I am doing land cover classification with an hyperspectral image of Santa Barbara, CA. The spatial resolution is 6.4 meters but I couldn't find any high spatial resolution land truth data for validation. Someone recommends me that I can generate random points in the classification map and see if those points are correctly classified by using Google Earth. However, I don't know how to do that.
Should I download image from Google Earth and put it into ArcGIS? Or should I put the randomly generated points and their classified types into Google Earth?
Best Answer
You're going to be doing accuracy assessment. You want to bring the data for validation (Google Earth) into ArcMap. Then you can "truth" the classification for each of your random points in a shapefile. Afterwards you can sample those same points in your classified image and compute your error matrix.