International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
(Richards, 1993) are one of the most popular multisource image
analysis systems. They use the GIS data in cooperation with the
image analysis outputs to extract the desired information
[Richards, 1993].
Recently, another important issue of the interaction between
GIS and RS is discussed, which is called GIS and RS
integration. Integration deals with the higher level of the
cooperation, which leads to the better result or a result that
could not have been achieved. In this manner, GIS information
is used to guide image analysis, which extracts more complete
and accurate information, which is in turn used to update the
GIS databases [Baltasavias et al. 2000].
Today GIS and RS integration not only is important but it is
necessary in order to reach to the desired information. GIS
databases are provided for many arcas around the world. On the
other hand remotely sensed images are taken from the various
parts of the earth. These two kinds of the geo-spatial data can
solve many of the problems of our life in combination to each
other. Each of these powerful tools completes the other and the
best results can be obtained by their cooperation. Baltasavias et
al. [2000], have listed some major aspects of the GIS and RS
integration. As a consequence we can Say. “if we have
information we should make use of them".
In this paper we make use of the available GIS data and/or
knowledge in order to extract more accurate results by image
analysis. We use the GIS and RS data to generate some
hypothesis maps and our predefined cost function, is used to
choose the best hypothesis for the objects. In our method we
generate a likelihood map for each radiometric class and then
overlay the existing boundaries of land cover objects on each
likelihood map. After this, the average probability of each
polygon is calculated and then using a threshold (variable
parameter) we generate a hypothesis map for the class. Then for
finding the best hypothesis map (the best threshold) for the class
we compute the cost of it. Now we can choose the threshold
with the minimum cost as the final estimated parameter. This
procedure is done just for one class in each time (see Figure. 4).
2.1 Object Dynamics
During the existence of an object, it may be affected by various
activations and things. This can affects its representation in GIS
in three ways [Molenaar et a.. 1992]:
Firstly, the thematic aspects of an object may change. In this
simplest case the value of one or more attributes change, e.g.
the cover type of an agricultural field changes or it may be that
an object to be reclassified e.g. the landuse class of a field
changes from farmland into build-up area. Secondly. the
geometric aspects of an object may change. This might be a
change of position, shape, size, orientation, or combinations of
these. These changes may lead also to changes of topological
object relationships (Figure 1(a)).
Thirdly, an object may change its aggregation structure. The
aggregation structure indicates how a terrain object can be
considered as a composite of smaller objects. Here too several
possibilities exist for such a transition (Figure 1(b)). The fact
that only the internal structure of the composite or aggregated
object changes implies that its external relationships are not
changed.
Here. we use GIS data (existing boundaries) to extract the first
type of changes. Therefore, we assume that there are no changes
of the reminder two types. But as it will be shown our method
can detect a majority of the changes of the third type.
2.2 Biddinghuizen Study Area and Data
The area of interest for this study is located in Biddinghuizen
region. This area represents a modern agricultural region in the
Netherlands [Abkar, 1994]. The agricultural fields are large and
usually rectangular. The main crops are grass, potatoes. cereals,
sugar beets, beans, peas, and onion. The elevation differences in
the Biddinghuizen region are very small. This region is a well
known area that we have a good set of data and information
about it. The RS data that we use for our experiment is a
Landsat TM image that was acquired on 7 July 1987 (see Fig
2(b).). The image, was of good quality and no atmospheric
corrections were performed. The image was georeferenced to
the national triangulation system using a first-degree affine
transformation. The pixels were resampled to the original size
of 30 m by 30 m. In this paper bands 3, 4. and 5 of TM were
used for classification and to generate the likelihood maps.
Additionally, various data at the Biddinghuizen test area were
stored in a GIS. A land cover map of this area with information
about crop types for 1987 was available. Then we have polygon
boundaries for each agricultural field and its crop type in that
specific date. Figure 2 shows the color composite image of the 3
used bands of the TM and land cover map of the study area.
|
i
i
1
e
4
SAN shape
Figure 1. The second and third types of the changes for objects
[Molenaar et al. 1992].
a) geometric changes b) aggregation structure changes; bl.a
set of elementary objects dissolves into one larger object or an
elementary (non-composite) objects fragmented into smaller
objects. b2. a collection of small elementary objects building an
aggregated object is replaced by a new set building the same
aggregated object.
In this paper, the Land use/cover served three purposes:
. Training field selection for the classification of
the satellite image
. To ensure the prior knowledge (fixed
boundaries) for the MBIA
. Validation of the final results
1282
Internat
Ag
(199
knov
avail
assul
are: :
influ
secti
The |
This
class:
inclu
beets
GIS «
(band
diffic
exam
gener
are al
maps
maps
In or
perfoi
menti
know|
better
only :
are al:
the Fi
In this
extrac
results
algorit
and ch
leads |
in ord
procec
genera
with tl
must b
4 shov
In this
GIS an
Crop €:
MBIA
bound:
This à
Biddin
as a b:
the ren
As it
likelihc
Here tl
For ea
that are
