Iso Cluster: Maximum number of bands?

114
6
01-22-2020 05:56 AM
Highlighted
New Contributor

Does anyone know if Iso Cluster has an upper limit for the number of bands that it can process?

Reply
0 Kudos
6 Replies
Highlighted
MVP Esteemed Contributor

Iso Cluster—Help | ArcGIS Desktop 

The minimum valid value for the number of classes is two. There is no maximum number of clusters. In general, more clusters require more iterations.

Reply
0 Kudos
Highlighted
Esri Regular Contributor

Jeffrey

I will explore this and try to determine if there is a maximum.  I'm sure it does work with more than 3 bands but if you have 512 bands of hyperspectral data there may be limitations.

It's helpful if you can provide more information.  Are you using ArcGIS Pro 2.4?  ArcMap 10.7.1?  Are you trying it and running into limitations?  What format is your file?  Are you loading the file into the map, then adding the layer from the Table of Contents into the tool?

Cody B

Highlighted
New Contributor

Thanks to you both for replying. Cody, you've understood my question. Dan, you have not. And I just realized that I didn't ask a couple of associated questions.

My original question: Does Iso Cluster have an upper limit to the number of BANDS that it can process? Associated questions:

Does Iso Cluster Unsupervised Classification have an upper limit....

Does Composite Bands have an upper limit....

We are using ArcMap 10.7.1. Our files are geotiffs. We typically do not load the file into the map.  We use Composite Bands to combine a number of four- and five-band images and load that composite into Iso Cluster Unsupervised Classification.  We have used it with as many as 36 bands, and it seems to have run normally, at least no error messages. The output seems reasonable relative to the nature of the imagery and our objectives.  But who knows with so many dimensions? We classify the same imagery multiple times specifying different numbers of clusters. I don't think we ever specified more than eight clusters. 

Reply
0 Kudos
Highlighted
MVP Esteemed Contributor

my bad

Reply
0 Kudos
Highlighted
Esri Regular Contributor

Jeffrey

I'm told by the engineers that there is no theoretical limit for # of input bands.  In the range of the number of input bands/image that you've done in the past, if it works, it sounds like it's a viable solution.

However, if I understand your workflow correctly, you may want to consider alternatives for how you're doing the processing.  To clarify: you say "combine a number of four- and five-band images" and "max has been 36 bands" so does that mean you had 9 four-band images from the same sensor captured on different dates, and you combined those 9 images into a single multitemporal stack then ran IsoCluster?  

I'm just thinking that you may not get the best results given 9 bands of redundant spectral content in each of the four truly unique spectral bands.  It sounds like you're mixing spectral classification with temporal change detection in IsoCluster as a "magic black box".  If my interpretation is correct, that's interesting and innovative - but I am thinking you may want to process the multibands from each date separately, and address the multitemporal changes in a separate step.  There are some bright people here at Esri with more expertise, if you want to explore this.

Cody

Highlighted
New Contributor

Hi Cody,

Thanks for your response, please excuse my delay in continuing the thread.

You have understood our workflow.  The 36 bands result from 9 four-band images from the same sensor captured on different dates combined into a single multitemporal stack as the input to Iso Cluster.  We are indeed mixing spectral classification with temporal change, but we are not interested in temporal change detection. (Because of IP concerns, I can't really detail why we're doing what we're doing.) We are using Iso Cluster essentially as you describe: a black box (Without the source code, Iso Cluster is pretty much a "black box" anyway.(:>)...).  We are exploring PCA for data reduction. 

Reply
0 Kudos