20 24
IN DB
feature. Peaks
ion and
:a from fig. 2
¡plit and merge
:d. This brings
lse the field
.n the manual
r a small effect,
i interpretation
lows these
;d for optimizing
>6-8 represent
)t be considered
C
L A
S S
I F
I E
R L
A В
E L
1
2
3
4
5
6
7
8
T
1
33
1
1
R
2
36
U
3
53
E
4
1
1
5
1
5
2
1
1 1
L
6
3
1
A
7
3
В
8
1
4
E
9
3
1
1
L
10
1
Table 2. Classification result after automatic field
segmentation of the testarea (see table 1 for legend
of labels).
REFERENCES
Hoogeboom, P. 1983. Classification of Agricultural
Crops in Radar Images. IEEE Trans GRS Vol GE-21,
p. 329-336.
Hoogeboom, P.; Binnenkade, P. and Veugen, L.M.M. 1984.
An algorithm for radiometric and geometric
correction of digital SLAR data. IEEE Trans GRS
Vol GE-22, p. 570-576.
Smit, M.K. 1979. Preliminary results of an investi
gation into the potential of applying X-band SLR-
images for croptype inventory purposes. IEEE Trans
Geosci Electron Vol GE-17, p. 303-308.
Loor, G.P. de; Hoogeboom, P. and Attema, E.P.W. 1982.
The Dutch ROVE Program. IEEE Trans GRS Vol GE-19,
p. 3-7.
Gerbrands, J.J.; Backer, E. 1983. Segmentation of
multitemporal side-looking airborne radar (SLAR)
images. Proceedings SPIE volume 397, p. 173-179.
(Geneva, April 19-22) Applications of digital
image processing.
Gerbrands, J.J. 1982. Introduction to digital image
processing (in Dutch), lecture notes L73A, Delft
University of Technology, November.
Leeuwen, P.J. van, 1984. Analysis of SLAR images
(MSC thesis, in Dutch), Delft University of Tech
nology, Information Theory group, July.
Class 8 (beans) is not planted until July, so in July
these fields are still almost bare, and therefore
easy to recognize (see fig. 2). Classes 9 and 10 are
not considered in the classifier and therefore
identified as other croptypes.
3 CONCLUSIONS
In this paper a follow-on study into the possibilities
of crop identification was presented. The goal was to
improve the classification result from a previous
study by adding early season SLAR flights and to
enable crop identification as early as possible in the
growing season.
A hierarchic classification procedure is proposed.
The success of this classifier is based on the
separability of winterwheat or rather wintercrops at
low grazing angles (5° - 15°) in the early growing
season (April, May) and the ability to discriminate
other croptypes in the mid-season on basis of their
angular dependence in the grazing angle range 5° -
35°. Field averaged radar backscatter values are used.
The test of the classifier was performed on the same
dataset as was used for the design of the classifier,
although for the test the fields were segmented in a
different way (automatic instead of manual). Care
should be exercised in the interpretation of the test
results, since the success percentages may be over
estimated in this situation.
Further investigation should incorporate a test in
ecologically different areas and areas with different
and more varied crop distributions. Also the use of
angular dependence should be further investigated. In
the Netherlands a research project is running to cover
these subjects.
4 ACKNOWLEDGEMENT
This study has been perfprmed by members of the ROVE-
team. Part of it was financed by the National Remote
Sensing Steering Committee (BCRS). '
Contributions to the work at the Physics and
Electronics Laboratory TN0 The Hague came from:
National Aerospace Laboratory NLR in Amsterdam, image
collection, Centre for agrobiological Research (CABO)
in Wageningen, ground data collection, Information
Theory Group of the Delft University of Technology,
classification.
Mr. R. Vlaardingerbroek of the Physics and
Electronics Laboratory TNO is credited for the many
calculations and plots he produced on the subject.
