Lab 5: Unsupervised and Supervised Classification

 

This map displays land use in Germantown, Maryland from a supervised classification image using ERDAS Imagine. In this weeks lab, we learned the steps of unsupervised and supervised image classification giving unique values to all the pixels in an image. For our map deliverable shown above, I performed supervised classification and merged the pixels into 8 distinct classes - urban/residential, grasses, deciduous forest, mixed forest, fallow field, agriculture, water, and roads. For this lab, we were given X and Y coordinates of 3 urban/residential points, 2 fallow field points, 4 agricultural points, 1 deciduous forest point, 1 grasses point, and 1 mixed forest point. We were instructed to also collect our own spectral signatures for a roads class and a water class. I created my own spectral signatures by opening an Annotation Layer and using the "Seed" method. I used the Inquire (Legacy) tool and panned to a road feature and a water feature on the image. In the Region Growing Properties window, I clicked the At Inquire button which created a polygon within the feature. Then I selected the Create New Signatures from AOI icon in the Signature Editor window to create signatures for both roads and water. Once all my classifications were created, I merged them using the Recode tool. Within the Recode Table, I changed the numbers in the value columns for each new classification. For example, instead of all of the agriculture classes having different values 1-4, they were all given a value of "1." Once this was done, I had to go through my new attribute table and give each class the appropriate name and a color to represent the feature on the image. I also included an Area column in hectares which is shown in the legend on my layout.

There is also a Distance File inset map within my layout. Distance files show bright areas in an image to represent areas and features that are spectrally different than other ones within their spectral class. So, if a pixel is brighter in the image, it may be in the wrong classification.