International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
= My (1)
1 1 Jd
IN. : as PN uut Yun) : s ;
f; is the input state of context at time / on layer T and 7; is
the state of context at time / on space 7.
The evolution equation for / is as follows:
IN
HR (2)
where f is a free parameter and 0« Pi X] is chosen to
satisfy the constraint that lr; =1,
Spatial-temporal databases represent huge amount of data, new
technologies of processing and a new access and management
of large databases.
Trutnov - u staré lanovky
& occurence of the forest in
1953 1970 1977 1984 1989
=
8
BR B
g
Ei B
E
= t 8
€ W *
es OR El B
= 8 * =
= E E
® - E = 2
= E E
I. % 8 E
us B
= & t" 5
= Ë € ex ® =
= =
= E 8 à
ms Ei ë
= ® g
m E E &
mE 5 s
ms : = s
m E E & g
m = E & B
ms E ® ; 8
mu 2 5 ë
Ei : $i $ =
Figure 1. The result of temporal analysis is shown - the forest
body between years1958-1989 in Trutnov neighbourhood
(closed to power plant). Aerial data sets evaluation.
On the other hand, specific temporal based queries should be
used to analyse spatial-temporal oriented databases like:
Retrieve all locations that changed between f,and f, about at
least 1/10 of current value.
Calculate the changes between 7; and /, on given areas.
Compare the changes in given time strips.
Briefly speaking, great amount of data asks for automation of
the classification and interpretation process of spatial data
without the necessity to collect and utilize the large training sets
(Tyree, Long, 1997, Ocelíková, 2002). The useful approach in
the face of such conditions is to utilize hierarchy in analytical
process.
2. DECISION-TREE DESIGN
2.1 Tree Structure Design
Decision tree is often called multi-level classifier. It reflects the
fact that we deal with multi-class, multi-node and multi-peak
distribution. The tree structure hierarchy simplifies the problem
of multi-class recognition and convert it into more convenient.
space where the step-by-step decision-making can be applied.
The advantage of this approach is that the model construction is
approximately low cost and it is easy understand.
There are three main parts to solve:
Q Tree structure design
Q Decision rule selection
Q Feature selection
The relations defined by the class hierarchy cause the
inheritance of class descriptions (attributes, relations and
methods, it means for example the operations for the computing
of different parameters without depending on the selected scale
of the image, respective the size of the object) of child classes
on one hand and semantic grouping of classes on the other.
Inheritance: Class descriptions defined in parent classes are
passed down to their child classes. The class can inherit
descriptions from more than one parent class. The inheritance
hierarchy is a hierarchy of similarities that can reduce the
redundancy and complexity in the class descriptions.
Groups of aggregation: It is the combination of classes to a
class of superior semantic meaning. Beyond that, the group
hierarchy has the direct functional implication. Each feature that
addresses the class is automatically directed to this class and all
its child classes in the group hierarchy. The class can be part of
more than one group.
Structure: Differs slightly from the other two hierarchies,
although the structure can have parallels to the hierarchy of
groups. Different classes can be put together in structure groups
as a basis for the classification-based segmentation. This tool
make possible fuse even of previously heterogeneous regions to
single objects.
2.1.1 Boolean Matrix
To express any relationship of objects to each other objects we
can use the matrix. To model the hierarchy of objects, it means
inheritance relationship between object we can utilize Boolean
matrices. The true will be entered in the matrix element where
the row object (class) inherits from the column object (class)
and false everywhere else.
We can define the simple operations to identify the level of
inheritance transitivity, groups of aggregation, accounting of
restrictions and assumptions tailored exactly to the solved
application.
Boolean matrix expresses the inheritance of classes S1 — S9. We
can read it as follows: S2, S5 and S8 are subclasses of S1, $3,
S4 are subclasses of S2, S6 and S7 are subclasses of S5 and S9
is subclass of S8. In case of multi-class decision we deal with
sparse matrices.
0:0: 08:0 0,0 0:050
00000
gy g 0g g^g ^9 9
0 1.0.9.0, 0. 0.0. 0 (3)
8 lip 1050/40 0 00000
0990 91 d'g^9 0
9 0 0.01 0.0.0 0
n4 Qs oz
0 D700 0 Qf 9 0
108
Int
Thi
wh
the
ma
of i
cla:
unc
Thi
stat
resi
inf
On
kno
ana
ana
deri
The
net
can
obi
dec
22
The
Scal
the :
coh:
ima;
its n
We €