, 2012
S PAL SAR
TS
50
0
175
262
S PAL SAR
0
0
50
38
0
0
100
0
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
CONCLUSIONS
For the rapidly-growing oil spill as is in the Gulf of Mexico,
the improvement of detection and continuous monitoring are
the most important issues to effectively plan rapid emergency
response activities, and lessen its effects. In this context,
remote sensing technology is being used as an important tool
for both oil spill detection and monitoring the changes in its
direction.
In this study, testing the generalization ability of ANN MLP
classifier for oil spill classification and examining the
capacity of a very new heuristic optimization technique
(ABC) were aimed. The generalization ability of the classifier
was tested using a different regional SAR data. While the
training phase (optimization of model parameters) of the
ANN was done by using the objects extracted from the
Lebanon oil spill event, the objects from a different case area,
Gulf of Mexico, were used in the testing phase. In addition to
the regional variation, the effect of using different SAR
sensor (i.e. ALOS PALSAR) was also taken into
consideration.
The conclusions are mainly based on the means of the
producer accuracies computed from 30 different runs. The
multiple runs are significant for the algorithms initialized
randomly as in ANN in order to determine the robustness.
Each run was iterated both 1000 epochs and predetermined
optimum epochs. The means of the procedure accuracies
show that the generalization ability of ANN with LM and BP
for unseen oil data is not good enough. The best results from
optimum epochs are 41.9% for RADARSAT- 2 and 35.4%
for ALOS PALSAR. But, ANN with ABC is relatively the
best algorithm and gave moderate results, i.e. 72.4% for
RADARSAT-2 and 57.9% for ALOS PALSAR. Although
LM and BP are the well known algorithms, showing off the
capacity of the ABC which is the first time used for oil spill
classification is very important.
Since the oil slicks imaged with C- and L-band sensors show
different amounts of dampening under the same imaging
conditions, this will reduce the effect of look-alikes and false
positive alarms. The differences in Bragg scattering at
different microwave bands will allow for development of
better oil spill detection algorithms. In this paper,
RADARSAT-2 and ALOS PALSAR data were compared
showing different performances of the sensors for oil spill
detection. The results are attractive because the difference
between RADARSAT-2 and ALOS PALSAR is so evident.
The reason might be due to the original training data
extracted from RADARSAT-1 data, i.e. it works in same
band (C band, whereas ALOS PALSAR works in L band)
and have similar calibration parameters and similar
normalized radar cross section values.
Although results from ANN with ABC are significant enough
for oil data, since look-alike data is absent, which is the main
drawback of the study, it cannot be easily concluded that
ANN with ABC method could be effectively used in a semi-
automatic determination process of any oil spill data.
ACKNOWLEDGEMENTS
The authors would like to thank MDA and ASF for providing
SAR data.
71
REFERENCES
Alpers, W., Wismann, V., Theis, R., Huhnerfuss, H., Bartsch,
N., Moreira, J. and Lyden, J., 1991. The damping of ocean
surface waves by monomolecular sea slicks measured by
airborne multi-frequency radars during the SAXON-FPN
experiment. Proceedings of the International Geoscience and
Remote Sensing Symposium, Helsinki, Finland, pp. 1987—
1990.
Assilzadeh, H., and Mansor, S.B., 2001. Early warning
system for oil spill using SAR images. Proceedins of ACRS
2001—22nd Asian Conference on Remote Sensing,
Singapore, pp. 460—465.
Bava J., Tristan, O. and Yasnikouski J., 2002. Earth
observation applications through systems and instruments of
high performance computer. ASI/CONEA training course,
September-December, Matera, Italy.
Brekke, C., and Solberg, A.H.S., 2005. Feature extraction for
oil spill detection based on SAR images. SCIA 2005, LNCS,
3540, pp. 75-84.
Del Frate, F., Petrocchi, A., Lichtenegger, J. and Calabresi,
G., 2000. Neural networks for oil spill detection using ERS-
SAR data. IEEE Transactions on Geoscience and Remote
Sensing, 38, pp. 2282-2287.
Fiscella, B., Giancaspro, A., Nirchio, F. and Trivero, P.,
2000. Oil spill detection using marine SAR images.
International Journal of Remote Sensing, 21, pp. 3561—3566.
Haralick, R., 1979. Statistical and structural approaches to
texture. Proceedings of IEEE, 5, pp. 786—803.
Hovland, H.A., Johannessen, J.A. and Digranes, G., 1994.
Slick detection in SAR images. Proceedings of the
International Geoscience and Remote Sensing Symposium,
Pasadena CA, pp. 2038—2040.
Karaboga, D., 2005. An idea based on honey bee swarm for
numerical optimization. Technical Report-TR06, Erciyes
University, Computer Engineering Department.
Karaboga, D. and Basturk, B., 2007. A powerful and efficient
algorithm for numerical function optimization: Artificial Bee
Colony (ABC) algorithm. Journal of Global Optimization, 3,
pp. 459-171.
Karaboga, D. and Akay, B., 2009. Comparative study of
Artificial Bee Colony algorithm. Applied Mathematics and
Computation, 214, pp. 108-132.
Karaboga, D. and Ozturk, C., 2009. Neural networks training
by Artificial Bee Colony algorithm on pattern classification.
Neural Network World, 3, pp. 279—292.
Karathanassi, V., Topouzelis, K., Pavlakis, P. and Rokos, D.,
2006. An object-oriented methodology to detect oil spills.
International Journal of Remote Sensing, 27, pp. 5235-5251.
Keramitsoglou, L, Cartalis, C. and Kiranoudis, C., 2006.
Automatic identification of oil spills on satellite images.
Environmental Modeling and Software, 21(5), pp. 640—652.
Ozkan, C, Ozturk, C., Sunar, F. and Karaboga, D., 2011. The
Artificial Bee Colony algorithm in training Artificial Neural
Network for oil spill detection", Neural Network World,
Vol. 21 (6), 473-492.
