The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
344
co is the mutation index, n is the quantity of all the antibodies
in T (Ab). i is the antibody's index which indicates the order of
the OIF value in T*(Ab).
5) Select d As whose F(/1) is max from T*(Ab), and a set T* d
is composed. T* d is an better solution set, and will be used to
improve the original population.
6) Replace d worst individuals from original population S(Ab)
by T d , and a clonal selection of S(Ab) is completed.
7) Evaluate the original population S(Ab). If the result reach the
acceptable domain, S(Ab) is the final result and the bands sets
in S(Ab) are the best feature. If not, go to the step 2) and do
another clonal selection on S(Ab) until the evaluate result get
the given domain.
—
—
—
+
—
4
+
—
4
4
4
4
4
■i
4
+
*
4
+
+
4
4 +
i
F*"'
+
<
---4.--
î.
-#—
*■
. +
*
4
+
T
1
+
1 i
0 5 10 15 20 25 30 35 40 45 50
feature
4. RESULTS AND CONCLUSIONS
Figure 4. Criterion values in original population S( A ) set
After pretreatment, the experiment select 176 bands from a
hyperion image data for the input data. Let the size of original
population be 50, and the length of individuals be 4. The
experiment is carried out in the MatLab 7.0 environment. When
(3 is 0.631, co is 0.873, and n is set to 10, d is set to be 8,
HDRM displays fast convergence. With the increasing of the
clone selection times, the average OIF value in original set Sn
increases accelerated (Figure 3).
1 2 3 4 5 6 7 8 9 10
iterated times
Figure 3. The evaluation result for feasible solution set in clonal
selection progress
Figure 5. Criterion values in population S(/t) set after 10
iterated times
When S(/l) is clonal selected for 10 times, its individuals have
gotten to nice convergence (Figure 4 and Figure 5). In Figure 4,
the individuals are partial regular, because they were arranged
during the clone selection progress.
It could help us faster solving a feature selection problem by
involving the HDRM. The effect of indexes in HDRM is shown
clearly during the experiment. M which is the size of the
original population decides the calculating size during the
progress.
(3 is the increment index. When it increases, the iterated times
decreasing obviously, but the time calculation cost increasing
much. So, it is needed to be set to an adaptable value. This
paper set it 0.631 through experiment.
co is the mutate index. If it is too larger, the random possibility
increases. The information original individuals contains could
be loss too much. If it is too smaller, its search will be too
partial, and can not to reach the global domain.