Unsupervised and Supervised Classification

In this week's lab, we worked on completing unsupervised and supervised classifications of aerial imagery. Unsupervised classifications are pretty much when the software tries to figure things out on it's own. To create this, I just told the software how many classes to find and then had to sort through them to recode them into the appropriate classes I desired. The imagery we worked on feature a portion of UWF's campus. It was a bit tricky as some of the vegetation was similar in color to the buildings. It might have been easier if the imagery was collected in the spring or summer when it was a little less brown.

Supervised classifications are completed by giving the software some examples of classes and then letting it figure out which pixels fall into which class. For this classification, I found spectral signatures as examples of each class. These are created by either drawing an area of interest (AOI) by hand in a known area or using a pixel at a specific coordinate and expanding on this using a grow tool. We also had to figure out the best band combination to highlight the differences between the classes. Needless to say, I feel like the supervised classification worked much better, but it did take a little more time and knowledge on the front end. 

Here is the supervised classification I created this week, exploring the land use of Germantown, Maryland.