Supervised Classification Example


Currently this page is out of date.

New image:

Use a band combination that best shows the features.

Contrast enhancement may be helpful


Select Train and Classify on Imagery analysis menu.


Define categories.  For each pick a name and color.


Select Training sets.  Tells program rectangles belong to a particular category.  Do this on a map at the highest possible resolution, so that you can accurately pick regions. 

At any point you can show the training sets on the map.

You can select multiple training sets for a category.

As you select training sets, the computer will compute the mean and standard deviation within each region for all the bands. In this case, note that water has very small standard deviations--the reflectances are all very similar.

If you get really large standard deviations, you categories are not homogeneous.


Classify.  The map will show each category in color.

This might work best on gray shading for a single band.  If you do it on a 3 band color image, it can be hard to tell the classification apart from the image.

You can then adjust the opacity of classification overlay with the slider bar. 

The results tab shows the classification details:
  • The bands used.
  • The standard deviation
  • The percentage of the area in each category in the classification.
Band Scattergram
  • Each band is shown in color.
  • Rectangle centered on the point that has the mean reflectance in each each band.  The  rectangle extends in each direction the number of standard deviations used for the classification.
  • In this case the orange category has a bigger standard deviation in band 2 along the x axis, compared to band 4 along the y axis.  This category may not be homogenous.
  • Points are classified in a category is they are within the box for that category, and if the distance to the category mean is less than the distance to any other category mean.
  • Classification is done in n-dimensional space, where n is the number of bands used for the classification.
The all scattergrams options shows you all the band histograms in a matrix.  This allows easy visualization of which band pairs discriminate your training areas.

The graphs on the principal diagonal how well single bands discriminate the training areas.

Principles of supervised classification.

Unsupervised classification (clustering)

Last revision 12/15/2012