Oil spill pollution has a substantial role in damaging the marine ecosystem. Oil spill that floats on top of water, as well as decreasing the fauna populations, affects the food chain in the ecosystem. In fact, oil spill is reducing the sunlight penetrates the water, limiting the photosynthesis of marine plants and phytoplankton. Moreover, marine mammals for instance, disclosed to oil spills their insulating capacities are reduced, and so making them more vulnerable to temperature variations and much less buoyant in the seawater. This study has demonstrated a design tool for oil spill detection in SAR satellite data using optimization of Entropy based Multi-Objective Evolutionary Algorithm (E-MMGA) which based on Pareto optimal solutions. The study also shows that optimization entropy based Multi-Objective Evolutionary Algorithm provides an accurate pattern of oil slick in SAR data. This shown by 85 % for oil spill, 10 % look-alike and 5 % for sea roughness using the receiver-operational characteristics (ROC) curve. The E-MMGA also shows excellent performance in SAR data. In conclusion, E-MMGA can be used as optimization for entropy to perform an automatic detection of oil spill in SAR satellite data.

Lately, oil spills in coastal zones have received much critical anxiety for its great damages on the coastal ecological system. Synthetic aperture radar (SAR) has proved as appropriate sensor for oil spill surveying for its wide-area and all-day all-weather surveillance potentials. Owing to its extraordinary imaging mechanism, conversely, the accuracy of oil spill detection is challenged by multiplicative speckle noise and dark patches instigated by other physical phenomena. In this perspective, dark patches do not be related to oil spills are known as look-alikes. They can be acclaimed to zones of low wind speed, internal waves, biogenic films, grease ice, wind front areas, areas sheltered by land, rain cells, current shear zones and up-welling zones (Lombardini et al., 1989; Teivero et al., 1998; Marghany, 2001). Consequently, three steps are expected to automatically detect oil spills in SAR images: (i) dark spot detection, (ii) dark spot feature extraction, and (iii) dark spot classification. Various classification algorithms for oil spill detection have been utilized, including pattern recognition algorithms (Teivero et al., 1998), spatial frequency spectrum gradient algorithms (Lombardini et al., 1989; Nirchio et al., 2005) and algorithms based on fuzzy and neural networks (Barni et al., 1995; Calabresi et al., 1999; Garcia-Pineda et al., 2013). Consequently, the oil spill automatic detection from SAR data are requested standard algorithm to overwhelm the multiplicative speckle noise and look-alike phenomena appearances. Marghany (2001) introduced entropy algorithm which is based on texture coocurrenace matrix for oil spill automatic detection from RADARSAT-1 SAR data. He found that entropy algorithm is able to discriminate between oil spill and look-alike phenomena. Indeed, the entropy algorithm can support the automatic detection of oil spill by reducing the uncertainty on the basis of information produced by multiplicative speckle noise and look-alike phenomena effects. Further, Shi et al. (2008) have implemented entropy texture algorithm for oil spill detection from SAR and optical remote sensing data. They found that the oil spill pixels are smoother than the surrounding environment. Shi et al. (2008) confirmed the work done by Marghany (2001). Besides, Minchew et al. (2012) declared the variability of the entropy is consistent with the variability of the oil properties suggesting that the entropy is providing a qualitative measure of the oil characteristics. Specifically, when there is open water and a thin sheen, the entropy is close to 0, but in the presence thicker oil (e.g. emulsion) the entropy has values that are close to 1.

Conversely, Skrunes et al. (2012) reported several disadvantages
associated with oil spill detection using the current SAR sensors and
stated that SAR sensors cannot detect the thickness distribution,
volume, oil/water emulsion ratio or chemical properties of an oil
slick. Instead, that group recommended the use of multi-polarization
observations, i.e., the data acquired by the RADARSAT-2 and TerraSAR-X
satellites. In addition, quad-pol RADARSAT-2 SAR (Zhang et al., 2011)
can provide information about oil spill thickness compared to other
SAR single channel such as RADARSAT-1 SAR, ERS-1/2 and Terra SAR. In
this reagrd, range of theoretical polarimetric SAR developments has
gradually qualified the accurate distinction between mineral oil
slicks and biogenic slicks (Liu et al., 2011; Minchew et al., 2012;
Skrunes et al., 2012). Recently, Minchew et al. (2012) used
Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) L-band
polarimetric for retrieving the oil volumetric concentration in
a thick slick that is based on Cloude-Pottier entropy algorithm
(Cloude and Pottier, 1996). The work of Liu et al. (2011); Minchew
et al. (2012); Skrunes et al. (2012) the Cloude-Pottier entropy
algorithm (

Newly, Staples and Rodrigues (2013) stated that entropy cannot be obtained from single co-polarized radar data, but requires quad-polarized data. Quad-polarized data means that the radar acquires two co-polarized channels (HH and VV) and two cross-polarized channels (HV and VH), but equally as important, quad-polarized data are phase-preserving meaning that the inter-channel phase difference (e.g. phase difference between HH and VV) is available. In contrast, Marghany (2001) and Marghany and van Genderen (2014) entropy texture algorithm provides excellent performance for oil spill automatic detection from different single SAR data.

