Sample set editor

Functionality (1)

In the sample set editor you can select sample pixels also called training pixels for certain classes prior to an image classification. The sample set editor has a menu bar, a context-sensitive menu, a toolbar and a status bar.

Sampling creates a sample set which is needed for a multi-spectral image classification.

Sampling is the assignment of class names to groups of pixels which have similar spectral values and that are supposed to represent the classes that you want to obtain from the classification. The sampled pixels or training pixels are thus supposed to be characteristic for a certain type of a natural resource. For instance, to perform a land use classification, you might sample the classes: forest, grass land, arable land etc.

Requirements:

For sampling you need an image which can be easily interpreted (e.g. a color composite), a map list, that is the set of images which you want to classify in a later stage, and a class domain containing the set of class names that you want to assign to training pixels. All maps must have the same georeference (or georef None).

Tip:

When your graphics board is configured to use more than 256 colors, you can also use an interactive color composite; then, a background map on disk is not required. For more information, refer to How to create a color composite.

To start the sample set editor, refer to How to create a sample set.

General working of the sample set editor:

During sampling, you probably need to zoom in on the map. You can still add new classes to the domain while sampling. It is advised to use the same domains as far as possible (not every class has to be used).

Tips:

Selection of pixels:

Inspect the Sample Statistics to see the statistics of the current selection.

Assignment of a class name to a selection:

When a new class is added, also a color can be selected. Class names in the domain that you are using can be as long as you want: set the width of the domain through the properties of the domain. Sampled pixels appear in the color of the assigned class in the background map; when you move the mouse over a sampled pixel, the status line shows the class name of the pixel.

View Sample Statistics:

Sample Statistics are displayed in a separate window. In the upper half of the window, you can select a class name, then some statistics are calculated for the training pixels belonging to this class: mean, standard deviation, number of pixels with the predominant value, predominant value, and total number of training pixels belonging to this class. In the lower half of the Sample Statistics window, statistics are shown for currently selected pixels.

View Feature Spaces:

To view a feature space, press the right mouse button in the sample set editor, and select Feature Space from the context-sensitive menu. You can also click the Feature Space button in the toolbar, or choose Feature Space from the normal Edit menu. Select which bands you want to be displayed by the feature space.

You can open as many feature spaces as you like, but as each of them is updated after a new selection of training pixels, speed may decrease. Feature spaces can be shown with or without additional grid lines, can be copied to clipboard and printed.

On the basis of the sample statistics and the feature spaces, you can for instance decide to sample more training pixels of a certain class, merge classes with too much overlap, delete classes, etc. For more information on merging and deleting classes, changing the color of classes, see Sample set editor : functionality (2).

Exit the sample set editor:

To exit the sample set editor, double-click the Control-menu box of the sample set editor window or the sample statistics window. You can also click the Exit Editor button in the toolbar, or choose Exit from the File menu in the sample set editor.

See also: