Supervised Train and Classify

Steps:

• Open multi-grids for raster analysis on the Imagery analysis menu.  The operation will be on the grids; you cannot use the original image.
• On  Multi-grids raster analysis pick Supervised classification.
• Training set: in a new "sup_class" directory under the directory with the data, with the name  L8_2017_282_sup_class_NNN.dbf where NNN is a sequential number so you can try multiple different classification.
  • New training set;  create a new classification.
  • Open training set: select an existing classification
• Add training class
• New training points, in the Current Class
  • Train points: double click for points
  • Train box: outline a region.  You set the maximum number of points to add; if the box is too big, points will be subsampled.
• Pick the classification parameters:
  • Mean/Std will be the best option; you can experiment with others
  • Std Dev determines how many standard deviations from the mean the point must lie within in each band
  • Bands must be in box allows some flexibility if a pixel is far away from the band mean in just a few bands.
  • Classification distance power determines the distance function if a pixel is within the limits for more than one band's limits, in which case it will use the closest.
• Classify

Some history to explain the disabled options.  We are working to reenable them, but we aren't sure many of them are very useful since they were established for experimental options that did not really pan out.

You must have the following data sets:

• Single TIFF image with all bands, or for Landsat, each band in a file according to USGS standard naming conventions.
• MICRODEM format DEMs, one for each band, in "multi_grids" directory under the TIFF image. If the checkbox on Imagery tab of options form is set, multiband imagery will be converted to grids automatically.  This allows much better handling of statistics for imagery with > 8 bits.

On the options form, Imagery tab, you can select whether to do analysis with the original data or cloud removed data.

Principles of supervised classification.

Supervised Classification Example

Last revision 12/18/2017