International Journal of Advance Engineering and Research Development ANALYSIS AND DETECTION OF OIL SPILL IN OCEAN USING ASAR IMAGES

Transcription

1 Sientifi Journal of Impat Fator (SJIF): 5.7 International Journal of Advane Engineering and Researh Development Volume 5, Issue 02, February -208 e-issn (O): p-issn (P): ANALYSIS AND DETECTION OF OIL SPILL IN OCEAN USING ASAR IMAGES * Mukta Jagdish and 2 Jerritta.S Researh Sholar, Department of Computer Siene and Engineering, Shool of Engineering, Vels Institute of Siene Tehnology &Advaned Studies (VISTAS), Vels University, Chennai, TN, India. 2 Researh Supervisor, Assoiate Professor (Eletronis and Communiation Engineering), Shool of Engineering, Vels Institute of Siene Tehnology &Advaned Studies (VISTAS), Vels University, Chennai, TN, India. ABSTRACT :- Here different tehniques are used to find out whih is best suited for detetion of oil spill with low time omplexity. This work is arried out using ASAR RADARSAT-2 image, whih is apture from Gulf of Mexio region. Main objetive of doing this researh is to ompare algorithms to determine dark pathes, bands and traking of oil spills in given input satellite images with less time onsuming. After omparing algorithms morphologial losing tehniques at as good tools for oil spill detetion and advane syntheti aperture radar image serves as a good sensor for surveying of oil spill. In onlusion, after omparing five algorithms, morphologial tehnique serves as exellent method for detetion of oil spill with low time omplexity. KEYWORDS:- Oil spill, Morphologial Tehniques, pattern, traking. I. INTRODUCTION Oil spill is one of the most important problem ours in the world whih beome one of the biggest issue in marine life. So regular monitoring is important, whih helps to solve problems based on oil spill. Extration of oil from oean is a fundamental work done for agenies to regular monitoring the sea. In this senario satellite image play an important role for data aquisition. In ever year oil spill aident take plae in history it was on the Gulf of Mexio in Deepwater Horizon on April, 20, 200, with explosion in July 5, 200. It effeted on wildlife habitats and maritime spies. To overome this problem radar image was used for regular monitoring whih improves over all oil spill problems by various approahes. To survey oil spill SAR image provide various advantage for detetion and traking of oil spills. Several satellite SAR sensors are involved in the oil spill detetion and survey. These data are from ERS-/2, (Brekke and Solberg 2005) ENVISAT (Marghany 203), ALOS, (Zhang et al. 20, 202), RADARSAT-/2, (Zhang et al. 202) and TerraSAR-X (Velotto et al. 20) whih have been globally used to identify and monitor the oil-spill. Reently, the multi polarimetri SAR highresolution data have beome a vital researh area for oil spill detetion (Skrunes et al. 202; Shirvany et al. 202). Oil spill detetion and monitoring using SAR tehnology, data are sare job, beause of barely disrimination between oil spill and other features of look-alike,shadows, wind speed that appear pathes in SAR data as Dark pathes (Topouzelis 2008). The problems faed in oil spill automati using SAR data, is ahievements in past deades. Simultaneously, Frate et al. (2000) proposed semi-automati oil spill detetion by using neural network, in whih a vetor defining features of an oil-spill is used. Topouzelis et al. (2007, 2009) and Marghany, Hashim (20) onfirmed that neural network tehnique ould give preise differene among look-alike and oil- spill in SAR data. Topouzelis et al. (2007) has used neural networks in finding both oil-spill and dark pathes detetion. Experimental results shows, 89 % auray and 94 % dark pathes segmentation but ertain disadvantages like they annot effiiently detet small and fresh spills. Skrunes et al. (202), reports that there are several disadvantages assoiated with SAR sensors based oil spill detetion. So they suggested using multi-polarization aquisition data, suh as Terra SAR-X satellites and All rights Reserved 90

