In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
391
contain amplitude information of the SAR data pair
corresponding to the two dates from Figure 1. Visually, there
are no significant changes in amplitude values between two
images. We have tested two amplitude change detection
techniques, i.e., image differencing (Rosin, 1998) and image
rationing (Lu et al., 2004), which confirm that no significant
change in amplitude occurs between these two time intervals.
However, the phase-based change detection result shown in
Figure 1 demonstrates that changes occur in that period, which
justifies the need for using this part of SAR information in this
type of applications.
Figure 3. Amplitude of the SAR data acquired on 15 February
2008
4. OUR METHOD FOR DETECTION OF HUMAN
ACTIVITIES
This method is introduced in (Milisavljevic, Closson and Bloch,
2010) and we outline it here. Firstly, a noise removal method
using median filtering is applied to each of the three CCD
results. Then, each of the three CCD results is divided in three
classes, based on coherence value:
if a CCD pixel is dark (low coherence), it is labelled
as 1;
if it has a medium value, it is labelled as 2,
if it has a high coherence value, it is labelled as 3.
As an illustration, Fig. 4 contains the obtained result in case of
the CCD image from Fig. 1. Green colour corresponds to label
1 (so, low coherence), white colour is used for label 2 (medium
coherence), and red is used for label 3.
Figure 4. Classification result of Figure 1
In the following, results shown in Fig. 4 are referred to as A, the
classification results obtained on the CCD pair from period 2 -
B, and those obtained on the CCD pair from period 3 - C.
In a next step, we compare the results A and B, in order to see
whether within the first periods of time the coherence values
were similar (i.e., there was no change in the soil disturbance
from one period to another) or not. The comparison of values A
and B (AB) in each pixel can result in one of the following
labels: 11 (low coherence in CCD results in period 1 and in
period 2), 12 (low coherence in CCD results in period 1 and
medium in period 2), 13 (change from low to high coherence,
i.e., possible increased human activity), 21, 22, 23, 31, 32, 33.
Since the threshold between low and medium, on the one hand,
and medium and high, on the other hand, is crisp and since
these changes are not as radical as changes from very low to
very high coherence (or the opposite), we regroup these AB
values so that
- 11, 12 and 21 are grouped together, labelled as 1,
- 22 is labelled as 2,
- 23, 32 and 33 represent one group, labelled as 3,
-13 remains 13,
- 31 remains 31.
Figure 5. Comparison of values A and B: 1 - green, 2 - white, 3
- red, 13 - blue, 31 - purple
The obtained result is illustrated in Fig. 5. Similar reasoning is
applied in comparing the results B and C.
Figure 6. White pixels show where AB is 13 and BC is 3.
Region within the blue ellipse is zoomed in Figure 7