MINISTRY OF EDUCATION & TRAINING
UNIVERSITY OF MINING AND GEOLOGY LE MINH HANG RESEARCH PROPOSAL METHOD FOR IDENTIFICATION
AND CLASSIFICATION OF OIL SPILLS AT SEA
BY MICROWAVE REMOTE SENSING DATA

Research field: Geodesy and mapping
Code: 62520503 SUMMARY OF PHD THESIS

Hanoi – 2013

The thesis has been completed at Photogrammetry & Remote Sensing
Department, Faculty of Geodesy, University of Mining and Geology,
Hanoi

Full name of supervisors:
1. Assoc. Prof. Nguyen Dinh Duong
Institute of geographic, Vietnam Academy of Science and
Technology

Nội.
3. Le Minh Hang, Nguyen Dinh Duong (2010), Practical implementation of
vectorization of oil spills detected at sea on SAR image, The 31
th
Asian
Conference on Remote Sensing, Hanoi, Vietnam
4. Lê Minh Hằng, Nguyễn Đình Dương (2010), Xây dựng chương trình đọc
tư liệu viễn thám siêu cao tần phục vụ phân tích vết dầu trên biển, Tuyển tập
Báo cáo Hội nghị khoa học lần thứ 19 – Quyển 06 Trắc địa, tr.61- 66,
Trường Đại học Mỏ - Địa chất, Hà Nội.
5. Nguyễn Đình Dương, Nguyễn Mai Phương, Lê Minh Hằng (2010),
Chuẩn hóa tư liệu ảnh SAR trên biển trong mặt cắt ngang, Tuyển tập các
công trình khoa học, Hội nghị khoa học Địa lý – Địa chính, tr.5 – 14,
Trường Đại học Khoa học tự nhiên, Đại học quốc gia Hà Nội, Hà Nội
6. Lê Minh Hằng, Nguyễn Đình Dương (2011), Tổng quan về các phương
pháp nhận dạng và phân loại vết dầu trên biển bằng tư liệu viễn thám siêu
cao tần, Tạp chí Khoa học kỹ thuật Mỏ - Địa chất, Số 35/7-2011, tr.66-71,
Hà Nội.
7. Lê Minh Hằng, Nguyễn Đình Dương (2011), Xây dựng chương trình đọc
dữ liệu ảnh vệ tinh EnviSAT ASAR chế độ thu nhận WSM, Tạp chí Khoa
học kỹ thuật Mỏ - Địa chất, Số 36/10-2011,tr.68-73, Hà Nội.
8. Nguyen Dinh Duong, Nguyen Mai Phuong, Le Minh Hang (2012),
OilDetect 1.0 - A System for Analysis of Oil Spill in Sar Image, Vol 12, No
2, tr.12-18, Tạp chí AJG (Asian Journal of Geoinfomatics)
9. Lê Minh Hằng, Nguyễn Đình Dương (2012), Nghiên cứu tách vết dầu
trên dữ liệu ảnh SAR bằng thuật toán nở vùng, Tạp chí Khoa học kỹ thuật
Mỏ - Địa chất, Số 38/4-2012, tr. 68-72, Hà Nội.
1

INTRODUCTION

- To propose the method for identification and classification oil spill at sea
2

by SAR image data consistent with the Vietnam conditions.
3. Subjects of study
- The characteristics of the transmitting and receiving signals of microwave
satellite.
- The impact of oil spill in declining the intensity fluctuations of waves and
characteristics of backscatter at the microwave satellite sensor.
- The factors affecting the accuracy of the identification and classification
of oil spill at sea by SAR image.
- The method for identification and classification of oil spill at sea by
microwave remote sensing data.
4. Range of the research
- The methodology is proposed to identify and classify the unknown origin
oil spills at seamainly discharge from from ships by SAR image data.
- The range of study is Vietnam East Sea.
- The research ability RADAR image of synthetic aperture radar (SAR),
with Band-L (ALOS PALSAR data) and Band-C (ENVISAT ASAR data).
5. Research Contents
- Research methodologies and methods for identification and classification
oil spill at sea by SAR image.
- Propose the method for identification and classification oil spill at sea by
SAR image in accordance with characteristics of SAR images observed the
sea and in the wide mode.
- Develop a program which can identify and classify oil spill and look-alike
by SAR image.
6. The methodology
- Analysis, synthesis of materials including scientific articles published in
over the world and Vietnam, results of experiments for early detection and

affects by weather. Oil slick in this case is not high contrast with the sea
surface in the SAR image. As a result, oil slick on SAR image has many
gray levels.
8.3. The method for identification and classification oil spill at sea by
microwave remote sensing data proposed in the thesis can be applied in the
condition of materials, infrastructure and information of Vietnam.
9. The new ideas of the thesis
9.1. Propose the method for automatically identifying and classifying oil
spill by SAR image data.
4