2 International Journal of Advane Engineering and Researh Development (IJAERD) II. DATA ACQUIRED In this study, RADARSAT-2 SAR data aquired by RADARSAT-2 operating with San SAR Narrow single mode beam on 27 th April, 200; st May 200; and 3 rd May, 200 are investigated for detetion of oil spill in the Gulf of Mexio. The satellite armed with Syntheti Aperture Radar (SAR) with multiple modes of polarization, whih inludes fully polar metri mode of operation in whih HH, VV and VH polarized data s were aquired (Maurizio et al. 202). It has got highest resolution of m in Spotlight beam mode (Ultra Fine mode of 3 m) with 00 m of positional auray. In the San SAR Wide Beam mode (WBM), the SAR has nominal width of 500 km and 00 m imaging resolution. The ground data obtained are based on study of Garia-Pineda et al. (203) where majority of oil types are emulsion and silver sheen. Sl. No Beam mode Plae Date Nominal pixel spaing( m) Resolution (m) Iniden t angle Polarizatio n ENVISAT ASAR ENVISAT ASAR ENVISAT ASAR Gulf Of Mexio Gulf Of Mexio Gulf Of Mexio 27 April 200 May May x x x x x x HH HH HH Table. ENVISAT ASAR IMAGE Figure. Input images III. MATERIAL AND METHODOLOGY Oil spill detetion Markov random field Fuzzy C mean Adaptive Thresholding Morphologial losing K means lustering Figure2. Methodology A. Markov random field Markov random field tehnique provides mutual influenes in given data, whih helps to provide more lear detetion on oil spills in images with objetive funtion modifiation. With the help of modifiation proess it helps in reduing spekle noise effet in radar images. It works together with membership funtion using neighborhood pixels. Steps are as follows- () initially two satellite image has to be taken, onsider as P = {P I, h, I M, h N} and {P 2 I, h, I M, h N}, where M x N. (2) in seond step generate logarithmi and mean operator to over over all area taken for monitoring it is denoted with l = k, mean = min ( k, k 2 ), Where as,, f f f f & f 2, k & k 2 are logarithmi and mean values. Here Mean 2 produes differene images using information of pixels. Log value used to find out intensity for images and it overs All rights Reserved 9

3 International Journal of Advane Engineering and Researh Development (IJAERD) x area. Let us onsidered iteration x = with standard deviation σ r and mean μ r values. Now find out energy funtion E pq with iteration x. Now apply Gibbs expression, to find out prior probability (π pq ) π pq = x ) exp( E pq exp Ex pq + exp( E pq x ) Now (p r ) determine onditional probability then generate the distane matrix (d pq ) for the given input image. p y j r μ r, σ = r σ r 2π exp y 2 j π r 2(σ r ) 2 Now ompute objetive funtion- d pq d pq = n p y j r μ r, σ r J pq = n p r y j J pq δ μ r,σr, + } Now determine Membership matrix {μ pq π μ + pq exp( d pq ) pq = π pq exp d pq + π pq exp( d pq ) + Update deviation and mean values σ r and μ + r respetively, =+ μ r + = (μ pq y j ) x (μ x pq ) σ r + = x u pq (y j μ + r ) 2 (u x pq ) B. Fuzzy -Mean: It is a lustering tehnique, whih allows one objet belongs to two or more objet in lusters form. Similarly objet or data will be plaed in one plae and other plae. This method improves frequently using pattern reognition. Minimization objet funtion M b n J x = p= q= u pq x p q 2, n < Whereas m, belongs to real number whih is greater than, u pq is a degree of membership, p is a dimensional measured data, C q is a luster enter. * it denotes similarity between enter and measured data. In fuzzy - mean lustering, with the help of iterative optimization objetive funtion, fuzzy partitioning an be done, using update membership u pq and enter of luster p. u pq = b i= xp q xp i 2. n, pq = M un p = pq x p M un q= pq (i+) Iteration stops when max pq u pq u (i) pq < Es E denotes terminal point between 0 and for k th iteration; it also overs loal minimum pointj x. Following steps are. First initialize proess U so it denoted as [u pq ]matrix, u [0] 2. Now alulate enter of vetors (i) =[ q ] with u (i), pq = 3. Update the proess by using u (i), u (i+) Therefore U pq = b i= xp q xp i 2. n M un p = pq x p M un q= pq (i+) If u pq u (i) pq < E then stop the proess for further operation otherwise return to step All rights Reserved 92