Recently, Marghany (2014) utilized the Genetic algorithm (GA) as automatic detection algorithm for oil spill in RADARSAT-2 SAR data. Marghany (2014a) confirmed the work of Topouzelis et al. (2009). Both studies have agreed that the genetic algorithm is able to extract oil spill footprint boundaries automatically from the surrounding pixels without using a separate segmentation algorithm, as was done by Skrunes et al. (2012). Consistent with Marghany (2014), the genetic algorithm has the ability to determine the optimal number of regions of oil spill segmentation or to choose certain features, i.e., the size of the analysis window or selected heuristic thresholds. Further, The GA is shown to be able to identify and remove pixels that do not significantly contribute to oil slick footprint in SAR data. This conclusion has approved the findings of Mohanta and Sethi (2012).

The novelty of this work is designing optimization tool for the real time oil spill automatic detection using Entropy-Based Multi-objective Evolutionary Algorithm without involving others tool such as neural network or any image processing classification tools. Indeed, previous studies have executed artificial neural networks (Topouzelis et al., 2009; Mohanta and Sethi, 2012) or post-classification techniques (Barni et al., 1995; Calabresi et al., 1999), which are considered to be semi-automatic techniques (Marghany, 2001). Furthermore, both artificial neural networks and post-classification techniques are time-consuming and the probability of misclassification does not always decrease as the number of features increases, especially when sample data are insufficient.

Incidentally, the main objective of this work is to minimalize the look-alike dark pixels for accurate oil spill automatic detection in COSMO-SkyMed SAR satellite data which could be involved with oil spill footprint was detected by entropy and genetic algorithm. The Entropy-Based Multi-objective Evolutionary Algorithm uses both basic and advanced operators. For illustrative purposes, the method has been operated to oil spill footprint boundaries shape optimization which allows local and global optimizations. Indeed, global optimization which involves finding the optimal oil spill boundary shapes in COSMO-SkyMed data. Look-alike pixels can be removed to reach the optimal oil spill automatic shape detection.

This section describes the main equations of entropy algorithm and entropy-based multi-objective Evolutionary Algorithm (E-MMGA). These two algorithms are used for detection of oil spill from observed SAR satellite images.

Be a consequence of Harmancioglu (1981), entropy is a quantitative
compute of the information content of a series of data since reduction
of uncertainty, by making observations, equals the same amount of gain
in information. Therefore, Marghany (2001) and Marghany and van
Genderen, (2014) stated that entropy is a measure of the degree of
uncertainty of random oil spill footprint discrimination. In
a definition adopted from information theory (Cloude and Pottier,
1996), entropy is the numerical expression of oil spill footprint
boundaries in SAR images. In using this concept, oil spill footprint
can be measured indirectly based on the degree of the reduction of
multiplicative speckle noises and uncertainty of look-alike
effects. The main hypothesis is the oil spill footprint boundaries
have larger entropy compared to surrounding environment. Hence, in
order to quantitatively assess the cumulative effect of uncertainty in
oil spill footprint, entropy can be used as a metric for population
diversity of oil spill footprint boundaries which are stored at each
intersection of the column

Marghany (2001); Staples and Rodrigues (2013); and Marghany and van Genderen (2014) have proved the efficiency and validity of the entropy on oil spill detection in SAR data. Nonetheless, this approach is required range of threshold procedures to discriminate between oil spill footprint quantities and surrounding environment. As a result, the multiplicative speckle noises are not totally vanished around the boundary of oil spill footprints. In this prospective, multi-objective optimization algorithm can involve in entropy metric (Gunawan et al., 2004) to preserve the diversity among different solution to minimize the influence of the look-alikes and multiplicative speckle noise (Lathi, 1968; Marghany, 2001; Zhang et al., 2013).

Take the advantage of E-MMGA of preserving the diversity of solution
set (Gunawan et al., 2004) and solving the multidisciplinary of
uncertainty of random oil spill footprint discrimination in SAR
data. The uniqueness of this study is to deal with entropy of oil
spill detection as multi-objective Genetic Algorithm
(GA). Comprehending Coello et al. (2002), the multi-objective
optimization (MOP) has already been successfully adopted to solve
uncertainty of object detection in SAR images (Marghany, 2014a). In
general, MOP consists of

Entropy of oil spill footprint boundaries (

Total of entropy of oil spill footprint boundaries is (

If

If there is no

Then the weighted sum to combine entropy of multiple objectives into
single objective is given by Zhou et al. (2006).

To determine the diversity of entropy of multi-objectives which is
mostly more than two objectives for instance, oil spill, look-alikes,
rough sea, and low wind zone, compute the distance from a given
footprint centre to its nearest neighbour boundaries. This can be
computed by following equation adopted from Zhou et al. (2006) and
Zhang et al. (2013).

