Geography Reference
In-Depth Information
In order to overcome the limitations of the first technique, one can use tech-
niques based on a supervised classification of multi-temporal images: Direct
Multi-data Classification (DMC), Neural networks (NNs) (Bishop
1995
),
Knowledge-Based Systems (KBS), Support Vector Machines (SVMs) (Vapnik
1998
), Post-Classification Comparison (PCC) (Del Frate et al.
2005
).
The fame of the above techniques may be because they can be freely applied on
available created single date classifications, where they are based on separate
single-date classifications whose results are later compared with the result of the
second independently classified image (Weismiller et al.
1977
). This simple
technique includes: (1) producing the classified image based on the classification
process; and (2) assessment the occurred changes based on the principle of
identifying the areas of change as pixel per pixel differences in class membership
(Castelli et al.
1999
).
Advantages: (1) the ability to clearly identify the kinds of occurred LULC-con-
versions; (2) the robustness to the various atmospheric and light conditions at the two
recording times (Bruzzone and Fernàndez-Prieto
2000
); (3) where the two datasets/
imagery are separately classified, so it is not needed to normalize these data (Singh
1989
); (4) it is more flexible than those used the comparison of multi-temporal raw
data; (5) it allows one to make change detection also by using different sensors and/or
multi-source data at two times; and (6) the possibility in entering several modifica-
tions on the used classifier in classification process (e.g., contextual information as
using the texture of an image) would increase the change detection mapping accuracy
(Pacifici et al.
2007
). Also, the new image classification algorithms, other than the
traditional MLC, can be used to increase both accuracy and effectiveness. Disad-
vantages: (1) requires more human supervision for classifying the images; (2) despite
its potential, this category is not relevant to quick change detection, because user
supervision is required to pre-classify the images; (3) limitations also include cost in
terms of money and implementation time, and generated errors from classification of
imagery, where the generation of a suitable training set has the two drawbacks, i.e.
the difficulty and the high cost (Bruzzone and Fernàndez-Prieto
2000
); and (4)
finally, the accuracy of the change thematic map will be equal to the accuracies of
each individual classification for each date.
2.4.2 Change Detection in Arid- and Semi-Arid-
Environments
Approximately 50 % of the total surface areas of the world are arid and/or semi-
arid regions (Meadows and Hoffman
2002
). Arid and semi-arid areas feature
irregular, low precipitation, dry ecosystems, and have a limited sustained eco-
nomical potential (Adam et al.
1978
). Because of the sensitive nature of these
areas, it may only require a small amount of turbulence to cause clear changes
within the environment (Okin et al.
2001
). As a result, remote sensing is quickly
becoming an essential tool to use in the study of these areas (Zhou et al.
1998
).
Search WWH ::
Custom Search