mainly with the top layer of the forest canopy but not the en-
tire canopy. As a result, the C-band image presented a subtle
difference in volume scattering between forests and banana
trees. By contrast, the L-band image fully interacted with the
crown layer of forests and banana trees, thereby revealing a
substantial difference in volume scattering between forests
and banana trees (Figure 17a). Therefore, the combination of
L- and C-band images yielded less confusion between forests
and banana trees than the sole use of C-band data (Figure 7c).
Conclusions
Aiming at steering the selection of optimal combinations of
PolSAR
frequency bands for different land cover classification
schemes, this study investigated the capabilities of all the pos-
sible combinations of L-band
ALOS
PALSAR
fully
PolSAR
data,
C-band
RADARSAT-2
fully
PolSAR
data, and X-band
TerraSAR-X HH
SAR
data for land cover classification, which involve water,
bare/sparsely vegetated land, crop/rangeland, banana trees,
forests, and built-up areas. A method that integrates polari-
metric decomposition,
OBIA
,
DTs
, and
SVMs
was employed
for the land cover classification. Polarimetric decomposition
theorems were used to investigate the effect mechanisms of
radar frequency variation on the classification capability. The
main conclusions and findings of this study are as follows:
X-band
HH
SAR
data are not necessary for classifying the
land cover types involved in this study when C- or L-band
fully
PolSAR
data are used. Although the combination of X-,
C-, and L-band data achieves the highest overall accuracy and
kappa value, which are 98.31% and 97.93, respectively, the
addition of X-band data improves the overall accuracy and
kappa value by only 0.68% and 0.84, respectively. Unlike the
sole use of L-band data, the combination of X- and L-band
images increases the overall accuracy and kappa value by
only 0.25% and 0.30, respectively. The combination of X- and
C-band data even decreases the overall accuracy and kappa
value slightly, unlike the sole use of C-band data. Among all
the classifications that use a single scene, the X-band image
yields the lowest overall accuracy and kappa value.
C-band fully
PolSAR
alone is adequate for identifying primi-
tive land cover types, namely, water, bare/sparsely vegetated
land, vegetation, and built-up areas. Thi
C-band
PolSAR
is sensitive to the volume
caused by the difference in surface roug
and bare/sparsely vegetated land. With t
band
PolSAR
generates a more significant difference in volume
scattering between water and bare/sparsely vegetated land
than L-band
PolSAR
. We also find that C-band
PolSAR
produces
a larger dihedral scattering difference between forests and
built-up areas and a larger volume scattering difference be-
tween bare/sparsely vegetated land and crop/rangeland than
L-band
PolSAR
. This condition is probably because C-band
Pol-
SAR
interacts mainly with the crown layer of vegetation given
its relatively small penetration power. Therefore, L-band
PolSAR
can be discarded when C-band
PolSAR
data are used
for the classification of water, bare/sparsely vegetated land,
vegetation, and built-up areas.
L-band fully
PolSAR
alone is adequate for distinguishing
between various vegetation types, such as crops, banana trees,
and forests. This study finds that L-band
PolSAR
is more capa-
ble of revealing the volume and dihedral scattering differences
between forests and crop/rangeland, the surface and dihedral
scattering differences between banana trees and crops, and
the volume scattering difference between forests and banana
trees than C-band
PolSAR
. The superiorities of L-band
PolSAR
are due mainly to the high penetration ability, which allows
for the full interaction between the radar signal and vegeta-
tion to reveal the scattering mechanism variation caused by
the differences in canopy size, leave size, and trunk thickness.
Compared with the sole use of L-band
PolSAR
, the synergy of L-
and C-band
PolSAR
creates little improvements in the classifi-
cation accuracies of crop/rangeland, forests, and banana trees.
The conclusions and findings of this study can be useful
for determining the optimal combination of spaceborne
PolSAR
data for different land cover studies, especially when the cost
of data collection and processing is an important consider-
ation. However, this study examined only the contribution
of X-band
HH
SAR
data to land cover classification. With the
availability of X-band fully
PolSAR
data in the near future
(Bello
et al.
2015), further studies are needed to investigate
the improvement made by the addition of X-band fully
PolSAR
data in land cover classification.
Acknowledgments
This work was supported by the National Natural Science
Foundation of China (Grant No. 41601445), the Natural Sci-
ence Foundation of Guangdong Province of China (Grant No.
2016A030313230), the Young Scholar Research Fund of Sun
Yat-sen University (Grant No. 16lgpy05), the National Key
R&D Program of China (Grant No. 2017YFA0604402), and
the Science and Operational Applications Research—Educa-
tion International (SOAR-EI) Initiative of the Canadian Space
Agency and MDA Geospatial Services Inc. (Project No. 5167).
Xia Li served as the corresponding author for this article.
References
Baatz, M., U. Benz, S. Dehghani, M. Heynen, A. Höltje, P. Hofmann,
I. Lingenfelder, M. Mimler, M. Sohlbach and M. Weber, M.
2004.
eCognition Professional User Guide 4
. Munich: Definiens
Imaging.
Bello, J.L.B., M. Marione, C. Gonzalez, T. Kraus and B. Brautigam.
2015. First interferometric performance analysis of full
polarimetric TanDEM-X acquisitions in the pursuit monostatic
phase. Pages 1242–1245 in
Proceedings of 2015 IEEE
International Geoscience and Remote Sensing Symposium
, held
in Milan, Italy.
Benz, U. C., P. Hofmann, G. Willhauck, I. Lingenfelder and M.
Heynen. 2004. Multi-resolution, object-oriented fuzzy analysis of
for GIS-ready information.
ISPRS Journal of
d Remote Sensing
58 (3–4):239–258.
ss, M. Lachaise, A. Niedermeier and M.
aSAR-X
SAR
processing and products.
IEEE
Transactions on Geoscience and Remote Sensing
48 (2):727–740.
Chang, C. C. and C. J. Lin. 2011. LIBSVM: A library for support
vector machines.
ACM Transactions on Intelligent Systems and
Technology
2 (3):1–27.
Chen, K. S., W. P. Huang, D. H. Tsay and F. Amar. 1996. Classification
of multifrequency polarimetric
SAR
imagery using a dynamic
learning neural network.
IEEE Transactions on Geoscience and
Remote Sensing
34 (3):814–820.
Cloude, S. R. and E. Pottier. 1996. A review of target decomposition
theorems in radar polarimetry.
IEEE Transactions on Geoscience
and Remote Sensing
34 (2):498–518.
Cloude, S. R. and E. Pottier. 1997. An entropy based classification
scheme for land applications of polarimetric
SAR
.
IEEE
Transactions on Geoscience and Remote Sensing
35 (1):68–78.
Congalton, R. and K. Green. 2009.
Assessing the Accuracy of
Remotely Sensed Data: Principles and Practices
. Boca Raton,
Fla.: CRC Press.
Costa, H., G. M. Foody and D. S. Boyd. 2018. Supervised methods of
image segmentation accuracy assessment in land cover mapping.
Remote Sensing of Environment
205:338–351.
Dobson, M. C., L. E. Pierce and F. T. Ulaby. 1996. Knowledge-based
land-cover classification using
ERS-1
/
JERS-1
SAR
composites.
IEEE Transactions on Geoscience and Remote Sensing
, 34
(1):83–99.
812
November 2019
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
