al
1
al
y
WM ^ Re "4
NM CN me Um Cu - NT
--
Pakzad, Kian
During the interpretation an instance net is created. With the concept net as prior knowledge hypotheses are created and
verified in the instance net. At the end of the process the instance net shows which parts of the concept have been
verified correctly. Therefore, the instance net contains at the end a description of the scene. The creation of nodes in the
instance net is called instantiation. The instantiation starts with a predefined seed node. According to the strategy and its
priority of rules the instantiation proceeds in a particular order along the relations postulated in the concept net, until no
more rules can be applied and the instance net is complete. To show the instantiation process in our case, it is described
in the following. Here the instantiation process starts with the creation of a hypothesis of the concept moorsegment. At
this point one segment is taken from the segment image. The interpretation for this segment will now be performed. As
shown in Fig.3 there are four different possibilities of interpretation (states) for the segment. These possibilities exclude
each other and therefore compete with each other. The first state to be verified is area of peat extraction: A concept
node area of peat extraction is created. Two obligatory parts of this node have to be present: harvester tracks and low
vegetation density. This leads to the top-down instantiation of the concept harvester tracks along the part-of relation.
The concretization of harvester tracks is parallel lines, which also leads top-down to a creation of a hypothesis parallel
lines. Now the bottom layer is reached and this hypothesis has to be verified. The node calls a special segment analysis
operator. The operator examines the aerial image within the given segment and answers back whether parallel lines
were found or not. If the result is positive the operator returns a certainty value to the node, which describes the quality
of the result, and the instance node parallel lines changes its status from hypothesis to complete instance. This leads
bottom-up to a complete instantiation of the node harvester tracks. In the same way the second obligatory part of the
node area of peat extraction is be verified and for the second verification also a certainty value is determined. Now all
obligatory parts of area of peat extraction are present and the node is instantiated completely. Also a certainty value for
this node is computed from the nodes below. The result is a possible interpretation of the moorsegment with a certainty
value. If the certainty is not good enough the other competitive interpretations have to be verified in the same way. In
Fig.5 the result of the interpretation based on the initial segmentation (section 5.2) and on the CIR aerial image of the
test area (Fig.4) is shown. Results based on biotope mappings are depicted in Pakzad et. al. (1999). The result of the
interpretation reveals, that most segments were interpreted in the same way as a human operator would interpret them
using only the aerial image. The misinterpreted segments were mostly small, narrow or not typical for the land use
states. Using grayscale aerial images instead of a CIR in led to similar results (see Pakzad et. al. (1999)). This result
shows, that while color information in general contains additional information, for most unproblematic regions texture
information is sufficient for the interpretation
area of
peat extraction
agriculturally
used area
Figure 4. Aerial images of the used test area Figure 5. Interpretation result
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1107