C.
:
*
>
unshrinkable
-
-
“----
Unexpansib
Un
shrinkable
"a
EE
Unexpansible
Figure.3 Possible changing patterns of landuse classes in temporal and spatial distribution
observation. Observational location and time/frequency can
be represented by locating observational pixel in the two
dimensional string. Spatio-temporal resolutions can be
represented by setting an aggregation formula over the range
of observation.
3) Total Fitness: Total fitness is computed from
behavioral fitness and observational fitness by the following
formula. Total fitness = Behavioral fitness * Observational
fitness or In(Total fitness) = In(Behavioral fitness)+In
(Observational fitness)
No
Observational Fitness = IT PCC
n=
Po, (Cp 7 Cp r05,) : Probability that observational value
C v ros i$ given when actual value is C »z
3.4 Definition of Operators
3.4.1 Reproduction
Reproduction is a process in which individual strings
are copied according to their objective function values or
the fitness values. Copying strings according to their fitness
values means that strings with a higher value have a higher
probability of contributing one or more offspring in the next
generation (Figure.4). This operators, realizing an artificial
version of natural selection, a Darwinian survival of the
fittest among string creatures.
duplicate
Individual with High Fitness Value
Figure.4 Reproduction of Individual
There are several proposals for selecting survival
individuals. The most basic scheme is called the roulette
wheel scheme, the deterministic sampling and the elitist
scheme. In order to efficiently find the best solution in a
search space, in our search, we proposed the selection
scheme based on the combination of the deterministic
sampling and the elitist scheme. The selected survival
possibility in next generation of each individual is calculated
as in the deterministic sampling. And the best individual is
kept into the next generation as in the elite scheme.
3.4.2 Crossover
Crossover operator first randomly mates newly
reproduced individuals in the mating pool. Then it randomly
locates a window with random size for a pair of individuals.
Finally, the contents of individuals within the window are
swapped to create new individuals (Figure.5).
Individual! „“ Individual2
Figure.5 Crossover of Individuals
3.4.3 Mutation
Mutation operator plays a secondary role in the
simple GA. It occasionally alters the value in a individual
position (Figure.6).
' a
' 1
' 5^ ' 253
Fe % Az al LA
Team Amen
$ 1) 1 ° tL
= pan = T LT
e e
d (3 change | .°
°° Individual D?
Figure.6 Mutation of One Individual
4. IMPROVEMENT OF THE SEARCH IN GAs
4.1 Hill-Climbing method to improve the efficiency of
genetic algorithm
Searching a complex space of problem resolutions
often involves a tradeoff between two apparently conflicting
objectives: exploiting the best solutions currently available
and robustly exploring the space (Lashon Booker, GA&SA).
Generic algorithms have been toured as a class general
purpose search strategies that strike a reasonable balance
between exploration and exploitation. The power of these
algorithms is derived from a very simple heuristic
assumption: that the best solutions will be found in regions
806
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
of
of
of
po
at
ob
the
an
ea:
op
ex
the
ha
ite
to
pre
co
en
the
str
co
str
the
of
ex
it |
ca
int
lai
lai
Sr
to
un
ha
dii
fo
