Supervised classification involves manually choosing representative areas ("signatures") in order to tell the program to classify all areas with similar pixel values to the same class. Unsupervised classification is the opposite, in that the program sorts through and assigns a set number of categories based on pixel values. After this, the categories are grouped manually. For this map of Germantown, Maryland, the categories were determined via supervised classification methods from a image with a band combo of 5, 4, 3 (Red, Green, Blue). This band combo was chosen because it differentiated the vegetation from the urban or fallow areas by including the Near Infrared band (Band 5), which is reflected by healthy vegetation and shows as red areas in the image. Using this image, signature areas were chosen for each of the 8 categories of land use, such as the lake, urban areas, and deciduous forested areas. The image was then classified based on these chosen signatures and the new image above was created. Colors were chosen to highlight particular areas, such as magenta for urban/residential areas, dark green for deciduous forests, and blue for water. The distance image in the top right of the map indicates potential areas of error (bright areas = higher likelihood of error). Only a small portion or the image shows to have a higher likelihood of error, so this method is viable to determine land use. This is good news, as both supervised and unsupervised methods of image classification save a lot of time for land use classification, as opposed to digitizing each area of land by hand! This image also shows how complex land use classification can be in a relatively small area, as seen in the high variation of color (category) and shapes (area).
Comments
Post a Comment