2

I am using ArcGIS Pro 2.4.0, and am attempting to classify a field into a schema with three categories: Leaves, Shadows, and Soil. Here is a sample of the original RGB image:

enter image description here

Through supervised classification (SVM), I was able to create training samples and classify the image into the the desired categories:

enter image description here

Note: Green = leaves, black = shadows, brown = soil. Despite not having the accuracy statistics generated yet, it looks very accurate.

When I try to do the same thing with an unsupervised pixel-based classification (ISO is the only option on ArcGIS Pro that I am aware of), it will not let me divide it into three classes. Instead, it only gives me two:

enter image description here

The only setting I changed from the default ISO cluster settings was the maximum number of classes. I changed that from 5 to 3:

enter image description here

Does anybody know why this would happen? I don't have as much experience with unsupervised classifications as I do with supervised, so I'm not sure why I am not able to get 3 classes. Do I need to change some additional settings?

2
+50

Instead of following the Wizard approach, using the Iso Cluster Unsupervised Classification tool from the Geoprocessing Toolbox inside Spatial Analyst Tools > Multivariate will allow you to choose the exact amount of classes you want.

As you can see, the classes are not the maximum but an exact choice:

enter image description here

I've tried it with 3 classes not so successfully:

enter image description here

And this is an example using 5 classes:

enter image description here

Considering it's been made from a screenshot it is fairly ok.

2
  • Btw, I'm curious. Is this an artichoke field? – Albert Nov 26 '20 at 13:27
  • 1
    Grain sorghum actually! – ihb Nov 29 '20 at 13:30
2

Maximum number of classes is a maximum - no more than 3 classes can be created. It does not imply that three classes must be created. It looks like to me ISO isn't being able to tell the difference between soil and shadows, and thus is grouping them together into a single class. I don't know much about ISO specifically, but I would try playing around with the other parameters too to see if they enable the classifier to differentiate between more than two classes.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.