Supervised Classification of Images
Supervised classification is used when there are image regions where the user has information that allow interest class identification.
The user must identify in the image a representative area of each class. It is important that supervised area is a homogeneous sample of respective class but at the same time must include all gray level variation of the theme.
It is recommended user acquire more than one supervised area using the biggest number of available information such as field work, maps, etc.
To obtain reliable statistical classes, it is necessary from 10 to 100 pixels of samples per class. The number of pixels to an acceptable accuracy of a class increases with the variation increase among classes.
Select IMAGE PROCESSING → IMAGE PROCESSING → SUPERVISED CLASSIFICATION in the main menu.
In opened window select bands 1 and 3 and click on the right arrow button.
Click on the button Sample Acquisition.
Region of Interest window opens.
Options:
Click on Create New Samples By and choose one of the 3 options: ROI, Points Theme or Polygons Theme.
If previously saved samples will be used instead of create a new one click on Load Samples and click on the File button to browse and find the file.
Select the option Region of Interest.
Click on the Next button.
Name: enter the name of the first class, e.g. water.
Click on the Color button to define the class color, e.g. blue.
Click on the Create button.
Enter another name to create one more class.
Click on the Color button to define color of the second class, e.g. red.
Click on the Create button.
Create and attribute a color to all selected bands repeating the same procedure. When finished click on the Next button.
Class: select one class created in previous window to insert the samples.
Click on the icon
or
to draw polygons or rectangles in visualization area and do sample
collection, to close the polygon click with mouse right button.
Acquire as many as samples needed.
An alternative to
sample pixels is the magic wand. Click on icon
and then use the mouse to select correspondent pixel to the class.
Besides the pixel selection button there is a slide button to adjust
??????.
Transparence: ?????????
Select the second class and repeat same procedure until finish all available classes.
Click on the OK button.
To save created
samples click on the icon
and define a path and file name.
Click in OK button.
Click on the Classification button.
Enter the name of the classified image in Output Image Name field.
Select the classifier.
To verify created samples, click on the Samples Analysis button.
Click on the Export button if the result is to be saved in a text file.
If not, click on the Close button and return to the previous window.
Click on the OK button to execute classification process.
Click on the Post Classification button.
In Weight and Threshold slide buttons define and input values.
Enter the name of the resulting image that will be generated in Image Name field and it will be displayed in the view/theme tree.
Supervised Classification using a Point Table
Select IMAGE PROCESSING → IMAGE PROCESSING → SUPERVISED CLASSIFICATION in the main menu.
In opened window select bands 1 and 3 and click on the right arrow button.
Click on the button Sample Acquisition.
Region of Interest window opens.
Click on the box Points Theme and choose the point view.
Click on the Next button.
Attribute: select the attribute that will be used to create classes.
Colors: choose the colors scheme to be used in the classification.
Click on the Apply button and classes will be automatically created.
Click on Next button.
Choose
a tolerance and click on the icon
to automatically create samples.
Note: The difference between this tool and the magic wand tool is that it uses the coordinates of points as the basis for polygons creation, not the coordinate of the mouse click. This tool only use the points that intersects with the visible area.
Click on the OK button.
Click on the Classification button.
Output Image Name: enter the name of the classified image.
Classifier: select the classification method.
Click on the OK button.
After the image classification, the user can be used to validate the classification to determine the reliability of the result. This tool is located on menu IMAGE PROCESSING → PALLETE.
Classified Image:
Layer Name: select the layer from classification result.
Sample Layer:
Layer Name: select the point layer in sample acquisition.
Class Column: select the attribute that indicates the classes.
Select the layer from classification result, the layer of points used in samples acquisition and the attribute that indicates the classes and make the association of classes.
Click on the OK button.
A report showing the number of computed samples, sampling factor, error rate and kappa index is generated.