Unsupervised Classification

Photo Interpretation & Remote Sensing - Mod 9

This week's less in Unsupervised Classification was pretty straightforward and simple, though somewhat time consuming.  Most of the work was done in Erdas Imagine starting with running the Unsupervised Classification tool on a high resolution aerial photograph of the UWF campus.  This

resulted in a thematic raster that allowed us to simplify the image into fewer classes by selecting the pixels and changing them.  Our task was originally to classify the image into four categories; Trees, Grass, Building/Roads, and Shadows.  This seemed simple enough until some of the clusters affected multiple features.  In order to deal with this problem a Mixed class was added as well and anything that affected more than one feature could be added to that.  Sometimes the affect on a second feature was so limited it made more sense to keep that class as the first feature.  Once we had the image classified we merged the classes so we had only the five specified.

Next we added fields to the attribute table for Class Name and Area.  This allowed us to calculate the area for each of the classes so we could determine the percentage of permeable verses impermeable surfaces.  Permeable surfaces include Grass and Trees, impermeable, Buildings\Roads.  The other two classes included a combination of both, so before we could make that calculation we first had to calculate what percentage of each of those classes were made of each surface.