Feb 23, 2016
Supervised Classification in Imagine
Concept: Supervised Classification
• The goal of this exercise is to use the spectral signatures of different land covers to create a supervised classification.
• We will attempt to map the same land cover classes covered in the last exercise.
Geospatial data fundamentals
• Geospatial information types:– Raster: “images” composed of “pixels”– Vector: points, lines, polygons (“shapes”)
• Raster data types:– Continuous
• Single attribute (panchormatic = “black & white”)• Multiple attribute (multi-spectral = “color”)
– Discrete:• Quantized continuous• Categorical
Continuous vs. Categorical• “Feature space” – set of all attributes
describing an object.• Student feature space:
– Height (continuous)– Weight (continuous)– Hair color (weirdly continuous)– SSN (categorical) -doesn’t make sense to take an “average” SSN
• GIS attributes– Continuous – How warm? How bright? How much photosynthesis?
What’s the mean population density? Crime rate per 100,000?
– Discrete – what type of land cover? In which country is it located?
Categorical – Land Cover
How to ClassifyMultispectral Images
RGB: decomposing images
RGB red green blueClass Red Green Blue
Tomato Bright Very dark Very dark
Background Very dark Kinda dark Medium
Green pepper Kinda dark Medium Very dark
Yellow pepper Very bright Kinda bright Very dark
Orange pepper Very bright Kinda dark Very dark
Garlic Very bright Very bright Very bright
Bowl Medium Medium Medium
RGB: spectral signaturesClass Red Green Blue
Tomato Bright Very dark Very dark
Background Very dark Kinda dark Medium
Green pepper Kinda dark Medium Very dark
Yellow pepper Very bright Kinda bright Very dark
Orange pepper Very bright Kinda dark Very dark
Garlic Very bright Very bright Very bright
Bowl Medium Medium Medium
BrightVery bright
Kinda brightMedium
Kinda darkDark
Very darkRed Green Blue
Supervised Classification• Very widely used method of
extracting thematic information
• Use multispectral (and other) information
• Separate different land cover classes based on spectral response, texture, ….
• i.e. separability in “feature space”
Supervised classification
• Want to separate clusters in feature space
• E.g. 2 channels of information• Are all clusters separate?
10
Tools• Identify spectral signatures of different land
cover types using tools within Imagine: – Signature editor
• Alarm feature• Signature editor statistics
– Areas of interest (AOI’s)• AOI tool
– Supervised classifier (“maximum likelihood”)– Raster Attribute Editor
Supervised Landsat Classification• Open “germtm.img” from the data folder (RGB=5,4,3)
AOI tool• Open AOI -> AOI Tool• Open AOI -> create polygons around training sites
Signature Editor• Have the Classification menu open• Utility -> inquire box and locate given x,y coordinates
Classify the image• The goal of this exercise is to use the spectral signatures
collected in the previous to classify the reflectance image: germtm.img (open this in a viewer, r,g,b->5,4,3)
• Open the previous AOI for germtm.img from the “spectral signatures” exercise. In the viewer menu bar: File-> Open-> AOI Layer to see the training polygons.
Input image with AOI’s
Classify the image• In the Imagine Toolbar, click on the “ Classifier”
button to get the Classifier menu; click on “Supervised Classification”
Classify the Image• Input file: “germtm.img”• Signature file:
“germtm.sig” (from before) • Output file:
“germtm_sup.img” (in results folder for the current exercise)
• Parametric rule: Maximum Likelihood.
• Click “Okay”
Classify the image• Open classified image in the same viewer as the input
image (deselect “clear display”)• Select the “Arrange Layers” icon in the Viewer and move
the AOI layer to the bottom to hide the polygons (“Apply”).
Classify the image• Swipe between the input and classified image. Move around and swipe between different areas to observe the results.
Refine the classification• From the viewer window, select Raster->Attributes
Refine Classification• In the raster attributes editor, click column properties icon to edit the location and size of the columns
in the editor.• Move the “Class Names” column heading to the “top” and change it’s wide to 10 (makes it leftmost
column).• Move the “color” heading “up” just below “Class Names”
Refine Classification• Make various “classes” red to evaluate it’s
accuracy (good urban classification)
Refine Classification• Make various “classes” red to evaluate it’s
accuracy (questionable urban classification)
Refine the classification• One solution: delete the problem class in the signature file (iterate for
all classes).• Rerun classification with updated signatures.
Compare to Unsupervised classification
• Open “xiso.img” from the previous exercise (DO NOT CLEAR DISPLAY
• Use swipe to make a quantitiative comparison with germtm_sup.img
• Using the raster attributes editor, compute the number of pixels in each class for both the unsupervised and supervised classification