Unsupervised Image Classification (Clustering)

Unsupervised classification attempts to find clusters in n-dimensional space based on the reflectance values, and assigns to those clusters to a group.  Hopefully there will be a reason for the cluster, which will correspond to a particular category, but that can only be interpreted by the user.  The selected colors are random.  On one scene the water might be blue, and on the next it will be vegetation.

Cluster for 6 band Landsat TM image of  Hanging Rock quadrangle.	Legend.  This will be in a .VAT.DBF file, which you can edit with descriptive names for clusters and change color assignments.

Pick option

• Image analysis, Multi-grids raster analysis menu.
•   Open hyperspectral imagery
• Set clustering options.  Pick parameters (all defaults should work).

Clustering will be based on a limited subsample of points.  They will be selected from the current screen, so you should subset  the satellite map to cover just the area of interest.  The entire image can be classified, but the results will be most valid in the map region.

Manual, equivalent options

• Create DBF file on Image analysis menu.
• Cluster on Stats button popup menu of table display window

You can also use the Orfeo toobox for unsupervised classification.

Clustering Delphi code from Fred Edberg; 11/30/02 (fedberg@teleport.com) and uses K mean clustering.

Last revision 7/1/2020