pd pe Ju) V e (D UO ÉD
EU ted =p oo
D CM ooo 0€
Pakzad, Kian
4 RECOGNIZABILITY OF DIFFERENT MOORLAND CLASSES
In order to recognize different classes in moorland areas an investigation about the recognizability of moorland classes
has been carried out (Suffrian 1999). As input data analogue CIR-aerial images with an image scale of 1:10000 in the
“Tote Moor” near Hanover were used. The task was to investigate, which moorland classes can be distinguished from
the given images by human operator. The visible differences between the classes could be differences in color
information just as differences in texture, structure and form.
As mentioned the aim of our research is to realize a system for the automatic interpretation of moorland. In order to
estimate which moorland classes can be distinguished automatically by the system, we assume that the upper limit of
classes can be estimated by the number of moorland classes, which can be distinguished interactively by an experienced
human operator. That means that in our opinion the judgment if a moorland class can be recognized automatically by a
system requires in any case the recognizability by an experienced human operator in the same input data. There is one
exception from this assumption: If the input data consist of many spectral bands, which usually can be handled better by
computer systems than by human operators. But this case applies not for the used input data.
Refinement Level1 Refinement Level2 Refinement Level3
4 Classes 10 Classes 14 Classes
excaved peat extraction
During the investigation as a first step the
eee operator began to distinguish between all areas
eatareaodracion | which show visible differences. The result of this
step were forty different classes. Then, an
accumulation of classes, which belong together
was performed. The result can be seen in Fig. 1.
area of re-/degeneration
heather state
area of re-/degeneration
birch state
area of re-/degeneration
State 3
milled p
area of peat extraction
manually extracted
block peat
For a better estimation which of the classes can be
recognized automatically the moorland classes
were included into three lists with three levels of
refinement. The first level shows the classes,
which can be distinguished and recognized very
well. The third level with highest degree of
refinement shows the classes, which have in some
cases only fine differences in the features.
area of re-/degeneration
agriculturally used area
high grassland
short grassland
birch forest birch forest The interpretation system described in section 5.3
forest pine forest pine forest is based on the classes in refinement level 1. As
shown in Fig.1 a distinction between an area of
other forest other forest
regeneration and an area of degeneration is very
difficult, because the vegetation is similar. A
distinction can be performed by using temporal
knowledge for the interpretation of multitemporal
data as shown in section 5.4.
Figure 1. Recognizable moorland classes
5 MOORLAND INTERPRETATION
5.1 Input Data
Our test area is the moor area in the northwest of Hanover near Steinhude in Lower Saxony. We work with aerial
images with a resolution of 0.5m/pel. The main input sources are CIR-images (from July), but we also tested the results
with grayscale images. The reason is that although color images contain more information, most available aerial images
are grayscale images. Also, for the multitemporal approach which is described in section 5.4 we used grayscale aerial
images from different epochs.
The second input source is a segment image. The segment image masks the different segments of the aerial image,
which is to be interpreted. The segment image can be based on a biotope mapping. Biotope mappings were performed
for many moorland areas in Germany by ground survey. For regions without biotope mapping we show in section 5.2 a
possible initial segmentation method which uses GIS information as prior knowledge.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1105