9.2. Propose the method for limiting the influence of near-far range effect on
SAR images data applied for identification and classification of oil spills at
sea. The near-far range effect exists on microwave remote sensing data,
especially for the wide mode.
9.3. Research the application of neural network MLP for identification and
classification of oil spills and look-alike on SAR image with the number of
various input parameters.
10. Volume and structure of thesis
The structure of the thesis is presented in 118 pages, 62 figures and
diagrams and 05 tables.

Chapter 1. OVERVIEW THE RESULTS OF RESEARCH
IN THE WORLD AND VIETNAM
1. Introduction
1.2. Overview of the research in the world
The research of using SAR image data to detect oil spill on marine
has studied since 1992 by Bern [11]. The author has used ERS-1 data (band-
C) to investigate the possibility of detecting oil spill on sea surface. The
research result includes:

and Radarsat (Band - C). There has not been much research on material
Band-L. The results of identification and classification of oil spill on SAR
image primarily based on expert knowledge. The completely automated
classification methods are still researched and experienced by different
mode. Vietnam have not invested a research method for monitoring and
detecting oil spill at sea.
1.5. These issues are developed in the thesis
The content of the thesis inherits the research which the student
have been done in KC09.22/06-10. Based on the results of research
archivement and published scientific journals, the student will continue to
study the application of image processing algorithms to improve the ability
to identify and classify oil spill in SAR image data such as:
- Research on application of contrast limited adaptive histogram
equalization (CLAHE) to remove influence of near-far rang effect on SAR
images.
- Research on application of automatic threshold algorithm to detect dark
spots on SAR image which adjusted near-far range effect.
6

- Research on using region growing algorithm to detect dark spots in the
case the oil spill was weathered and had low contrast on SAR image.
- Research on ability to discriminate oil spill and look-alike based on neural
network models MLP with geometric characteristics index of each slick.
- Experiment with 2 type data content of Band-C and band-L ranges. There
are the main remote sensing data being used in Vietnam.

Chapter 2. THE METHODOLOGY OF IDENTIFICATION AND
CLASSIFICATION OIL SPILL AT SEA BY SAR IMAGE
2.1. Principles of Synthetic Aperture Radar (SAR)
2.1.1. RADAR image system

2.3.1. Oil spill on SAR imagery
The viscosity of oil slick will reduce the short-wave oscillations,
increase surface tension and reduce wind pressure at oil spill location. So,
energy backscattering at that position will be reduced and as a result oil spill
on SAR image is dark spot, contrast to sea surface (Figure 2.10). The
contrast between oil spill and sea surface on SAR image data is the
important characteristic for identification and classification oil spill and it is
the advantages of SAR image to others remote sensing data. However, due
to fluctuations of the sea surface are complex with the natural conditions at
sea, the accuracy of the identification and classification results depends on
the objective conditions.
Figure 2.10. Oil spill on
SAR image
(a) Backscattering at oil
spill position and region
surrounding;
(b) Oil spill on SAR
image

2.3.2. Identification and classification of oil spill at sea by SAR images
According to research agency Aerospace Europe (ESA) [16], 45%
of the oil pollution comes from operative discharges from ships. The ships
often discharge waste oil on the road and oil slick has linear shape. Scientists
base on this shape to identify and classify oil spill on SAR image.
2.4. The affection of identify and classify oil spill on SAR image Bề mặt biển
Vết dầu
Vết dầu

