Create a training sample dataset from existing points

[WadeWall]

Hi all,

I have run through the tutorial on performing image classification from both an object-oriented approach and a pixel-based approach. However, one of the things that is unclear to me is if it is possible to create a training sample dataset from a point or polygon file. For example many of us have binary data (e.g. presence(1), absence(0)) that we could use to create the training samples. Is it possible to do this? All of the methods I have seen start with an existing schema (bonus points if you can tell me how these are created). For example, here:

Creating training samples—ArcGIS Help | ArcGIS for Desktop 

and here:

Managing training samples—ArcGIS Help | ArcGIS for Desktop 

Does anyone know how I would create from scratch?

[Anonymous User · Accepted solution]

Hey Wade - you can definitely use an existing point or polygon feature class for a training sample dataset. The feature class must contain the following fields:

• classname—A text field indicating the name of the class category
• classvalue—A long integer field containing the integer value for each class category

Check out the in_training_features parameter in any of the training tools to see this requirement, e.g.:
Train Support Vector Machine Classifier (Spatial Analyst)—ArcGIS Pro | Documentation 

So the process would be:

1. Add the required classname and classvalue fields to your existing training sample dataset.

2. Run one of the training tools (support vector machine, random forest, maximum likelihood) using the imagery and the training samples you already have.

3. Run the Classify Raster tool using the imagery and the output from the classifier in step 2.