In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
390
interpretation of the decorrelation due to too many possible
human activities.
The paper is organized as follows. Firstly, we point out main
ideas of the CCD technique. Then we describe our study area
together with the data, as well as our method for analysis of
CCD results in order to obtain information about potential
human-induced scene changes, such as building up areas of
hard surface on a terrain that was previously a part of a desert.
In order to cover a wide range of applications and situations, the
proposed method is intentionally designed so that it does not
need any specific knowledge source about the terrain. Still,
once CCD results are obtained, various knowledge sources can
be taken into account in order to improve the final
interpretation. These knowledge sources can be related to the
sensors (operational principles and experience) or to the
situation at hand (context - terrain type, land-use, historical
background etc.). A way to include them in the reasoning
process is discussed here too. In addition, in order to further
improve the quality of the obtained output, usefulness of
applying a spatial regularization technique is tested as well.
Finally, we validate the obtained results and propose ways to
continue the work.
2. ON COHERENT CHANGE DETECTION
Starting points for CCD processing are the complex images of
an interferometric SAR image pair (two images from
approximately the same geometry collected at two different
times). Namely, CCD uses the correlation between them and
provides information on the stability of the target (Matikainen,
Hyyppa and Engdahl, 2006; Price and Stacy, 2006). For
instance, forests have a low coherence value, which means that
such a type of target has changed much from one image to
another (thus in time corresponding to the collection of these
two images), while urban areas typically have high coherence
values even between image pairs separated by several years
(Luckman and Grey, 2003; Matikainen, Hyyppa and Engdahl,
2006).
The coherence can be expressed as the product of five dominant
contributions, as shown in (Zebker and Villasenor, 1992;
Preiss, Gray and Stacy, 2006):
1) the relative backscatter signal to radar receiver noise ratio in
the interferometric image pair,
2) the volume decorrelation,
3) the baseline decorrelation,
4) the decorrelation related to mismatch between the coherent
acquisition apertures and image-formation processing stages
used to produce the primary and repeat-pass imagery, and
5) the decorrelation in the scene over the repeat-pass time
interval (temporal decorrelation); this type of decorrelation is
the one we are interested in, as it is determined by various
sources of scene change, such as environmental effects
(moisture changes, atmospheric effects) or man-made
disturbances.
The value of the product of the first four contributions
mentioned above is close to one if the repeat-pass imaging
geometry is designed carefully and if interferometric processing
steps are performed, such as compensation for aperture and
processor mismatch as well as image registration (Preiss, Gray
and Stacy, 2006). Therefore, under such conditions, the
coherence of the scene image reflects the true scene coherence
over the repeat-pass interval. For the data used in this paper,
these conditions are fulfilled.
Therefore, a starting point in our analysis is the fact that in this
type of scene/terrain, low coherence values can refer to the
moving ground (sand), which is perturbed all the time, while
high coherence values can be related to a hard surface (e.g.
concrete).
3. STUDY AREA AND THE DATA USED
The study area is an airfield that is located in a desert part of
Israel. This type of terrain is a good start for analysing the
usefulness of the CCD method for detecting potential human
activities due to low soil moisture as well as low vegetation.
As far as the data are concerned, four ALOS PALSAR (Phased
Array type L-band SAR) images (Rosenqvist, Shimada and
Watanabe, 2004) of the scene in the descending mode are
processed, corresponding to four different dates of acquisition:
15 November 2007, 15 February 2008, 1 April 2008 and 17
May 2008. Starting from the Single Look Complex data, three
CCD results are obtained using the following pairs: 15
November 2007 and 15 February 2008 (period 1), 15 February
2008 and 1 April 2008 (period 2), as well as 1 April 2008 and
17 May 2008 (period 3). As an example of the three obtained
CCD results, Fig. 1 contains the CCD result for the images of
period 1. The images of the other two periods can be found in
(Milisavljevic, Closson and Bloch, 2010). The coherence
images are generated using the module Insar of Erdas Imagine
2010 platform.
Figure 1. CCD result for the images acquired on 15 November
2007 and 15 February 2008 (period 1)
Figure 2. Amplitude of the SAR data acquired on 15 November
2007
In order to illustrate the usefulness of the SAR phase
information used to obtain the CCD results, Figures 2 and 3