2.4.5. The SAR image data on the experiment
According to the results published in the document [35], the value
of signal attenuation when sattelite observes an oil spill at sea by band- C
and band-L data are different.
2.4.6. Effected by meteorological conditions on sea surface
Under the impact of the environment at sea and by the physical and
chemical characteristics, the new oil spill is extracted easlier than the old one
on SAR image [35].
2.5. Conclusion Chapter 2
Methodology of the identification and classification oil spill at sea
9

by microwave remote sensing data based on the interaction of
electromagnetic waves and oscillations of the waves on sea surface. Oil
spills are dark spots on SAR images due to the decline Bragg scattering at
the oil slick position compare with region surrounding. Identified and
classified an oil spill on SAR image result is major influenced by wind
speed above the sea surface, look alike, characteristics of microwave data,
meteorological conditions of sea surface, chemistry and physical properties
of oil spill.
Chapter 3. PROPOSED METHOD FOR IDENTIFICATION AND
CLASSIFICATION OIL SPILL AT SEA BY SAR IMAGE
3.1. Data preprocessing SAR image
3.1.1. Converting the origin format to GeoTIFF
3.1.1.1. The format GeoTIFF data
3.1.1.2. Conversion ALOS PALSAR format data
The flow chart of converting from origin format of ALOS
PALSAR data to GeoTIFF is described in Figure 3.1.
ASAR
image data 3.1.2. Masking land and islands
The land and island are masked automatically base on the Coast East
Sea Vietnam data (reference Appendix 3).
3.1.3. Adjusting near-far range effection on SAR images
In the thesis, the author uses contrast-limited adaptive histogram
equalization (CLAHE) to adjust near-far range effection [30] (Figure 3.9).
11

(a) (b)
Figure 3.9. ALOS PALSAR image adjusted to normal of azimuth
(a) Before adjustment; (b) After adjustment
(a) (b)
Figure 3.10. Graph profile of backscatter by normal of azimuth
(a) Before adjustment; (b) After adjustment
Equalization process is done on each of the image window. In the
thesis used to adapt the window size 8x8 and contrast limit coeffection is
0.03. After adjusting near-far range effect by CLAHE algorithm, the

(3.17)
(a)
(b)
13

Shannon function in Equation (3.17) is a linear increase in the range
of [0, 0.5] and decreases linear from [0.5, 0]. The process of fuzzy
measuring using the formula (3.17) was calculated with a loop until
max
1tg
in which
1tt
. Optimal threshold value will be
determined by the smallest fuzzy value (refer to Appendix 4).
3.2.2. The region growing algorithm
The region growing algorithm is computed by a number of
seed points and grows the search area depending on the proximity of
threshold of these points. Formula (3.18) describes the growing
algorithm by gray-scale value of seed points and the pixel under
consideration.
 
:
i seed
P R True if z z T  
(3.18)
To compare the results of Huang method and region growing
method in Figure 3.17 and Figure 3.20, the region growing method have
been appropriated to detect dark spots which have low contrast with
surrounding sea surface on SAR image.


Set label for each region
Coordinate of points on boundary
Computer the index of region
Converted to geographic coordinate
Save data to Shapefile format
BEGIN
Origin image data
Detect dark spots in SAR image

END
15

Data input of neural network include geometric shape index of oil spill and
look-alike on SAR image. The output is the reliability of the classification
oil spill and look-alike. The number of experience test in this thesis is 100
samples including 67 oil spill and 33 look-alikes. The program choose 70
samples to train the network, 15 samples to validation and use 15
independence samples to test. The neural network structure are 8:8:2 which
input data consist of 8 shape indexs of dark spot including area, perimeter,
shapes (Sf), the complexity (PT), the standard deviation of gray values
(OSD), the average gray values (Osm), the largest value of gray, smallest
gray level values. Classification results are presented in Table 3.3 to achieve
93% accuracy. To confirm the role of the shape index affects the
classification result by neural networks, the author experiences other
structure model 4:4:2 which input data have 4 indexs including only area,
perimeter, and shape complexity; 4 hidden layer and output layer is 2 oil
spill and look-alike. The analytical results are presented in Table 3.4 to
achieve 96% accuracy.
Comparison of classification results with oil spill and look-alike by
neural network (MLP) with the structure 8:8:2 and 4:4:2, we have some

