5.3-Image Classification

Prepared by:
George McLeod
With support from:
NSF DUE-0903270
in partnership with:
Geospatial Technician Education Through Virginia’s Community Colleges (GTEVCC)
Satellite images capture light by sampling over
predetermined wavelength ranges which are
referred to as “bands” or “channels”
To extract additional information from digital
images use image processing techniques such
False Color Composites,
Image Ratios, and
Classification (Supervised and Unsupervised)
 Unsupervised
Image source: Dr. Ryan Jenson
Requires minimal amount of input from user
Based solely on numerical information in the data
Matched by the analyst to information classes
Pixels with similar digital numbers are grouped
together into spectral classes using statistical procedures
such as cluster analysis ISODATA
Iterative Self-Organizing Data Analysis Technique Automated spectral clustering
User then identifies which class membership for each
User selects area in image that represent each
unique class (“Training” sites)
Pixel values for each band are recorded for class
sample set
Computer matches rest of pixels to user defined
classes based on closest distance in multidimensional image space
This outputs a classified image
Image source: Dr. Ryan Jenson
Used in supervised classification
Homogeneous areas of land cover
Information derived from:
field studies,
 thematic maps,
 other areas of knowledge
Each site should have at least 10 times
‘n’ number of pixels, where n is equal
to the number of bands used in the
Map digitizing
On-screen digitizing
Minimum Distance to the Means
Maximum Likelihood
The data points for DNs from two bands are dots; the mean for each clustered data set
are the squares. For point 1, an unknown, the shortest straight-line distance to the
several means is to the class "heather". Point 1, then, is assigned to this category. Point 2
is slightly closer to the "soil" category but lies within the edge of the "urban" spread.
Here, the classification seems ambiguous. By the minimum distance rule, it would go to
"soil" but this may be erroneous ("urban" would have been a greater likelihood). Point 3
is not near any of the class DN clusters, but is about equidistance between "urban",
"water", "forest", and "heather". If one plays the odds, "urban" is just a tad closer to 3;
but this situation indicates how misclassification might occur.
Not shown is the fact that inside each ellipse are contours that indicate the degree
of probability. Associated with each ellipse is a separate plot that expresses a
statistical surface (bell-shaped in three dimensions) called probability density
functions. Using these functions, which relate to the contours, a likelihood that any
unknown point U is most probably associated with some one ellipse is
determined. A Bayesian Classifier is a special case in which the likely occurrence of
each class (common to rare) is assessed and integrated into the decision making.
Growth or shrinkage of urban areas
Deforestation of tropic areas
Fire and burn damage
Damage done by hurricanes,
earthquakes, and tornados
Autralis 1999
Autralis 2002
http://glovis.usgs.gov/ USGS data viewer
er/ USGS New Earth Explorer
Global Landcover facility at the University of
jsp NOAA NexRad Radar Data

similar documents