Research Methodology - Historical Land Use/Land Cover Classification Using Remote Sensing - page 143

Fig. 5.58 The general classes classification, irrigation mapping, and agriculture classification
results that were produced based on the spatial extent of each irrigation project
are often found in mountain areas in Syria, but food-trees (cultivated plantations)
are essentially located in plain areas, with sparse houses and a low population
density. So, expert knowledge can be performed based on the relationships
between the high factor and the agricultural conditions to separate food trees from
forested areas. Classical pixel-based classification methods often lead to ''salt and
pepper'' effects in final classification results maps, caused by the isolated pixels of
some classes within another dominant class. That is due to the complexity of
biophysical environments, which potentially decrease the classification accuracy.
It is more logical to join these isolated pixels to the dominant class that they are
first assigned to. A suitable enhancing filter applied after the classification process
on the produced thematic map will not only ''clean up'' the map and make it
visually less noisy, but also increase the classification accuracy.
To improve the classification results, the majority/mode filter in ENVI, v. 4.6
was used as a post-classification procedure. This procedure is a low-pass filter that
reduces the created effects and noises from the classification process, where it
replaces the isolated pixels by whatever value constitutes the majority in their
neighborhood. It could be regarded as a kind of post-classification spatial inte-
gration. This filter is simple, where it smoothes a thematic map without any
numerical operations (Liu and Mason 2009 ). The classification results (the the-
matic maps) were filtered using a 3 9 3 majority filter window, followed by a