STUDY
ha, which is
f survey is a
irrounded of
1e surface of
territory. is
classify this
ing a humid
he region of
the level of
sent 56,14%
a.
allows us to
uisition. The
f the life of
hy] activity
L. We notice
. It is for this
supposed to
DOPTED
1s based on
ll as on the
je approach
ellite data.
f land cover
and to make
Into account
stratification
lv for the
d allowed us
nd cover»
nd grazings
ctation.
nous.
clérophvil).
IOUS tissu.
the photo
1c region of
technics of
fication by
of distance)
hers data as
Auxilhary Data
(Agricultural statitics)
Remote Sensing Data
(high resolution)
Terrain Informations
|
|
Photointerpretation
Pattern Recognization
Soil Occupation
Fig. 1: Methodological approach.
The classification being important, we defined several layers in
which will be taken different samples to investigate. Several
criterias as the type of met cultures. the parcels structure, the
localization, allowed us to establish this stratification, which
allowed us to define 7 layers in a first time:
= cercal zoncs:
= markets zones;
- Courses;
= Bare soil.
- Forests;
— Reforcstations;
= agglomcrations.
In order to get optimal classification, the image processing is
donc according to two approaches which were compared:
< a so-called supervised approach (classification by maximum
Likelihood or classification by minimum of distance):
= an approach no supervised (dynamic cloud).
For the first approach. thc selection of the test parcels was made
on onc hand, to lcave some information collected on the ground
and on the other hand. from the image of color composite. Two
algorithms of supervised classification were tested, maximum
Likelihood and minimum distance (Fig. 2).
Orpunal Data (SPO)
SPOT Data Masked
Y
Supervised Classiticaton
(Maximum [ikelihood and Minimum
Instance)
Ground
Data
: fier
A
analv sis
10 Classes
Image classified of
soll occupation
Fig. 2: Supervised classification (treatment by pixel).
The radiométric characteristic of the different cultures was
determined from the analysis of the whole test parcels defined
previously. We choose the most homogencous possible and
rejected samples which present an important tv pe eap. in order to
avoid the risks of confusion.
4.1 The results obtained:
The supervised classification chosen is the one of maximum
Likelihood of the XS channels with a mask which allowed us to
eliminate all what is not vegetation in our region of survey.
Indeed, this mask was achieved bv using a vegetation index
(NDVI). This indication allowed us to conceal all the non
agricultural part where the chlorophyll vegetation is absent
(image 3).
image 3: NDVI range betw cen 140-180.
We determined a threshold on this image thereafter in such a way
to take into consideration the classes which interest us. This
classification gave the following results:
N° Classes Nbre. Pts % image Strate agr.
0 Non classified 677258 64.59% -
| Cereals G.Y 52436 5.00% 14.12%
2 Cereals W.Y 303531 28.95% 81.74%
3 Courses ] 0.00% 0.00%
5 Bare soıl 1639 0.16% 0.44%
6 Forest 776 0.07% 0.21%
7 [rriguated zones 12935 1.23% 3.48%
Total 1048576 100% 100%
Array 4: Result of the classification by maximum Likelihood of
the XS channels (+ NDVI mask).
The masked region represents 677 258 pixels of the total image
ic : 64.59% (Array 4). On the 371318 classified pixels we notice
that 95.86% is occupied bv the two classes of cereals. These
results reflect the cercal aspect of the zone.
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 295
In
il
|
|
1H
(I
