confusion between new clear-cuts and other objects than
between new roads and other features. Visual comparison with
the topographic map indicates that most of the other areas with
high membership values are marshes. The image was registered
in August, at the end of an extremely dry summer, which would
explain the bright reflectance from these areas. There is most
likely no water in the marshes and the vegetation is dry and
partly woody. In the future a forest mask will be used, which
will exclude not only old clear-cuts but also marshes, and rock
outcrops.
x nec en Z m
Figure 6. The result from the rule for extraction of new clear-
cuts. Three known clear-cuts were detected. The pixels in these
clear-cuts have medium to high membership values.
Erroneously included areas mainly correspond to dry marshland
and a powerline corridor. Further processing is required to
remove these errors. These areas mostly have medium fuzzy
membership values.
A power line corridor runs through the area from north to south
and is clearly visible in the data layer showing areas with high
spectral reflectance (figure 4). It is also present in the map layer
with clear-cuts, but it has low membership values (figure 6). It
was not identified as a road since the filters which were used to
enhance linear features were constructed to detect thin linear
objects. However, powerlines are also present in the T5 map
and masking will consequently help solve this problem.
The difference in spectral signatures, and membership values,
between the new clear-cuts indicates the need of some ancillary
data for identification purposes. The age of the clear-cuts is the
same, but they face different directions which might explain the
difference. A digital elevation model will be available later in
the project, and it will be possible to use factors like slope and
aspect for identification. Some other spatial and contextual
factors which will be tested are:
e size and shape of individual regions
e existence of straight line segments that make up the border
of the region
e the spatial distribution of several adjacent regions
e closeness to existing roads/clear-cuts.
Inclusion of these factors will limit the risk of confusion
between, for example, clear-cuts and previously undetected
marshes, and make the identification more robust. Naturally the
rules will become more complicated when the number of factors
is increased. However, the rules should still be constructed in a
clear and simple way to keep the transparency high. They
should be generally formulated, and they should not be scene
specific.
The major advantage of using fuzzy membership is the ability to
formulate “soft” decision rules. No “hard” decisions are made at
an early stage of the analysis process, and thus the risk of
omitting pixels that potentially belong to a revision object is
reduced. The confidence in an identification is represented by
the combined fuzzy membership values from separate
statements. The fuzzy membership values will also be useful for
delineation of revision objects, since connected pixels with
varying membership values can be allowed to form an object,
The possibilities of deriving an accurate object boundary is
thought to increase this way. Iterative changes of the borders are
possible, based on different degree of membership.
5. CONCLUSIONS
This study has shown that a rule based approach can be used for
detection and identification of map revision objects (roads and
clear-cuts) for the topographic map of scale 1:50,000 (TS-
version), from SPOT panchromatic data. The application of
fuzzy logic in the rules results in a more flexible detection of
potential revision objects, than does classical set logic.
The detection of new roads was successful, with negligible
confusion to other objects of similar characteristics.
The detection of clear-cuts was, at this stage, almost entirely
based on spectral reflectance and, as in traditional image
classification, there was some confusion with other, similar
objects. When the rules are extended to include a larger number
of factors, the identification procedure will be more clearly
separated from ordinary classification, and this type of problem
may be avoided. The T5 map will also be used for masking to a
larger extent in the future.
6. REFERENCES
Ahern, F. and D. Leckie, 1987. Digital remote sensing for
forestry: requirements and capabilities, today and tomorrow.
Geocarto International, 2(3), pp. 43-52.
Altman, D., 1994. Fuzzy set theoretic approaches for handling
imprecision in spatial analysis. Int. J. Geographical Information
Systems 8 (3) pp. 271-289.
Centeno, J.A.S. and V. Haertel, 1995. Adaptive Low-Pass
Fuzzy Filter for Noise Removal. Photogrammetric Engineering
and Remote Sensing, 61 (10) pp 1267-1272.
Dymond, J.R., and P.G. Luckman, 1994. Direct Induction of
Compact Rule Based Classifiers for Resource Mapping. Int. J.
Geographical Information Systems, 8 (4) pp. 357-367.
Eastman, J.R., 1993. IDRISI Version 4.1 Update Manual. Clark
University, Graduate School of Geography. 209 pp.
538
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
Go]
Acc
Phc
181
Joh
SPC
Sen
Leu
She
189
Mal
for '
Swe
sens
Mel
rule
Eng
Mid
Tem
Sate
Sen:
Olss
segn
base
Sym
SP-?
765.
Piloi
appl
expe
Srini
Data
Infor
US (
State
Labo
Wan:
appr
Syste
Zade
338-.