In this study, COSMO-SkyMed image is acquired on 29 July 2010 at
11:23:33 UTC which is implemented for oil spill detection in the Koh
Samet island, Thailand. This data covered
12

The Satellite has a Synthetic Aperture Radar (SAR) with multiple
polarization modes, including a fully polarimetric mode in which HH,
HV, VV and VH polarized data are acquired. Its meduim resolution is
5

Figure 4 shows the variation in the average backscatter intensity
along the oil slick footprint. The average backscatter intensity was
damped by

Figure 5 shows the entropy algorithm result. Clearly, the oil spill footprint has lower entropy value of 1.5 as compared to sea roughness and land. The land has highest entropy value of 3.5 entropy and sea roughness has entropy value of 2.7. Indeed, non-Bragg scattering is existing on land as backscatter becomes depolarized (Shi et al., 2008; Skrunes et al., 2012). Additionally, entropy algorithm has identified oil spill footprint boundaries by entropy value of 3.3. However, land entropy and oil spill footprint boundary having close entropy. In fact, entropy represents the randomness of scattering mechanism (Shi et al., 2008). According to Marghany (2001); Fukunaga (2013); and Marghany and van Genderen (2014) entropy is measure of uniformity in SAR image. In general, the entropy is a measure of variability or randomness because the concentration of the backscatter changes in relatively few locations would be non-random essentially. This confirms the study done by Shi et al. (2008).

Figure 6 shows the output result of E-MMGA. Clearly, E-MMGA is able to produce four different segmentation boundaries. Besides Fig. 7 shows that the thick oil spill footprint has highest E-MMGA value of 2 than medium and light oil spill. This is mainly because each multi-objective function in E-MMGA tends to bias its population towards the extreme edges of the Pareto frontier. This is confirms the work was done by Gunawan et al. (2004). Compared to entropy algorithm, E-MMGA is able to identify the look-alike footprint boundaries and discriminate accurately between, oil spill and look-alike, and surrounding sea surface. E-MMGA can accurately identify the morphological boundary of oil spill and assigned by different segmentation layer in COSMO-SkyMed satellite data. In fact, the Entropy-Multi-Objective Evolutionary Genetic Algorithm (E-MMGA) provides a set of compromised solutions called Pareto optimal solution since no single solution can optimize each of the objectives separately. The decision maker is provided with the set of Pareto optimal solutions in order to choose solution based on the decision maker's criteria. This sort of E-MMGA solution technique is called a posteriori method since decision is taken after searching is finished. This confirms the work done by Coello et al. (2002). In this context, the Pareto-optimization approach does not require any a priori preference decisions between the conflicting of oil spill, look-alike, land, and surrounding sea footprint boundaries. Further, Pareto-optimal points have form Pareto-front as shown in Fig. 6 in the multi-objectives function of the COSMO-SkyMed data space.

Entropy-Multi-Objectives Evaluation Genetic Algorithm (E-MMGA) which
based on the Pareto optimal solutions provides excellent
discrimination of oil spill footprint boundaries. This can be
confirmed by the receiver-operator characteristics (ROC) curve
(Fig. 8). In this regard, the existing of weight sum of objective
function converts a conflicting multiobjective problem of oil spill
and surrounding sea feature objectives. This can be seen in ROC curve
where oil spill has an area difference of 85 % which is larger
than look-alike and sea surface areas. Further,

On the word of Gunawan et al.,(2004), E-MMGA is able to preserve diversity and converge as fast as most of the single-level approaches (which are expected to be more efficient but less practical for large-scale problems of multidisciplinary nature). Besides, it improves overall quality of solutions by explicitly optimizing the entropy index at every system-level iteration, and then using this information to bias the search process toward obtaining a solution set with maximum diversity.

This study has demonstrated work to optimize the oil spill footprint detection in synthetic aperture radar (SAR) data. Therefore, Entropy-based Multi-objective Evolutionary Algorithm (E-MMGA) has implemented with COSMO-SkyMed data during the oil spill event along the coastal water of along Koh Samet island, Thailand. Besides, Pareto optimal solution is implemented with E-MMGA to minimize the difficulties of oil spill footprint boundary detection because of the existence of look-alike in SAR data. The study shows that the implementation of Pareto optimal solution and weight sum in E-MMGA generated accurate pattern of oil slick. Furthermore, thick oil spill has highest value of 2 E-MMGA than thin and medium spills. The E-MMGA, is able to preserve the morphology of oil spill footprint boundaries i.e. thick, medium, and light. In addition, the receive-operational characteristics (ROC) curve confirmed accurately performance of E-MMGA with 85 % oil spill detection, 10 % for look-alike and 5 % for surrounding sea surface boundary identification. In conclusion, E-MMGA is considered as excellent algorithm to discriminate oil spill from look-alikes and also to identify thick oil spill from thin one.

The author would like to thank Geo-informatics and Space Technology Development Agency (GISTDA) of Thailand for providing COSMO-SkyMed data