Classification Tutorial 2: Collect Training Data

Classification Tutorial 2: Collect Training Data

This tutorial is the second in a two-part series about preparing data for image classification. It
requires ENVI 5.4 or later and will take approximately one hour to complete. You will learn
how to interactively define regions of interest (ROIs) that represent known land-cover types.
The ROIs serve as ground truth, or training, data. Training data is required for supervised
classification.

Files Used in this Tutorial

Tutorial files are available from our website or on the ENVI Resource DVD in the
classification directory. Copy these files to your local drive.
Data files were provided by the National Ecological Observatory Network, 2014. Files were
accessed in July 2016, available online at http://data.neonscience.org from the National
Ecological Observatory Network (NEON), Boulder, CO, USA.
File Description
NEONOrthophoto.dat Digital photograph from the NEON RGB camera at 0.25 meter
resolution
NEONOrthophoto.hdr ENVI header file for above

Open and Display an Orthophoto

In this section you will open a digital orthorectified photograph that was captured at the same
time as the NEON image used in "Classification Tutorial 1: Create an Attribute Image." The
two images are co-registered, so you can collect training data from the orthophoto and apply
it to the multispectral image or the attribute image. You will collect training data from the
orthophoto because it has a higher spatial resolution (0.25 meters) than the NEONUrban.dat
image (0.5 meters) used in the first tutorial. The colors in the orthophoto are also more
natural because it was captured with an RGB camera.
Follow these steps:
1. Start ENVI.

2. Click the Open button in the toolbar.

3. Navigate to the directory where you saved the tutorial files.

Page 1 of 8
© 2016 Exelis Visual Information Solutions, Inc., a subsidiary of Harris Corporation. All Rights Reserved. This
information is not subject to the controls of the International Traffic in Arms Regulations (ITAR) or the Export
Administration Regulations (EAR). However, this information may be restricted from transfer to various embargoed
countries under U.S. laws and regulations.
 Classification Tutorial 2: Collect Training Data

4. Select NEONOrthophoto.dat and click Open. The orthophoto appears in the Image
window.

Create Regions of Interest

In this section you will define training data ROIs for five different land-cover types:
n Building
n Tree
n Grass
n Concrete
n Asphalt
This is a simple use case with a few classes. Real-world land-use classification typically
involves many different feature types and training samples.

Page 2 of 8
© 2016 Exelis Visual Information Solutions, Inc., a subsidiary of Harris Corporation. All Rights Reserved. This
information is not subject to the controls of the International Traffic in Arms Regulations (ITAR) or the Export
Administration Regulations (EAR). However, this information may be restricted from transfer to various embargoed
countries under U.S. laws and regulations.
 Classification Tutorial 2: Collect Training Data

Collect Training Data for Buildings

1. Click the Region of Interest (ROI) Tool button in the toolbar.

2. Click the Add ROI button in the Region of Interest (ROI) Tool. A new Regions of
Interest layer appears in the Layer Manager. The default name is "ROI #1" and its color
is red.
3. Change the ROI Name to Building, and use the color picker to choose a different color
if desired.
4. Zoom into the orthophoto.
5. The Polygon icon is active in the ROI Tool. Click and drag the cursor to define a polygon
around some building pixels. To complete the polygon, right-click and select Complete
and Accept Polygon. For example:

Page 3 of 8
© 2016 Exelis Visual Information Solutions, Inc., a subsidiary of Harris Corporation. All Rights Reserved. This
information is not subject to the controls of the International Traffic in Arms Regulations (ITAR) or the Export
Administration Regulations (EAR). However, this information may be restricted from transfer to various embargoed
countries under U.S. laws and regulations.
 Classification Tutorial 2: Collect Training Data

To pan around the image while creating ROIs, press the middle mouse button and move
around the display. You can also click the Pan button in the toolbar. To resume drawing
ROIs, you must (1) click the Select button in the toolbar and (2) click inside of the
ROI Tool. The cursor will change back to a crosshair.
6. Create several more polygons around building pixels. You do not have to draw ROIs for
every building. The key is to draw ROIs around buildings with different colors,
intensities, and features (including solar panels and vents) so that you have a wide
representation of building types.

Collect Training Data for Other Land-Cover Types

1. Click the Add ROI button in the Region of Interest (ROI) Tool. A new Regions of
Interest layer appears in the Layer Manager. The default name is "ROI #1" and its color
is green.
2. Change the ROI Name to Tree.
3. Define polygons that represent tree pixels. Choose different colors of trees.
4. Repeat Steps 1-3 for the other ROIs:
n Grass: Select pixels that are purely grass, not a mixture of bare soil and grass.
n Concrete: Select pixels from sidewalks and driveways.
n Asphalt: Select pixels from roads and asphalt parking lots.
Here is an example with all five ROIs defined:

Page 4 of 8
© 2016 Exelis Visual Information Solutions, Inc., a subsidiary of Harris Corporation. All Rights Reserved. This
information is not subject to the controls of the International Traffic in Arms Regulations (ITAR) or the Export
Administration Regulations (EAR). However, this information may be restricted from transfer to various embargoed
countries under U.S. laws and regulations.
 Classification Tutorial 2: Collect Training Data

5. When you are finished creating ROIs, select File > Save As from the ROI Tool menu
bar.
6. In the Save ROIs to .XML dialog, click the Select All Items button to select all five
ROIs.

Page 5 of 8
© 2016 Exelis Visual Information Solutions, Inc., a subsidiary of Harris Corporation. All Rights Reserved. This
information is not subject to the controls of the International Traffic in Arms Regulations (ITAR) or the Export
Administration Regulations (EAR). However, this information may be restricted from transfer to various embargoed
countries under U.S. laws and regulations.
 Classification Tutorial 2: Collect Training Data

7. Click the button next to the Enter Output File field.

8. Select an output folder, and name the output file TrainingDataROIs.xml.
9. Click OK in the Save ROIs to .XML dialog.
10. Exit ENVI.

Next Steps
Now you can use the attribute image (from "Classification Tutorial 1: Create an Attribute
Image") and training data ROIs from this tutorial for classification. Here are some different
options:
n See the following code examples in ENVI Help for instructions on running an end-to-end
classification workflow using the attribute image and training ROIs:

Page 6 of 8
© 2016 Exelis Visual Information Solutions, Inc., a subsidiary of Harris Corporation. All Rights Reserved. This
information is not subject to the controls of the International Traffic in Arms Regulations (ITAR) or the Export
Administration Regulations (EAR). However, this information may be restricted from transfer to various embargoed
countries under U.S. laws and regulations.
 Classification Tutorial 2: Collect Training Data

n Code Example: Support Vector Machine Classification Using API Objects

n Code Example: Softmax Regression Classification Using API Objects
n Or, use the ENVI Classification workflow by selecting Classification > Classification
Workflow from the ENVI Toolbox. When the workflow starts, select an input Raster
File and click Next. Select the Use Training Data option, then select
AttributeImage.dat for the source image and TrainingDataROIs.xml for the training
data.

Copyright Notice:
ENVI is a registered trademark of Harris Corporation.

Page 7 of 8
© 2016 Exelis Visual Information Solutions, Inc., a subsidiary of Harris Corporation. All Rights Reserved. This
information is not subject to the controls of the International Traffic in Arms Regulations (ITAR) or the Export
Administration Regulations (EAR). However, this information may be restricted from transfer to various embargoed
countries under U.S. laws and regulations.