Geography Reference
In-Depth Information
2.4.1 Change Detection Techniques
There are numerous change detection approaches applied on remotely sensed data,
as a result of increasing versatility in processing digital data and increasing
computing power (Pacifici et al.
2007
). Generally applied approaches are: image
differencing; and image rationing (Singh
1989
). Some of the proposed supervised
and unsupervised approaches in the literature are: write function memory inser-
tion; image algebra; multiple-date composite; post-classification comparison;
image differencing; image rationing; change vector analysis; etc. (Nelson
1983
;
Singh
1989
; Sohl et al.
2004
). Expert systems and neural networks were too used
in change detection (Seto and Liu
2003
). These approaches use multi-date imagery
from multi- and hyper-spectral sensors, so that alterations, in feature or phe-
nomena, be accurately recognized, measured and if needed observed (Jensen
2007
), each of which could be spatially, spectrally, or temporally controlled (Lu
et al.
2003a
). Figure
2.7
illustrates how the various frequently used techniques are
located in this framework.
Returning to Fig.
2.7
, change detection techniques can be separated into two
general groups, depending on whether the technique needs classification before or
after change detection process.
1. Techniques which first detect change and then assign classes (e.g., image dif-
ferencing or PCA)-Unsupervised Approach- Pre-classification method.
(a)
MLC
Fuzzy
Bayesian
Neural networks
Texture
Object-based
Others
(I)
Post-classification
change comparison
Classifid Images
Unclassified
Images
(b)
Original pixel
Direct
(d)
Differencing
Change -vector
Multidate
comparison
Regression
Classification
of „from-to“
classes
(II)
(c)
Ratio
Principal component
Texturaltransformed
Chi-squared
transformed
Others
Indirect
Fig. 2.7
A framework for classifying change detection methods (Source modified from Lam
2008
)
Search WWH ::
Custom Search