Figure 3.27. The method
automatic identification and
classification oil spill at sea by
SAR image Detect dark
spots
Conversion data to GeoTIFF format
Removing mainland and island
Adjusting near-far range effect
Filtering speckle noise in SAR image
Automatic threshold by Huang
Dark spots detected image
(Binary format)
Extracting boundary of dark spot
Computer shape index of dark spots
Input data to

by SAR image Detect dark spots

Conversion data to GeoTIFF format

Removing mainland and island

Adjusting near-far range effect

Filtering speckle noise in SAR image

Choose seed points in oil spill
Dark spots detected image
(Binary format)

Extracting boundary and computer shape index of dark spot

range effect creates the difference gray values of oil spill near range and far
range at the same image. It makes difficult to use automatic threshold
algorithm to detect dark spot on SAR image. We need to use region
growing algorithm to detect dark spots on SAR image in the case of oil spill
weathered by time and have many gray levels in oil spill.
According to the test results of identification and classification of
oil spill and look alike in Chapter 3 has confirmed the possibility of using
the shape index to discriminate between oil spill and look alike at sea by
SAR image. The shape index which is characteristic illegal discharges oil
spill are higher shape index (Sf), lower complexity index (C), lower area
and perimeter index compared with look-alike at the same time observation.
Besides, the experimental results confirm the ability application of multi-
layer neural network (MLP) to identify and classify oil spill and look alike,
improve the capability of the fully automated process identify and classify
oil spill and look alike at sea by SAR images.

Chapter 4. EXPERIMENT IDENTIFICATION AND CLASSIFICATION
OIL SPILL AT SEA BY SAR IMAGE
4.1. Analysis and system design detection of oil spill by SAR image
19

4.1.1. Design the functional component modules
4.1.2. Diagram of experimental program

Figure 4.1. Flow chart of the experimental program
4.1.3. Integrated modules and program design
4.1.4. Analysis the main modules of the program
4.1.4.1. Module analysis oil spill at sea by SAR image
4.1.4.2. Module display analysis results
4.1.4.3. Exit program

Saving data to Shapefile format

END
F
T
Analysis by ArcGIS or Neural Network (MLP)

20

4.1.5. Practical solutions in the experimental program
4.1.5.1. Image processing solution for larger image
The adjustment algorithm near-far range effection, filtering speckle
noise on SAR images are computed on each block and adjusts the gray level
which is difference between the block borders.
4.1.5.2. Limited the region to extract boundaries
The solution of limited the region in extracting boundaries include:
finding and detecting edges on the border of the image (reference Appendix
6) finding and cropping the smallest rectangle surround the entire oil spill
(reference Appendix 7) automatically. The process extracting boundary of
an oil spill would be carried out on limited image.
4.2. Experience results identification and classification oil spill at sea by
SAR image
4.2.1. Database of experience image
4.2.2. Experience results
Experience results by ALOS PALSAR image for example
PASL4200706130316260911040000 image which is supported by
ERSDAC, level SCG 4.2, collected on 13/06/2007 is shown Figure 4.6 and
Figure 4.7.
images is based on the difference in backscattering energy at the oil spill
position and the surrounding sea surface on SAR image. Energy
backscattering response obtained at microwave sensor of satellite when
observed at sea mainly Bragg scattering. Bragg scattering is created by the
interaction of microwave signals and oscillations of the waves. However,
the reliability of the identification and classification of an oil spill at sea by
SAR images depend on wind speed on sea surface, incident angle and
physico-chemical properties oil.
2. In general, there are three methods for identification and classification of
oil spill at sea by SAR images such as manual method, semi-automatic
method and fully automatic method. The manual method have been
identified and classified by the knowledge of experts. The semi-automatic
method is used widely because this method can be automatically detected
dark spots and calculated the geometric shape indexs of oil spill and look
alike. The fully automatic method is studying to discriminate between oil
spill and look alike by MLP neural networks.
3. The proposed methods in the thesis include in multi-step process image
with 3 main parts: 1) Image pre-processing; 2) Detecting dark spots on the
SAR image; 3) Identification and classification of oil spill and look alike.
4. The experience results prove of the proposed method well on band - L
data (ALOS PALSAR) and band C data (ENVISAT ASAR). There are two
main data which is being used in Vietnam. The proposed results are
experienced in normal meteorological conditions on the Vietnam East Sea
and without extreme weather such as hurricanes, thunderstorms …
5. It is reduce much processing time to analysis context by detecting dark
spots by automated threshold Huang algorithm, extracting the boundaries of