Hi - I would like to do an unsupervised classification using i.cluster from GRASS 7 on my Ubuntu 64-bit machine. When I run i.cluster then i.maxlik on a Landsat image I only get one output class, the entire image. Only one class appears in the signature file so the problem is with the i.cluster step. I tried running i.cluster on different images and always get just one class.
Here is the command line I'm using:
i.cluster group=TM_r15c33@PERMANENT subgroup=AllBands signaturefile=IsoDataSigs2 classes=20 sample=5,5 convergence=99.9 min_size=10000
Is anyone able to get i.cluster to work properly? It would be helpful to know if it's a user problem or a bug.
Thanks in advance to anyone who can help.
All the best,
Ned
Hi Ned,
I have been using i.cluster in GRASS 7 on a 64-bit fedora server,
without any trouble. However, I managed to replicate your problem
and create a single class output by setting the parameter
‘sample=5,5’ as you had done in your example. If I run i.cluster without
that parameter, the signature file and classification output both
have the number of predefined classes.
Can you see if this helps with your data?
Regards,
Daniel.
On Thu, Feb 17, 2011 at 7:49 PM, Ned Horning <horning@amnh.org> wrote:
Hi - I would like to do an unsupervised classification using i.cluster from GRASS 7 on my Ubuntu 64-bit machine. When I run i.cluster then i.maxlik on a Landsat image I only get one output class, the entire image. Only one class appears in the signature file so the problem is with the i.cluster step. I tried running i.cluster on different images and always get just one class.
Here is the command line I’m using:
i.cluster group=TM_r15c33@PERMANENT subgroup=AllBands signaturefile=IsoDataSigs2 classes=20 sample=5,5 convergence=99.9 min_size=10000
Is anyone able to get i.cluster to work properly? It would be helpful to know if it’s a user problem or a bug.
Thanks in advance to anyone who can help.
All the best,
Ned
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user
Hi Ned and Daniel,
I have run into this problem in the past, and I believe it may have to do with your convergence, min_size, and classes settings. Try decreasing some of those and see if you don’t get better results. You can also try increasing your separation from 0.0 (the default) to 1.0 or 1.5.
Nick
On Fri, Feb 18, 2011 at 4:24 PM, daniel mcinerney <daniel.o.mcinerney@gmail.com> wrote:
Hi Ned,
I have been using i.cluster in GRASS 7 on a 64-bit fedora server,
without any trouble. However, I managed to replicate your problem
and create a single class output by setting the parameter
‘sample=5,5’ as you had done in your example. If I run i.cluster without
that parameter, the signature file and classification output both
have the number of predefined classes.
Can you see if this helps with your data?
Regards,
Daniel.
On Thu, Feb 17, 2011 at 7:49 PM, Ned Horning <horning@amnh.org> wrote:
Hi - I would like to do an unsupervised classification using i.cluster from GRASS 7 on my Ubuntu 64-bit machine. When I run i.cluster then i.maxlik on a Landsat image I only get one output class, the entire image. Only one class appears in the signature file so the problem is with the i.cluster step. I tried running i.cluster on different images and always get just one class.
Here is the command line I’m using:
i.cluster group=TM_r15c33@PERMANENT subgroup=AllBands signaturefile=IsoDataSigs2 classes=20 sample=5,5 convergence=99.9 min_size=10000
Is anyone able to get i.cluster to work properly? It would be helpful to know if it’s a user problem or a bug.
Thanks in advance to anyone who can help.
All the best,
Ned
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user
Also, make sure you mask any Nulls since that often seems to be the problem when i.cluster doesn’t work as expected, especially if you are using landsat 7 slc-off imagery - although I’ve also experienced it with Landsat 4 and 5 imagery.
[Sent from a mobile device]
On Feb 18, 2011, at 4:38 PM, Nick Jachowski <njachowski@gmail.com> wrote:
Hi Ned and Daniel,
I have run into this problem in the past, and I believe it may have to do with your convergence, min_size, and classes settings. Try decreasing some of those and see if you don’t get better results. You can also try increasing your separation from 0.0 (the default) to 1.0 or 1.5.
