work? i.cluster requires at least
> > two input raster maps. And, honestly, I have never tried it the way you
> > did, i.e. by using a composite map. Also, r.composite will use by
default
> > "32 levels" whicih means that you loose important info for the
subsequent
> > classification, if I am not wrong.
Same error (of mine) as above, you correctly used a group with three bands.
> The supervised classification does work, i.cluster generates a signature
> file and i.maxlik success to create the classes raster map
You mean the _un_supervised I guess.
Yes, I meant the unsupervised, with i.cluster
> #Supervised classification
> >
> > > i.class map=lsat7_2000_RGB group=lsat7_2000_RGB
subgroup=lsat7_2000_RGB
> > > outsig=classlsat7
> > > i.maxlik group=lsat7_2000_RGB subgroup=lsat7_2000_RGB
sigfile=classlsat7
> > > class=lsat7_class_sup
> > >
> > > Here I get the error in which i.maxlik can't read the signature file
> > > created by i.class (classlsat7) but read the one created by i.cluster
> > > flawlessly
> > >
> > > This is what I'm selecting on the i.class monitor, I named it
> > > vegetation:
> > > [image: Imagen integrada 1]
> >
> > Only on class? Can you try with more please?
>
> Yes, the error I'm getting is only with the i.class module
Sorry for my typo -- I meant to write "only with 1 class?", "can you try
with
more than 1 class, i.e. 3 or 4 classes?".
I was using two classes, vegetation and construction, I tried two more
classes and got the same error
> I'm doing some tests creating a test vector with my classes as areas,
how many classes?
> then converting it to raster with v.to.rast,
can you post the exact command? What values are you attributing to the
rasterised classes?
I used the grass gui to draw some polygons on a test vector with
cat(integer),class(integer),desc(varchar) and assigned 1 for
vegetation or 2 for constructions
Then issued
v.to.rast in=test out=test use=attr col=class labelcol=desc
And got a raster with those areas, then created the signature with
i.gensigset group=lsat7_2000_RGB subgroup=lsat7_2000_RGB sig=lsat7_testsig
training=test
i.smap group=lsat7_2000_RGB subgroup=lsat7_2000_RGB sig=lsat7_testsig
out=veg_const
and then using that raster to generate statistics for i.smap with
> i.gensigset
> I don't know if I'm not using the i.class module the right way or if it's
> broken
I don't think it is broken as it usually works/ed fine. But we'll try to
nail
that down.
Nikos
I just tested with Grass7.0svn and the GUI tool for supervised
clasification (g.gui.iclass) generates a signature (with just one class)
that is read just fine by i.maxlik
--
*César Augusto Ramírez Franco*
Laboratorio de Sistemas Complejos Naturales
Escuela de Geociencias
Facultad de Ciencias
Universidad Nacional de Colombia - Sede Medellín
Teléfono: (57-4) 430 9369 - 301 389 5607
