[GRASSLIST:1860] Re: i.cluster

[Markus Neteler]

Hi Martin,

On Mon, May 21, 2001 at 02:23:57PM +0000, Martin Wass wrote:
> Why is i.cluster only produsing one class in lover sample as 3,3
> 
> Exampel
> subgroup: cluster2 = Landsat TM  Band 1,2,3,4,5,7
> 
> 
> i.cluster group=sachsen subgroup=cluster2 sigfile=sig.25 classes=100 
> sample=3,3 iterations=30 convergence=98 separation=0 min_size=2
> 
> Reading image ...  100%
> Iteration 1: %Convergence 2.30 (2m30s elapsed, 35m00s left)
> Iteration 2: %Convergence 92.48 (2m40s elapsed, 24m00s left)
> Iteration 3: %Convergence 100.00 (2m49s elapsed, 18m18s left)
> 
> Sig25 have now only one class why ? I expected it to be 100

you provided "100" - this is a theoretical value which might not
be found in your data. So in reality it may be smaller. Assume you
have an image group with everywhere the same value: Even if you
specify class=100 you will get one class only as only one class
is inside the image. So it depends on your data.

> It works better white sample=10,10 and sample=20,20
Might be possible, depends on your data.

> Is it true that i.cluster never create more clusters than the number of 
> start classes ?
Yes, I think so.

For "separation" you may try 0.5. "iterations" I wouldn't specify but leave
this to GRASS. "sample" I wouldn't specify.

> Thanks for the help!
> / Maritn Wass

You will have to try several parameters as i.cluster only works from
the image statistics. Better results you will get from i.gensigset/i.smap,
an excellent algorithm.

Cheers

 Markus

-- 
Markus Neteler *  University of Hannover
Institute of Physical Geography and Landscape Ecology
Schneiderberg 50 * D-30167 Hannover * Germany
Tel: ++49-(0)511-762-4494  Fax: -3984