Classification

Four methods can be used to classify the objects created by the Object Extraction tool. Furthermore, the easiest way to change the class of one object is to double-click on it or to select Properties in the contextual menu to display a dialog which permits to change the class.

 Note: Only the checked classes, which are visible on the image (see Display), are used by the classification methods. If you do not want to involve a specific class during a classification, just uncheck it.

Class Editor: Make your own list of classes.

Result View: Open the Data Viewer directly in the Classes Mode of the Object Extraction tool.

Manual

Select a class in the combo box and then click on a object to assign the selected class.

Delimit an area

Draw one or more polygonal areas and then click on the Apply button to assign the selected class to all the objects that are inside the polygons or touching them.

 Note: Press the selection toggle button in the toolbar to stop drawing polygons.

Single parameter

Select a parameter, define a range by entering the min and max values, and then press the Apply button to assign the selected class to all objects included in this range.

 Note: Displaying the values distribution of the Histogram may help to choose the range.

Parameter: Select one of the object descriptors. The Min and Max text fields values are the extrema of the selected parameter.

Reset (on top): Set again the extrema (Min and Max) of the selected parameter.

Keep the last result: When this box is unchecked, the previous classification is canceled when a new classification is applied.

Histogram: Show the histogram of the selected parameter.

Supervised classification

The supervised classification uses the nearest neighbors or the k-nearest neighbors classifier. It involves a set of training objects. New samples are classified basically by calculating the smallest Euclidian distances (parameters space) to the nearest training objects (the most represented class is chosen).

Performing a classification

  1. Create the training set. Firstly, add the desired classes (do not use existing classes that contain objects) with the Class Editor. Then, assign for each class some training objects by using the Manual classification (see above).
  2. Select the relevant parameters for the classification in the New Learning Session tab.
  3. Click on Apply and then check the class boxes that include objects to classify. The other class boxes are the training classes defined in the first point. If the k-nearest neighbors classifier is selected, choose a k number (by default 10), which represents the number of the nearest training objects involved in the classification decision.

Weight: Set a weight (1 to 100%) to the parameters involved in the classification.

 

Load and Save Learning

Once the objects have been classified (see Performing a classification), the statistics of parameters of each class are displayed in the Current Learning Session tab. Learning can be saved, and then reloaded and applied to objects in another project.

 Note: Learning containing additional parameters (see Object Descriptors) cannot be saved (does not appear in the Current Learning Session tab), because these descriptors have variable parameters. Involving them could produce erroneous results.

Add objects: Add new training objects to the selected class.