4 International Journal of Advane Engineering and Researh Development (IJAERD) C. Adaptive Thresholding It helps to determine dark pathes, bands and pattern for the given data set. Adaptive is important beause radar ontrast depends on ondition of weather for detetion of dark spot, look alike and surrounding water, whih may hange in investigation. In image threshold levels depends on angle inidene even in onstant onditions in image. Aording to pre determined rule threshold level should hanged. In mask operation mapping from one pixel to a new one s proess take plae, pixel modifies aording to equation and are not dependent on neighborhood. f(p,q)=f[g(p,q)], Where F is input and output image with one to one mapping. f(p,q)=(p+,q),(p-,q),(p,q+),(p,q-) Adaptive help to produe segments with similar intensities, to determine boundaries and shape for the given images. It produes gray sale value with global and loal image information. σ q 2 = n (μ μ 2 ) 2 + n 2 (μ μ 2 ) 2 If denotes variane for the threshold silk smooth extrated using (R)= - ( σ 2 ), with standard deviation. D. Morphologial operation Morphologial operation is based on the struture analysis, in whih some omponents are seleted among all, whih satisfy Gestalt priniples suh as () Proximity state that two objets are easier regard as a single objet by a human being if those objets are lose to eah other. (2) Good ontinuation states that objets are easier regard as a single objet if they an be ontinued from one to other. Aording to these priniples riteria have been developed. If both analyzed omponents are lines then first group of riteria is applied. Based on this riteria proximity is state as minimum of Eulidean distanes between starting and ending points of these lines and the exellent ontinuation is determined by heking orientation differene. If both objets are not lines then seond group of riteria is applied. Relation of lines to non-lines then third group of riteria is applied. The analysis is done by two routines: In-level routine and inter level routine. In-level routine: It Analyzes level N = {3, 2, } omponent to gain oil slik struture. Struture is onsist element that satisfy proximity riteria and good ontinuation riteria. The oil slik struture is seleted from eah omponent of level N that alls the inter-level routine to group or luster it with the strutures levels from lower. In Inter-level routine the omponent is given from a level N >, all omponent are searh from level N- and omponent are find that satisfies riteria of Gestalt-based. Set urrent omponent, set level as N = N-, then all Inter-level routine again, then stop the analysis when urrent level show N = or no omponents present to analyze. Analysis start for gray level mask from level 4. This level represents lowest baksatter area in radar images. If level 4 gray ontain few pixel then merge level 4 and level 3 that represent dark areas of the Syntheti aperture radar fragments with slik orrespond look-alikes areas. The levels N= {3, 2, } routine is run to find oil silk struture whih orrespond to less-dark areas of SAR image. Then this struture passed to subsequent analysis for inter-level routine analysis. Components are seleted in stages of analysis are retained in the image and other onsidered as noise. E. K-mean lustering: K mean lustering is a vetor quantization method used for oil spill detetion. Here eah element is partition into k lusters whih belongs to nearest mean, at as prototypes for the luster. It works on dividing data ell into voronoi ells. K mean luster determine omparable spatial extent lusters. It lassifies data whih is new into existing lusters whih alled as entroid nearest All rights Reserved 93

5 International Journal of Advane Engineering and Researh Development (IJAERD) Let as onsider set observations (a, a 2, a 3,.. a n ) with D real vetor dimensional. Here means lustering partition number of observation into n sets whih is denoted as S = {set, set 2, set 3. set k }, it will minimize sum of squares with luster. arg S minimum set = a set i a μ p 2 = arg set minimum p= set p vars p, μ i is points mean of set p whih is similar to squared pairwise deviations in same luster. arg set minimum p= x y 2 2 set p a,y set p similar features an be deleted by using formula a set i a μ p 2 = a y set p (a μ p (μ p y), whih shows number of total variane is onstant between points in luster. Let as onsider k mean initial set n (), n k (), with assign observation in luster, where mean has Eulidean distane set p (i) = {a p a p n p t 2 a p n j t 2 j, j }, where a p assigned one s and s (t) an be assigned to three or more luster. In update step it alulates and observed new mean to be entroid in the new luster. IV. RESULTS AND DISCUSSION In this approah, five different tehniques were used whih helps to regular monitoring and detetion of oil spills in the oean. This researh work is arried out using SAR RADARSAT-2 image. This tehnique examined SAR image to find struture of the oil spill with levels of gray orresponding to less damp / most damped area of sea surfae roughness. Radar images onfirmed grey level mask ontaining struture of the slik in Gulf of Mexio. Oil spill happened on 27 April 200 where rude oil spread in 49,500 km2 aross 9,2 square miles in Gulf of Mexio. As we know oil spill is one of the biggest issue in marine environment. Here five algorithms are applied to find out pattern, dark pathes and traking of oil spill with low time omplexity in the given ENVISAT ASAR images. In this researh different days images has been taken for regular monitor and observe ourrene of oil spill in oean. For detetion of oil spill inidene angle with HH polarization is suitable for researh. Aording to HH polarization and inidene angles it helps to redue noise whih is reated during bad weather onditions. For detetion of spill ASAR width inrease to 300km- 350km. Advane syntheti aperture radar provide high level of sensor images. Figure-3 indiates spills with patterns, dark pathes, oil spill traking and surrounding area of the images. To determine positive and negative pattern it ompared with neighborhood pixels. In this paper after omparing five algorithms, it define morphologial losing tehniques is more onvenient and good for oil spill detetion beause in ASAR images it slowly varies gray level point based on image loation and positions whih help to monitor and detet oil spill region in fast way with low time omplexity based on different weather ondition. Figure3. Oil spill Detetion Results Markov random field Time taken for MRF e+00 ses Time taken for MRF e+00 ses Time taken for MRF.70396e+00 All rights Reserved 94

