Urban
Forest Mapping
Classifying land cover
The two land cover classification
approaches included a pixel-based
(per-pixel classification) approach using the maximum
likelihood
classifier in ERDAS Imagine and an
object-oriented approach with a nearest neighbor classifier
in eCognition. These classification approaches are compared
below.
The table below shows that
the overall accuracies of pixel-based and object-oriented
classifcation
methods were very similar. However, by comparing to some known areas’ land
cover type, we accuracy of the object-oriented
approach was a little higher, while the 3 by 3 majority
filtering
using the pixel-based approach in ERDAS is a little more
realistic.
Comparing the two classified maps using these two approaches,
we also found the object-oriented classification maps
look more homogeneous, whereas pixel-based maps look heterogeneous.
To some extent, the object-oriented map looks
more promising.
CITYgreen
Application >> |
