Hi class,
Today, we’ll look at how to randomly sample from an image. The idea is that you want to assess how well you did in classifying an image. One way is to select a bunch of locations and see if you got those locations correct or not. The problem is that you need random samples, and those are hard to do by hand.
So we’ll look at several ways to randomly select sample points. NOAA has an Add-in that is in the ARCGis Gallery called ‘The Sampling Design Tool’. It is useful if you have a polygon to select from (like town boundaries, or a polygon class map like those distributed by MassGIS. Installing the Add-in is a little tricky, but I’ll walk through it today.
- Here is the Add-in file: https://udrive.oit.umass.edu/xythoswfs/webui/_xy-11979220_1-t_H075j9IR
- Here is the Documentation for the Add-in: https://udrive.oit.umass.edu/xythoswfs/webui/_xy-11979221_1-t_6I8hm788
For rasters, you do it by creating a uniform random grid, and use the setnull statement to select the number of random points you want. You may then want to convert that raster of selected cells to a point file that is easier to work with. I’ll go over this in class too. You might need to do this for each class; this is a statified random sample.
Once you have a random point file, you can use the points to store the class that you map says that point is, and the class that you (or you ground truth) say it is. You then have a way to calculate the %error in your map byt class.
Jack