Geography Reference
In-Depth Information
Fig. 5.33 Integrating the
ISODATA-clustering with
the supervised classification
algorithms in a so-called
hybrid-procedure
Training sites
selection from
ISODATA -clustering
Training sites selection
from field-work
Signature generation
Signature evaluation
Chosoing the supervised classification
algorithm
Thematic map/s
light color-tones bare areas and fallow on light soils; very dense irrigated trees
(especially Poplar) and dark water; and between vine and sugar beet.
5.7.1.2 Supervised Classification
The Multi Stage Classification Approach
The decision tree classifier is a hierarchically based classification method which
compares data with a variety of well-chosen features. The selection of these
features is controlled by an estimation of the spectral distributions or separability
of the classes. There is no commonly confirmed formula and each decision tree or
set of rules must be constructed by a specialist. If a decision tree presents just two
outputs at each stage, then it will be named a Binary Decision Tree Classifier
(BDTC). This procedure was applied in many cases due to its flexible character-
istics. In agriculture applications, the rules of a decision tree are acquired via
analyzing the specific attributes (understanding the various spectral responses, the
agricultural calendar, etc.) of different crop types (Chen et al. 2008 ). Figures 5.34 ,
5.35 illustrate the steps applied in the multi stage classification approach to gen-
erating the classification results of the four region study area.
Training sites and testing areas are fulfilled separately and compared to satellite
images for each classification algorithm after applying the masking-process. This
is done because, for example, the mask that represents the distribution of the
irrigated agriculture (separation and classification of the irrigated agriculture areas
and the rain-fed agriculture areas) using the MLC-algorithm covers areas differing
Search WWH ::




Custom Search