International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
represent a non linear solution to a non linear problem, whose
effectiveness grows up with the increasing of the number of the
Ground Control Points.
input layer
TOL |
—_— x
IN
= / a tmt
v" Ÿ hidden layer
Figure 1 - MLP NN with 2 computational layers (hidden e
output).
MLP NN belongs to the feed-forward NN family. Adopted
training algorithm is the Back Propagation Levenberg-
Marquardt one. Procedure make use of the Neural Netorl
Toolbox i MATLAB 5.3.
NN architecture is the one shown in Figure 2. The most
appropriate number of neurons has to be defined time to
time according to the number of GCPs and image type.
Only an expert user can successfully control it.
Indications for the best architecture can be derived from
RMSE (Root Mean Square Error) analysis. The NN
approach is quite sensible to the initialization of the
weights of the neurons.
| INPUT LAYER HIDDEN LAYER OUTPUT LAYER |
TT [7 7 7 7 0E [CUESTA 4
| ES Ls rh i
| ALPES |
| 1 |
| ; edd sb. |
IC /
8, 6
s +b LU |
TNE M i
YE bh | y |
— ——— |
| = (fr 2t.
I m |
| |
| 7 |
*b |
|
ed Le ts ce dis
Figure 2 - MLP NN mathematical model with 2 computational
layer (hidden e output) designed for the ortho-correction
problem.
1.2 Rational Function Model
This is the most famous and used non-parametric model. It is
present within almost every remote sensing commercial
software. It allows to relate image coordinates (6,7) with
object-terrain 3D coordinate (X,Y,Z) through rational
polynomials as shown in (1):
bar ATA.
F PG Y,Z)
- PULY ZI
BU y. Z)
(1)
5
P4 PPS Pa polynomials of maximum 3? degree (78
parameters to be estimated) whose equations are (2) o (3):
P(X.Y,2)-a, +a X +a,Y +a,Z +a,X" +
Ye XV td, Cra, V0 +4
Hy my my
PUOYZeXYYv.4r (3)
i=0 j=0k=0
0<m <3;0<m,<3;0<m, <3 em +m, +m, <3
Equations (1) are known in literature as RF M Upward.
2. EXPERIENCES
Here are presented two experiences carried out on images
of the type previously described which both take into
consideration the geometric quality of the orthocorrected
images.
2.4 Technical Map Updating
Problem is to evaluate if SPOT 5 images can be successfully
used for map updating problems, and which scale map they can
be rigorously suitable for.
As far as such validation is concerned two test sites have been
chosen in the outskirts of the city of Turin (Piedmont, Italy):
Stupinigi (Figure 3) and Venaria Reale (Figure 4). These places
are sited in an area that has been subject to intensive changes
over these years.
They represent two Italian cultural sites as they are old
residences. Stupinigi was built in the first half of 18th century
by the architect Juvarra as a hunting lodge and residence for the
Savoia family; it is the nucleus around which a National Natural
Park develops. An ancient medieval castle, Castelvecchio,
which is well preserved, can be found near the hunting lodge.
Figure 3. Royal hunting lodge and national park in Stupinigi
(Turin)
1068
Intei
The
Hun
It w
cons
for &
the a
A SI
The
Tabl
The
duri
grid
Acci
depe
1:10
refer
2.1.2
In €
atter
prob
to ui
is as
A de
map
cons
valu
inter
map
Resi
each
Tabl
appr
Tabl
MLI
imag