Nick
On Fri, Feb 18, 2011 at 4:24 PM, daniel mcinerney <daniel.o.mcinerney@gmail.com> wrote:
Hi Ned,
I have been using i.cluster in GRASS 7 on a 64-bit fedora server,
without any trouble. However, I managed to replicate your problem
and create a single class output by setting the parameter
‘sample=5,5’ as you had done in your example. If I run i.cluster without
that parameter, the signature file and classification output both
have the number of predefined classes.
Can you see if this helps with your data?
Regards,
Daniel.
On Thu, Feb 17, 2011 at 7:49 PM, Ned Horning <horning@amnh.org> wrote:
Hi - I would like to do an unsupervised classification using i.cluster from GRASS 7 on my Ubuntu 64-bit machine. When I run i.cluster then i.maxlik on a Landsat image I only get one output class, the entire image. Only one class appears in the signature file so the problem is with the i.cluster step. I tried running i.cluster on different images and always get just one class.
Here is the command line I’m using:
i.cluster group=TM_r15c33@PERMANENT subgroup=AllBands signaturefile=IsoDataSigs2 classes=20 sample=5,5 convergence=99.9 min_size=10000
Is anyone able to get i.cluster to work properly? It would be helpful to know if it’s a user problem or a bug.
Thanks in advance to anyone who can help.
All the best,
Ned
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user
Nick and Daniel,
Thanks for the suggestions. I did manage to get it to run after playing with some of the variables. If I used the default values for “sample” and “min_size” it worked. Unfortunately the default values are less than ideal. I think the main problem is with the “sample” parameter. When I enter “sample=5,5” I assume that every fifth pixel from every fifth line is being sampled for use in calculating the cluster statistics. This is apparently too many samples for i.cluster to handle. If I use “sample=100,100” the program works as expected (more or less). I have a feeling that the sample array is not able to handle a large number of samples.
Thanks again,
Ned
On 02/18/2011 09:22 AM, Nick Jachowski wrote:
Also, make sure you mask any Nulls since that often seems to be the problem when i.cluster doesn’t work as expected, especially if you are using landsat 7 slc-off imagery - although I’ve also experienced it with Landsat 4 and 5 imagery.
[Sent from a mobile device]
On Feb 18, 2011, at 4:38 PM, Nick Jachowski <njachowski@gmail.com> wrote:
Hi Ned and Daniel,
I have run into this problem in the past, and I believe it may have to do with your convergence, min_size, and classes settings. Try decreasing some of those and see if you don’t get better results. You can also try increasing your separation from 0.0 (the default) to 1.0 or 1.5.
Nick
On Fri, Feb 18, 2011 at 4:24 PM, daniel mcinerney <daniel.o.mcinerney@gmail.com> wrote:
Hi Ned,
I have been using i.cluster in GRASS 7 on a 64-bit fedora server,
without any trouble. However, I managed to replicate your problem
and create a single class output by setting the parameter
‘sample=5,5’ as you had done in your example. If I run i.cluster without
that parameter, the signature file and classification output both
have the number of predefined classes.
Can you see if this helps with your data?
Regards,
Daniel.
On Thu, Feb 17, 2011 at 7:49 PM, Ned Horning <horning@amnh.org> wrote:
Hi - I would like to do an unsupervised classification using i.cluster from GRASS 7 on my Ubuntu 64-bit machine. When I run i.cluster then i.maxlik on a Landsat image I only get one output class, the entire image. Only one class appears in the signature file so the problem is with the i.cluster step. I tried running i.cluster on different images and always get just one class.
Here is the command line I’m using:
i.cluster group=TM_r15c33@PERMANENT subgroup=AllBands signaturefile=IsoDataSigs2 classes=20 sample=5,5 convergence=99.9 min_size=10000
Is anyone able to get i.cluster to work properly? It would be helpful to know if it’s a user problem or a bug.
Thanks in advance to anyone who can help.
All the best,
Ned
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user