6 International Journal of Advane Engineering and Researh Development (IJAERD) Fuzzy -mean lustering Time taken for fuzzy means e+0 ses Adaptive Thresholding Time taken for fuzzy means e+0 ses Time taken for fuzzy means e+0 ses Time taken for e-0 ses Time taken e-0 ses Time taken e-0 ses Morphologial losing Time taken e-0 ses Time taken e-0 ses Time taken e-0 ses K mean lustering Time taken e-0 ses Time taken e-0 ses Time taken e-0 ses Oil Spill traking Oil spill All rights Reserved 95

7 International Journal of Advane Engineering and Researh Development (IJAERD) Dark pathes Area e+02 Area e+04 Area e+04 Table2. Experimental Results for oil spill Detetion Time Tehniques used Image Image 2 Image 3 Markov Random Field e+00ses e+00ses.70396e+00ses Fuzzy means e+0ses e+0ses e+0ses Adaptive thresholding e-0ses e-0ses e-0ses Morphologial losing e-0ses e-0ses e-0ses K means lustering e-0ses e-0ses e-0ses Coverage Area (m 2 ) e e e+04 V. CONCLUSION As we know oil spill is a biggest issue in marine environment whih effet marine life yle. In this researh, omparing of algorithms take plae whih results in less time onsuming, oil spill traking, oil spill area, dark pathes and spill patterns. This work helps to regular day to day monitoring and detetion of oil spill in the oean. This approah helps to find out whih algorithm is best suited for detetion of oil spill with low time omplexity. This work is arried out using ASAR RADARSAT-2 image, whih is apture from Gulf of Mexio region. The work illustrate detetion of oil spill in the oean using satellite data with gray level masking, prepared with slik-relevant struture extrated by the algorithm with less time onsuming. Here morphologial losing tehniques an be used as a good tool for monitoring and identifying the ourrene of oil spill and Syntheti aperture radar image serves as a good sensor for surveying of oil spill. In onlusion, after omparing eight algorithms, morphologial tehnique serves as exellent method for detetion of oil spill with low time omplexity. VI. REFERENCES () T.Nishidaid, H.Harahsheh, T.Onumad, monitoring and detetion oil spill, Environmental Modeling and Software, Siene Diret, Marh (2) Albert Osei, Edmund C Merem, Data to detet environmental degradation, Remote Sens., Applied plant and soil siene 2, Marh 200. (3) Turgay Celik, unsupervised hange detetion for satellite image using Dual tree wavelet transform, IEEE transations on geo siene and remote sens. Vol-48, 3 Marh 200. (4) Istein Johansen, Mark Reed Oil spill modeling, siene and tehnology Bulletin, vol. 5, pp 6, 999. (5) M. Marghany, Multi-Objetive Evolutionary Algorithm for Oil Spill Detetion, ICCSA,355-37, Springer 204, (6) M. Marghany, Automati deteting oil Gulf of Mexio, Environment and Earth Siene, June 205, Springer. (7) Maged Marghany, Deteting of oil spills in Gulf of Mexio using SAR Data, ISRS June 206. (8) K.Karantzalos and D.Argialas, level set segmentation oil spill traking. International Journal of remote sensing (IJRS), Volume 29, November (9) H.S. Solberg, Detetion of oil spill, Elsevier, Siene Diret, Remote Sensing of Environment, November (0) M. Marghany, automati redution of oil spill from Satellite data. International symposium of the digital Earth IOP, July () D. Camassa, Adalsteinsson, R. Harenberg, Z. MLaughlin, S. Lin, Subsurfae trapping of oil plumes in stratifiation, Oil Spill Monitoring and Modeling the Deepwater Horizon, Geophysial Monograph Series, 20. Camilla Brekke, Anne H.S. Solberg. November 2004, Oil spill Detetion by satellite remote sensing. Siene Diret, Elsevier. Remote Sensing of Environment. (2) Van Genderen J and Marghany M, July 2009 Entropy automati redution of oil spill from RADARSAT-2 SAR Satellite data. International symposium of the digital Earth IOP onferene0.088/755-35/8/0/ (3) Zhao J, Ghedira and Temini M, September 204, Exploring the potential of optial remote sensing for oil spill detetion in shallow oastal waters a ase study in the Arabian Gulf, express All rights Reserved 96