[Freegis-list] classification with i.class,i.cluster

[Markus Neteler]

On Mon, Mar 24, 2003 at 09:46:56AM +0100, Heiko.Kehlenbrink at vermes.fh-oldenburg.de wrote:
> hi list, 
> i just got my marks for a homework i did using grass. 
> i had to classify several areas in a landsat scene. 
> it was the first time that a stundent used grass therefor and i passed. 
> usually we had to use pci-works but with about 35 students and only 10 dongles it is a pain... 
> some points i like to annotate: 
>  
> it would be great if one could define a trainingarea assembled out of severeal polygons  

You can also run (instead of i.class): 
 - r.digit, i.gensig, or:
 - v.digit, v.to.rast, i.gensig

or use i.smap instead of i.maxlik (i.smap is much better).

Summary of the standard GRASS reclassification techniques:

        |radiometric, | radiometric,| radiometric,| radio- and geo-
        |unsupervised | supervised  | supervised  | metric, supervised
--------+-------------+-------------+-------------+-------------------
Preproc.|i.cluster    |i.class      | i.gensig    | i.gensigset
        |             |(monitor)    | (from maps) | (from maps)
--------+-------------+-------------+-------------+-------------------
Compu-  |i.maxlik     |  i.maxlik   |  i.maxlik   | i.smap
tation  |             |             |             |
--------+-------------+-------------+-------------+-------------------

> we had to identify 6 main classes (water, wood,swamp,rural,grassland and urban) 
> afterwards we had to split those classes down to say coniferous forest and deciduous forest. 
>  
> using i.cluster with the six classes the outcome wasn`t so good, there were many forrest areas 
> identified as water, urban areas identified as swamp and so on... 
>  
> using i.cluster with the subclasses , about 12 basing upon the six main classes, the outcome 
> was ok. 

Again I recommend i.gensigset, i.smap.

> in the report generated by i.cluster there was a matrix regarding the seperability of the classes 
> but there wasn`t an explanation how to rate the values in the matrix. 
>  
> all in all it was again very instructional working with grass ... 

Best regards

 Markus Neteler