TY - GEN
T1 - Oil spill detection from TerraSAR-X dual-polarized images using artificial neural network
AU - Kim, D.
AU - Jung, H. S.
N1 - Publisher Copyright:
© 2017 SPIE.
PY - 2017
Y1 - 2017
N2 - Marine pollution from oil spills destroys ecosystems. In order to minimize the damage, it is important to fast cleanup it after predicting how the oil will spread. In order to predict the spread of oil spill, remote sensing technique, especially radar satellite image is widely used. In previous studies, only the back-scattering value is generally used for the detection of oil spill. However, in this study, oil spill was detected by applying ANN (Artificial Neural Network) as input data from the back-scattering value of the radar image as well as the phase information extracted from the dual polarization. In order to maximize the efficiency of oil spill detection using a back-scattering value, the speckle noise acting as an error factor should be removed first. NL-means filter was applied to multi-look image to remove it without smoothing of spatial resolution. In the coherence image, the sea has a high value and the oil spill area has a low value due to the scattering characteristics of the pulse. In order to using the characteristics of radar image, training sample was set up from NL-means filtered images(HH, VV) and coherence image, and ANN was applied to produce probability map of oil spill. In general, the value was 0.4 or less in the case of the sea, and the value was mainly in the range of 0.7 to 0.9 in the oil spill area. Using coherence images generated from different polarizations showed better detection results for relatively thin oil spill areas such as oil slick or oil sheen than using back-scattering information alone. It is expected that if the information about the look-alike of oil spill such as algae, internal wave and rainfall area is provided, the probability map can be produced with higher accuracy.
AB - Marine pollution from oil spills destroys ecosystems. In order to minimize the damage, it is important to fast cleanup it after predicting how the oil will spread. In order to predict the spread of oil spill, remote sensing technique, especially radar satellite image is widely used. In previous studies, only the back-scattering value is generally used for the detection of oil spill. However, in this study, oil spill was detected by applying ANN (Artificial Neural Network) as input data from the back-scattering value of the radar image as well as the phase information extracted from the dual polarization. In order to maximize the efficiency of oil spill detection using a back-scattering value, the speckle noise acting as an error factor should be removed first. NL-means filter was applied to multi-look image to remove it without smoothing of spatial resolution. In the coherence image, the sea has a high value and the oil spill area has a low value due to the scattering characteristics of the pulse. In order to using the characteristics of radar image, training sample was set up from NL-means filtered images(HH, VV) and coherence image, and ANN was applied to produce probability map of oil spill. In general, the value was 0.4 or less in the case of the sea, and the value was mainly in the range of 0.7 to 0.9 in the oil spill area. Using coherence images generated from different polarizations showed better detection results for relatively thin oil spill areas such as oil slick or oil sheen than using back-scattering information alone. It is expected that if the information about the look-alike of oil spill such as algae, internal wave and rainfall area is provided, the probability map can be produced with higher accuracy.
KW - Artificial Neural Network (ANN)
KW - Dual polarization, coherence, co-polarized phase difference
KW - NL-means (Non Local) filter
KW - Oil spill
KW - Probability map
UR - http://www.scopus.com/inward/record.url?scp=85038442954&partnerID=8YFLogxK
U2 - 10.1117/12.2278715
DO - 10.1117/12.2278715
M3 - Conference contribution
AN - SCOPUS:85038442954
T3 - Proceedings of SPIE - The International Society for Optical Engineering
BT - Remote Sensing of the Ocean, Sea Ice, Coastal Waters, and Large Water Regions 2017
A2 - Bostater, Charles R.
A2 - Neyt, Xavier
A2 - Mertikas, Stelios P.
A2 - Babichenko, Sergey
PB - SPIE
T2 - Remote Sensing of the Ocean, Sea Ice, Coastal Waters, and Large Water Regions 2017
Y2 - 11 September 2017 through 12 September 2017
ER -