Pakzad, Kian
5.4 Multitemporal Interpretation
In this section we describe the extension of the system described so far to a multitemporal interpretation system. This is |
resegmentation „(segment bor der) necessary for our application because besides the
ER ean assignment of classes to areas in moorland the
monitoring of changes is very important.
multitemporal
data ! | A |
ea The multitemporal interpretation begins with an initial |
v.50 08800 interpretation for the aerial images taken at the first
interpretation
epoch t to be interpreted. Then the next epochs t+n
have to be interpreted based on the results of the
interpretation before. These results restrict the search
[ space and hopefully lead to an improvement of the
; monitoring process.
interpreted
n ; :
segments Fig.6 shows an overview of the system structure.
diagram
i Beginning with the part knowledge based interpretation
oe an initial interpretation of the segments is performed.
The results are interpreted segments of moorland.
1 These segments are the input for a prediction of state
transitions.
predicted new
states
state transition
prediction of
state transition
Figure 6. Concept for multitemp. moorland interpretation
This prediction uses prior information concerning the
possible changes. The possibilities are represented in a state Pi
transition diagram. A description follows below. The output
of the prediction are predicted new states for every segment.
upland moor
agriculturally used area
The borders of the segments may change between the (s
interpretation intervals. Therefore, for the multitemporal Pr
area of degeneration
|
approach we include a module to perform segment splitting
by segmentation. The approach is described in Pakzad et. al.
(1999) and uses the information of the predicted new states.
The results of this step are updated segment borders, which
are integrated into the knowledge based interpretation for
the new epoch, just like the predicted new states and the |
multitemporal data.
| area of regeneration birch state |
forest | | area of peat extraction 0)
area of regeneration heather state |
Figure 7. State transition diagram
The temporal part of the prior knowledge described in section 3 is implemented in the state transition diagram, see
Fig.7. It describes the most probable state transitions.
Although many more state transitions are possible there are
3
Sram restrictions by law and nature, and we can use these
i
Layer IT restrictions in order to improve the interpretation. a
moorsegment
we €
agriculturally
o
forest used area
oc area of area of peat
degeneration extraction
Dan — m Bar pat “rte,
medium
vegetation harvester
Partof.
high
vegetation
densi
dismember: i
structure |
Segment analysis operators
Figure 8. Extended concept net for multitemporal
interpretation
In contrast to the concept net in Fig.3 this diagram contains
seven different states. The first state, upland moor, is
implemented only to complete the diagram. Because upland
moor does not exist anymore in the test area it will not be
used in the interpretation. The states area of degeneration
and area of regeneration are now separated. In addition the
state area of regeneration is also divided into two parts:
birch state and heather state. As mentioned in section 4 the
distinction between the states area of degeneration and area
of regeneration in aerial images taken at one epoch is very
difficult. But in a multitemporal interpretation with the prior
knowledge described in the diagram, the development of the
different segments can be used also. For example given an
area of peat extraction the system knows, that this segment
has passed the state area of degeneration, and if in a new
epoch a segment analysis operator finds for example vegetation, the only possible states are area of regeneration and
forest. This prior knowledge enables also the distinction between the two regeneration states. If no more parallel lines
1108 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.
