Creating Training Samples for Supervised Image Classification
Included in this tutorial:
Accessing and reviewing the Training Sample Manager
Creating training samples in a new class, including deleting sample polygons and collapsing and expanding samples within a class.
Saving your schema and training samples
Software version in examples: ArcGIS Pro 3.0.2
Tutorial Data: This tutorial demonstrates with Landsat 8 data downloaded from the USGS EarthExplorer. The dataset is a scene acquired in late 2018, covering Phnom Penh, Cambodia.
US Geological Survey. “Scene LC80_L1TP_126052_20181225_20190129_01_T1” [geoTIFFs]. Landsat 8 OLI/TIRS, Collection 1, Level 1. Acquisition date: 25 December 2018. (29 January 2019). Accessed 14 August 2020 from https://earthexplorer.usgs.gov/
Credits: L. Meisterlin (2022)
This tutorial demonstrates creating training samples from a false color image in ArcGIS Pro, for later use in supervised image classification.
In the example, we create training samples based on a Shortwave Infrared composite (also known as False Color Urban), and develop samples depicting wetlands.
Related tutorials:
This tool requires either the Image Analyst Extension in ArcGIS Pro. To verify that you have the Network Analyst Extension, see Which Esri Extensions Is My Software Licensed to Use?
For guidance on downloading Landsat data from the EarthExporer website, see Downloading Landsat Data from the USGS EarthExplorer.
The tutorial begins with a three-band raster created from a composite raster created from this dataset. For more on these tools in ArcGIS Pro, see Creating a Composite Band Raster Dataset and Making a Raster Layer from a Multiband Raster Dataset.
the Training Sample Manager
Accessing and Inspecting the Training Sample Manager
Access the Training Sample Manager in the main menu ribbon (with a composite layer selected in the Contents panel) under Imagery > Classification Tools > Training Sample Manager. This will summon the Training Sample Manager in the Image Classification pane.
The demonstration below walks through accessing the manager as well as some key aspects of its default configuration.
(Note that in all the demonstrations of this tutorial, left-clicks are represented as magenta and right-clicks are represented as yellow.)
accessing the Training Samples Manager and an overview of its options and default settings
Notice that, by default, the manager includes a classification scheme from the (US) National Land Cover Database (2011).
Confirming (or changing) the Active Layer
If you have more than one applicable layer in your map project, the menu at the top right of the Training Sample Manager allows you to confirm or change the active layer.
confirming the active layer in the Training Sample Manager
Creating training samples in a new class
In the demonstration below, we…
Collapse (hide) the default classification schema.
Create a new classification schema and change its properties by right-clicking on its name in the manager. In the example, we change the schema’s name to “Water.”
Create a new class (assigning it a color for visibility and a value) within the new schema. In the example, we also name the class “Water” because we will collect a few samples depicting water within this class. We assign it a color for the purposes of visibility (so we chose a bright magenta) while creating samples and give it a value. Here the example value is “1.”
Hover over the drawing tools for your reference. Options include rectangles and polygons as well as other shape-drawing tools.
Create the first training samples by drawing polygons. Choosing the polygon tool, in the example we draw two polygons around areas we know represent water (in this visualization, water is depicted in shades of dark blue).
creating the first training samples in a new class
Note: The demonstration shows adding a new class by right-clicking on the schema name and choosing “New class.” You can also add a new class by clicking the “+” button at the top of the panel’s menu.
Deleting (a bad) sample polygon
If you need to, you can individually highlight, modify, or delete the sample polygons. In the demonstration below, we highlight a sample polygon (which includes areas that are not dark blue and thus, we know, do not represent water) in the manager and remove it from our training set by clicking the “delete” button.
Collapsing (and Expanding) the samples within a class
The samples you create within a class must be collapsed (or merged) together. Essentially, together they comprise the representative samples used to train the automated classification. (Alone they each only represent one condition.)
To collapse (or merge) the polygon samples within the class, highlight the ones you want to combine and click the “collapse” button.
To expand a collapsed set of samples, you can highlight your choice and click the “expand” button.
The demonstration below indicates the locations of both of these buttons, and shows an example of collapsing multiple polygons of our example “water” class.
Saving your schema & training samples
To save your schema and training samples, the manager has several different “save” buttons. In short,
There are two save buttons for the schema: to save the current schema or to save a copy of the schema
There are two save options for the training samples: to save the current training samples within your map project and to save a copy of the training samples as a feature class.
Note: Notice that there is a “load” button for both the schema and the training samples. With these you can “load in” previously saved schemas and samples.
In the demonstration below, we walk through a few of these saving options.
saving the schema and training samples