5. Using the Semi-Automatic Classification Plugin

5.5. Import Training Data

The classifier needs training data. With SCP we can easily import the training data polygons from the previous chapter and calculate the spectral signatures for each class.

  1. Go to the SCP Dock.
  2. Select the Training Input tab at the left of the dock.
  3. Click to create a new training input.
  4. Save the file as training_input.scpx.
  5. In the main menu, go to SCP | Basic tools | Import signatures.
  6. Click on the Import vector section to expand it.
  7. Click to browse to select training_areas.gpkg that we have created in Section 4.3.

SCP uses a two‑level class hierarchy to help you organise training data and produce cleaner, more interpretable land‑cover maps. Macro Classes are the high‑level, thematic categories. Think of them as the main land‑cover groups, such as:

  • Vegetation
  • Water
  • Built‑up
  • Bare soil
  • Crops

They help you structure your classification and group related detailed classes together.

Micro Classes are the detailed, fine‑grained categories inside each macro class. Examples:

Macro Class Micro Classes
Vegetation Forest, Grassland, Shrubland
Water River, Lake, Irrigation Canal
Crops Wheat, Maize, Rice, Sugarcane

When you create training areas in SCP:

  • Each area is assigned a Micro Class ID.

  • Each Micro Class belongs to a Macro Class ID.

During classification:

  • The classifier predicts Micro Classes.

  • You can later aggregate them into Macro Classes for simpler maps or reporting.

In this tutorial we don't use this hierarchy. Therefore we define the same fields for both micro and macro classes.

8. Select the fields as in the screenshot below. Remember that the layer field contains the class names.

9. Keep the rest as is and click Import vector .

Wait until the process has finished. This can take a while.

Let's add the class colours to the training areas. Training areas are also called Regions of Interest (ROI).

10. Click on the Color column to change the colour of a Macro Class (the bold line in the table). We don't need to change this for the micro classes.

In the next section, we're going to explore the spectral separation of the training areas.