Unsupervised Image Classification (ISOdata classification)


Data:

  • October 13, 2020 Sentinel-2 image (eLC Link)
  • December 18, 2019 Sentinel-2 image (eLC Link)
  • Toolbox (extract the Toolbox.tbx and copy it to your working directory) (eLC Link)

Sentinel-2 Satellite Imagery:

  • Instructions for download from USGS Earth Explorer (eLC Link)
    • Data delivered as single-band rasters
  • Sentinel-2 band information
    • 10 meter bands (blue, green, red, near-infrared)

Source: https://sentinel.esa.int/web/sentinel/user-guides/sentinel-2-msi/resolutions/spatial

GENERATE A COMPOSITE IMAGE

  • Load bands 2, 3, 4, and 8 from the October 13, 2020 Sentinel-2
    • These are the single-raster bands containing the responses from the blue, green, red, and near-infrared portions of the spectrum
  • Generate a composite image using the COMPOSITE BANDS tool
    • Stack bands in this order: 2, 3, 4, 8
    • Save the output as S2_202013_b02030408.img
    • Don’t forget the “.img” extension
    • Explore this layer and try to find: water, forests, fields, urban, …

UNSUPERVISED CLASSIFICATION:

  • Perform an unsupervised classification of the October 13, 2020 image using the ISO CLUSTER UNSUPERVISED CLASSIFICATION tool
    • Number of classes: 8
    • Output classified raster: S2_202013_b02030408_iso8.img
    • Minimum class size: 20 (this means that the smallest grouping of contiguous cells is (20*10*10*3.28*3.28/43560) ~ 0.5 acres
    • Output signature file: S2_202013_b02030408_iso8_sig.gsg
  • Your classified image is a single-layer raster with values ranging from 1 to 8
  • Your task is to figure out what land cover type each class represents
    • Some classes like VALUE = 1 will be pure (water)
    • Some classes like VALUE = 6 will be a mixture of cover types (river and road)
    • VALUE = 2:
    • VALUE = 3:
    • VALUE = 4:
    • VALUE = 5:
    • VALUE = 7:
    • VALUE = 8:
  • Repeat the process using the December 18, 2019 image

What differences do you see in the classifications of the October 2020 and the December 2019 image?

Print Friendly, PDF & Email