Research Methodology - Historical Land Use/Land Cover Classification Using Remote Sensing - page 114

Fig. 5.35 The flow-chart of
the applied multi stage
classification approach in this
study
primarily non-vegetated areas class) will during the automated classification
process, be automatically fused with classes within the second general level (i.e.,
primarily non-vegetated areas), thus creating accumulated error/s in the classifi-
cation process. Part of the resolution of this problem is to re-classify the wrongly-
classified areas when moving to the next stage or level of classification, as long as
there are lands representing the wrong classified class within the various levels of
the multi stage hierarchical classification approach. When we return to the
example of the 1,000 ha, which were classified as non-vegetated areas and con-
sider this at the second level of classification, instead of training sites that represent
only the two classes of this level (i.e., terrestrial and aquatic), extra training sites
will be selected that represent the 1,000 ha area/s. If this 1,000 ha were completely
separated and classified within the second level, it will be appropriate. Otherwise,
if a further part of this area, such as 100 ha would appear within the next level,
again extra training sites would be trained to represent this class in the classifi-
cation process.