we
ral
jis
ISPRS Commission III, Vol.34, Part 3A »Photogrammetric Computer Vision“, Graz, 2002
6. INTERPRETATION OF MOORLAND
6.1 Prior knowledge about moorland
In this section some background knowledge about industrially
used moorland is given (see also Figner & Schmatzler, 1991).
Following that the implementation of that knowledge into
semantic nets, the knowledge representation language of our
system, is shown.
= ma
SE | S 8
$9 |. iE |i $s
<E s c 9 e
ox tt o9 dc Sa L-—
f su | = EG S 22
| / =- | & Se | o o8
|i © &s | 22 zs
| % 2 So |
| ot m ca | © | a
| ag j = |
| 58 & ES
| $5 L4 ? |
lo mn 2 £98 | $ 3
I E == | S =
\ 8S9 | + = Ip s
ft 8 | 5 = 5 59
i. os Mt 5 4:57 22 L4
| ou S - S 3
| 2e m $ 1 E^
= ©
JE 22 § z
[is N =
FA a w t T
E 2 SE TTS] [
I] @ US o8 2
Ei T. ez ©
s | = > 55 5
© / MS ee - o
2) + = 30 9 ©
I Ll Oo, 57i E i
I] 2 Ig o£ 8 9
IH = | ei =
le 7 k&l.&s 2
I? ) E E 3
lj / 5 AEST.
Il f i
I / o
m ; — J : a UT =
[ ] [ Hf 17 o
| | | t | np Pd
Pa 51e Ÿ sy ; | s t 5
= | = | c if 3 o = = IE 2 o
| uis e 2 set pe &£ Ls ]| =8 4° às o.
| 5 ss 5 % oc a = ME- TE der on |
| o = © A £8 | © 8| $S« IP8 mo Oo
| = S o N > z = gx
| & | \ 3 & | 3
| N E |\ | bom 2
| | jh | © | | | a
M e = =
M | | ©
ls NT | £z
1 | o
2 N s o <
V \ o o
© Ii ig $9 o
\\ 2 23 =
en E S NEA e
\[ m ds a gz ©
LA s | E | Ée IN 29 o
$i sE uuu sese 5 LL
$ | s 5 |355 148
(1.32 fei for 832 ^
L| £8 & EzE|
| «5 £
Uc] 9 | | £58 |
(8 E à
VE eed [2 1 3
| 59
\ 225
S m Ld £53 rrr)
| m H M
\ oo c a
| | = FEE ES
| $ 2 332
| errr SENS
| w | a os NS
| | 230 185 2
| | o Is o
a ILS S
Na >
ris pr o
se FA
S
TI
z
$
=
1
o
=
1 9 1
2 — N zz OO
© 8000
> © = © >
QO c 0 cnn
o - O« E.
Figure 2. Semantic net for greyscale images
Originally, moors were upland moors. In Germany these have
practically vanished. Today, in the former upland moors
agricultural areas, forests and areas of regeneration or
degeneration are found. The most important industrial use of
moorland is peat extraction. To enable peat extraction in a moor
the ground must be drained first. For this purpose ditches must
be dug. Thus, the water level decreases, and the area begins to
degenerate. The vegetation changes. During degeneration the
vegetation is inhomogeneous and irregular. Then, peat
extraction is possible. Usually harvester machines are used for
this task. These machines leave two straight tracks on the
ground, which in aerial images can be recognised as parallel
lines. It is possible that peat works stop for a short time, and
then continue again. After peat works have finished,
regeneration of the moorland can begin. In most cases people
simply stop working the land and leave it to regenerate, which
eventually results in increased vegetation. Hence, vegetation
can be found again on these areas, especially birches, because
of the dry ground. Remains of tracks from the harvester
machines may still be found. To start up regeneration in the
direction of the original moorland, sometimes supporting steps
are carried out, as for example filling up of ditches, and trees
are removed to raise the water level. If the water level further
rises trees die, and a homogeneous vegetation without trees
appears.
A representation of the temporal part of this knowledge can be
seen in the state transition diagram in figure 1. The
monotemporal part is represented in semantic nets. Figure 2
shows one of them, designed for the interpretation of greyscale
images. At the top the moor classes are shown. Below, their
obligatory parts contain the features and structures, which have
to be found. The nodes in the greyscale aerial image layer
describe the appearance of features and structures in the aerial
images. These nodes are connected to the feature analysis
operators (described in section 4).
As described below the interpretation of greyscale and colour-
images needs different semantic nets. The semantic net for
greyscale images is able to distinguish eight different states, and
the net for CIR-images 12. This shows, that the missing colour
information results in no more than 30% less classes, which can
be distinguished. These numerous classes are achieved by using
structural and texture information. An example for a semantic
net used for CIR-images, as well as a more detailed description
regarding the use of semantic nets for interpretation is described
in (Pakzad, 2001, Pakzad et al. 2001, Heipke et al. 2000).
6.2 System Overview
Figure 3 shows an overview of the multitemporal interpretation
system. The interpretation starts with an initial segmentation
based on Geo-Data and radiometric/textural information for the
images of the first epoch, as described above. This results in
segment borders, which are the basis of further interpretation.
An interpretation for every segment is performed inside the
segment borders.
SS Start |
ETT Y | »um | V I MEA
| Aeriallmage | An | Geo-Data
| | ci: | |
| | [ Initial | | |
| — Segmentation +—
| | | for t, | |
et A sors,
| Resegmentation L We B der) Knowledge Base |
| for t 1 [^ S emen Bor SI /—] | for mono- / multi-
RU temporal Interpr.
| M : P n crar) | DZ |
| ultıtemporal | | |
| | Aerial Images | | |
vod > eee
voee e | A I» Knowledge Based |
É 2 / | Interpretation
Predicted i a for t;
new States apa ee
\ ) f TN |
ee State Ye
ul o. Trasition | Scene Description |
| Prediction of | Diagram | | for t; |
| State | | Qu. ] | |
| Transitions E (e at) | | |
[e | | |
[Ieri NT T \ ]
D |
Figure 3. Multitemporal interpretation system
The interpretation procedure utilizes semantic nets as
knowledge base. The semantic net, which is used for the
interpretation of the first epoch differs from the semantic nets
for interpretation of the next epochs. The reason is that some
classes can only be recognized by using temporal history. This
A - 237