As the title suggests, this module's exercise focused on the concepts and methods required to carry out unsupervised and supervised classification. While I gained experience completing a classification of both types using instruction from the lab assignment, the key focus was to complete a supervised classification on my own using a fresh image.
For this assignment, I was given an aerial photo of Germantown, MD and tasked to create a land use map of the area using supervised classification. All of my data processing for this assignment was completed in ERDAS Imagine and then used to create a map layout in ArcGIS Pro.
I was given coordinates for a few land use types: urban/residential, grasses, deciduous forest, mixed forest, fallow field, and agriculture. For each set of coordinates I set a cursor using the Inquire (Legacy) tool in ERDAS Imagine. At each cursor, I created an AOI (Area of Interest) using the Grow tool. Each AOI was carefully created by adjusting the Neighborhood type and Spectral Euclidean Distance value. Once each AOI was deemed acceptable, I created a spectral signature by way of the Create New Signature(s) feature within the Signature Editor window. This is known as signature creation from "seed". I also needed to classify two feature types without predestined coordinates, water and roads, of which I applied two different methods. For water, I simply zoomed into two large bodies of water in the image and drew polygon AOI's using the Polygon tool in the Drawing tab. For roads, I recalled that in the lab it was mentioned that narrow features required an eight-way neighborhood setting to create a proper signature. I elected to place an Inquire Cursor at two different segments of highway and create my AOI's from "seed" once again. Once I had compiled my signatures into one file, I then needed to clear out any spectral confusion. To do so, I looked at histograms comparing two features for each band (1-6) and found the bands where histograms did not overlap. I also used a plot comparing each band of every feature to see where the bands had the most separability. I found that bands 4, 5, and 6 were the best for classifying my image. I would need to set these bands to my signatures to produce a proper supervised classification, so I changed the colors in the Signature Editor using Edit>Colors>Approximate True Colors. I set the colors to Red-4, Green-5, Blue-6. With the colors set, I ran a supervised classification in the the Signature Editor with Maximum Likelihood as the Parametric Rule. I also created a spectral distance image from the classification that would display areas based on how accurate the classifications were likely to be. With the supervised classification image created, I focused next on merging my classes to reduce each use classification to one code. I did so by way of Setup Recode in Raster>Thematic>Recode. The new, final image now had eight unique land use classifications displayed thematically by color. Before creating the final map layout in ArcGIS Pro, I added a field to the recoded image in ERDAS Imagine to calculate area for each classification by acre. The final layout with the land use image via supervised classification, spectral distance image, and all necessary map elements is posted below.
Comments
Post a Comment