Image Classification
Image classification is based on the principles of remote sensing. It is a means of converting spectral raster data into a different set of classifications that represents the surface type seen in an image. The surface type seen on an image varies depending on its wavelength or reflectance; we can have images that are in true color or infrared depending, on the band within which the image is located.
Image classification is important in identifying vegetation types, land use changes/types, mineral resources, to mention a few. In this exercise, an image of a study area showing 6 different land classes was classified using the Image Classification Tool. The land cover classes include forest areas, barren areas, wetland areas, cultivated areas, water areas and developed areas.
Description of Analysis Procedure
The appearance of raster image with six land cover classes was changed from a true color image to an infrared color image. This was done by changing the raster images symbology color bands. Image classification is more successful if image is in infrared color as opposed to being in true color. The image’s pixel values were also adjusted. The ‘training sample manager’ function form the image classification tool bar was used to create training sample polygons. These sample polygons were drawn around homogenous areas of all 6 different land covers. The first supervised classification was done using very few sample polygons per land cover class.
Original image showing land cover classes(click to enlarge)
Workflow of analysis procedure(click to enlarge)
training sample manager dialogue box(click to enlarge)
First set of training polygons(click to enlarge)
Map showing 1st set of image classification(click to enlarge)
Sample sets were then redefined to improve classification output. This was done by adding more sample polygons to the initial training samples created. These additional training sample polygons were added to all six land cover classes. These additional polygons were also run using the supervised classification option and another raster layer image was created showing better classification of the six different land cover types.
Workflow of reclassification on land cover classes(click to enlarge)
Additional training polygons added to land cover image(click to enlarge)
image classification after using redefined sample polygons(click to enlarge)
Application and Reflection
supervised classification can help one understand the components of an area. For example if an area is made up of mostly agricultural patterns, water or urban areas. This can be used in planning and assessments. It can also be used to view land use changes. I can use image classification in depicting previous and impervious surfaces which would aid in determining better planning development within that area. Also depending on the band used with the raster image, one can get more details on a certain land cover. For example, you can get more details on the different kinds of trees in a forest area, depending on the appearance of the original image.