[GRASS-user] Re: problem with i.gensigset

[Micha Silver]

Hello Natalia:

(Returning to the list - perhaps someone else can help also)


On 11/05/2010 03:20 PM, Sofina Natalia wrote:
> Am 04.11.2010 20:20, schrieb Micha Silver:
>>
>> The "too many subclasses" is not necessarily an error. I believe that 
>> it means one classification is too complicated, that is the spread of 
>> pixel values is such that it can't create a signature for that class.
>> Did the process finish?
>> How did you make the training map? Are all the training areas with 
>> the same category uniform? If you setup a training area that is too 
>> large, with a broad mix of pixel values, then gensigset won't be able 
>> to create unique signatures for each class.
>> HTH
>>
>> --------------------------------------------------------------------------------------------
>
> Hi, Micha!
>
> I can't create a "sig-file" still... Below it is what I got after 
> starting the i.gensigset:
>
> ---------------------------------------------------------------------------------------------------------------
>
> (Fri Nov 05 12:47:14 2010)
>
> i.gensigset trainingmap=my_cl at image_classification 
> group=train at image_classification subgroup=train_sub 
> signaturefile=train_sig
>

No need to explicitly enter the mapset here. It shoud be enough to do:
i.gensigset trainingmap=my_cl group=train \
     subgroup=train_sub signaturefile=train_sig

> Finding training classes...
>
> 6 classes found
>
> Reading raster maps...
>
> Clustering class 1 (3324 pixels)...
>
> Removed a singular subsignature number 5 (9 remain)
>
> Removed a singular subsignature number 8 (8 remain)
>
> Number of subclasses is 6
>
> Clustering class 2 (1898 pixels)...
>
> Removed a singular subsignature number 9 (9 remain)
>
> Number of subclasses is 3
>
> Clustering class 3 (2710 pixels)...
>
> Number of subclasses is 10
>
> Clustering class 4 (1347 pixels)...
>
> Removed a singular subsignature number 10 (9 remain)
>
> Removed a singular subsignature number 6 (8 remain)
>
> Removed a singular subsignature number 2 (7 remain)
>
> Number of subclasses is 6
>
> Clustering class 5 (1024 pixels)...
>
> Number of subclasses is 4
>
> Clustering class 6 (2107 pixels)...
>
> Number of subclasses is 1
>

So far this looks normal. You have a few thousand pixels in each 
training area which should be enough.

> unable to create signature file train_sig for subgroup train_sub of 
> group train at image_classification
>

Hmmm... What went wrong? One additional test below...

> ERROR: Unable to create signature file <train_sig>
>
> (Fri Nov 05 12:47:19 2010) Command finished (5 sec)
>
>
> My training map I make so as it is described in your paper. I've tried 
> to create it with different sizes of the training areas. And the color 
> of the area I define using the function r.what.color.
>

I'm not sure what you mean here. Each training area should cover a 
uniform feature in the landscape. Perhaps send the output of:
r.info  my_cl
r.univar my_cl


> I don't understand what is wrong... Which size should be a training 
> area? Does it depend on the number of pixels?
>
> Thanks a lot,
>
> Natalia
>
>
>
> This mail was received via Mail-SeCure System.


-- 
Micha Silver
Arava Development Co. +972-52-3665918
http://www.surfaces.co.il


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.osgeo.org/pipermail/grass-user/attachments/20101105/332cbbc7/attachment.html