Pakzad, Kian
can be found on an area which was used before for peat extraction and if the system finds homogeneous areas, the
system expects that there are no more ditches in order to dry the ground (see section 3) and the requirement for the state
area of regeneration heather state is correct (heather state is not defined as an area on which already heather is growing
but as an area which is changing toward the state).
Every link in the s/ate transition diagram has a priority, which describes the probability of the state transitions. This
value affects the order in which the different state transition hypotheses will be verified. As shown in Fig.7 every state
has a transition link back to itself. This is in each case the link with the highest probability. Consequently for every new
epoch this is the first transition concept to be verified.
The semantic net we used for the multitemporal interpretation is a refinement of the net in Fig.3 and is depicted in
Fig.8. It contains the separation into the states area of degeneration, area of regeneration birch state and heather state
with their obligatory parts. The probabilities depend also on the time differences of two epochs.
For a part of the test area the multitemporal interpretation was performed. As input data three grayscale aerial images
= taken at the epochs 1969,
1975 and 1988 were used.
The first image from 1969
was the basis of the initial
segmentation, which divided
the image into three
segments. For every segment
the system determined the
state transition. Fig.9 shows
the result.
The reduction of the search
space for the possible
Successor states leads to a
correct interpretation of the
segments. For segment 1 a
transition from area of peat
extraction to area of
regeneration heather state is
stated using the knowledge
about the previous land use of
1969 1988 the segment and about the
Ea EE mn missing parallel lines.
pe . rea of area of area of forest Without using this prior
cr oi Segersraton | information the system could
not distinguish the states
agriculturally used area and
area of regeneration heather state in grayscale images because both states are also characterized by a high
homogeneity. For segment 2 the land use state changed from area of peat extraction to forest although there is no direct
state transition between them represented in the state transition diagram. Due to the rather long time interval of 13
years between epoch t+1 and t+2 the regeneration states were not observed. Using the knowledge about the mean
transition times the system also generated the hypothesis for forest which was verified successfully for segment 2. The
transition state between 1975 and 1988 was with higher probability the state area of regeneration birch state, because
this is the direct way between the two states area of peat extraction and forest. But it is also possible that the area
changed first to the state area of regeneration heather state and after that to the other two states.
Figure 9. Result of multitemporal interpretation
The exploitation of the state transition diagram is reached through an extension of the semantic net in Fig.8. The
semantic net used for this purpose takes advantage of temporal links in addition to the other one. These links are
included for the interpretation of the next epoch (t+1) after the complete interpretation of the initial epoch t. During the
interpretation of every segment with a particular state several hypotheses are created along the temporal links. These
hypotheses exclude each other. According to the priorities the verification of the different state transition hypotheses is
processed in a particular order. The search tree splits. In case of a good result of a verification, the other competitive
hypotheses will not be verified anymore. At the end of the instantiation for epoch t+1 all instance nodes of the
interpretation for the time t will be removed, and the interpretation will continue for epoch t+2 in the same way.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1109
