Geography Reference
In-Depth Information
Fig. 5.32 The spatial extent of the four administrative regions (Athawra, Al-Jurnia, Ain Eysa
and Menbij)
5.7 The Choice and Evaluation of the Optimized Method
of Automated Classification
A comparative study of different remotely sensed data classification algorithms is
often conducted to find the optimized classification result for a specific study (Lu
and Weng 2007 ). Many considerations, such as: spatial resolution of the remotely
sensed data (how many meters?); spectral resolution (how many bands?); different
sources of data (which sensors?); a classification system (which scheme?); and
training samples (which statistical distribution?), must be taken into account when
selecting a classification algorithm for use. Each algorithm has its merits and
deficits. So, the issue of which classification algorithm is more fit for a specific
study in a specific area is not easy to answer. And, diverse classification results
could be obtained depending on the classifier(s) chosen.
Experiments were conducted on the testing study area to determine the suitable
algorithm to use on the entire ERB study area. The supervised classification
algorithms tested were: MLC: Maximum Likelihood Classifier, NN: Neural Net-
work, and SVM: Support Vector Machine (Fig. 5.32 ). Two classification proce-
dures were also applied: (1) one stage classification approach; and (2) multi stage
classification approach, to produce land cover maps.
Search WWH ::




Custom Search