research article a study on coastline extraction and its

Hindawi Publishing CorporationAbstract and Applied AnalysisVolume 2013 Article ID 693194 6 pageshttpdxdoiorg1011552013693194

Research ArticleA Study on Coastline Extraction and Its TrendBased on Remote Sensing Image Data Mining

Yun Zhang123 Xueming Li1 Jianli Zhang23 and Derui Song23

1 Liaoning Normal University Dalian 116029 China2National Marine Environmental Monitoring Center Dalian 116023 China3 Key Laboratory of Sea Areas Management Technology SOA Dalian 116023 China

Correspondence should be addressed to Yun Zhang cloud208163com

Received 31 January 2013 Accepted 21 March 2013

Academic Editor Jianhong (Cecilia) Xia

Copyright copy 2013 Yun Zhang et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

In this paper data mining theory is applied to carry out the field of the pretreatment of remote sensing images These resultsshow that it is an effective method for carrying out the pretreatment of low-precision remote sensing images by multisource imagematching algorithm with SIFT operator geometric correction on satellite images at scarce control points and other techniquesthe result of the coastline extracted by the edge detection method based on a chromatic aberration Canny operator has a heightcoincident with the actual measured result we found that the coastline length of China is predicted to increase in the future byusing the grey prediction method with the total length reaching up to 19471983m by 2015

1 Introduction

The coastline is an adjacent transition zone between land andocean that is the result of dynamic change processes involvingnatural changes and human activities At present 60 of theworldrsquos population is gathered near a coastal zone rich inmetallic mineral resources oil and gas resources tidal andwave energy sources and other renewable energy sourcesThe coastline is an important place for human economicand social activities A quick and accurate determination ofthe coastline length variation trend is not only a necessarytechnical activity for the study of land-ocean interactionscoastal reclamation port development and urban expansionbut also an important subject for marine economics andmarine multidisciplinary research

There are two coastline extraction methods One isthe field detection method which is the more commonmethod that is the photogrammetric technology and GPStechnology The other is the remote sensing image coastlineautomatic extraction technology and image interpretationSuch software and technologies for coastline extraction havegradually becomemoremature as discussed by Li and JigangLee et al Manavalan et al Holman et al Rudin et al

Chan et al Donoho Xiaofeng et al and Chaoyang [1ndash9] The automatic remote sensing image classification havebeen well researched such as the proposal by Jiang et alfor an automatic scheme for the classification of land usebased on change detection and a semisupervised classifier[10] Stavrakoudis et al [11] built a boosted genetic fuzzyclassifier for land cover classification of remote sensingimagery Although the coastline can be extracted by these twomethods they fail to include a systematic and comprehensiveanalysis of coastline length and type variation trend whichmeans that the automatic sensing image classifications donot meet the shoreline classification The spatial data min-ing theory presented in this study is aimed at the imageprocessing of remote sensing image data and is focused onthe discovery of potential hidden and useful models andrules between the image targets from remote sensing imagesas discussed by Xiaocheng and Xiaoqin [12] It combinesthe spatial data mining method as discussed by DengkeDeren et al Zhanquan Kaichang et al and Longshu etal [13ndash17] with classic remote sensing image processingtechnology as discussed by Aihua et al Deren and JuliangGuobao et al and Tao and Bin [18ndash21] A new coastlineextraction and analysis model was put forward to analyze

2 Abstract and Applied Analysis

Remote sensing data selection

Data preprocessing

Data preparation

Data mining

Result interpretation

Image registration correction toning mosaic and other pretreatment processes

Coastline extraction and verification based on the chromatic aberration Canny operator

Coastline result analysisand trend prediction

Figure 1 Coastline extraction flow chart based on data mining

the future variation trend of coastline length bymeans of greyprediction algorithm to achieve the resulting conclusions andto propose feasible recommendations for the exploitation ofthe future coastal zone

2 Coastline Extraction Process

Data mining refers to the process as discussed by Kantardzicet al [22] in which implicit unknown potential and usefulinformation and knowledge can be extracted from a largenumber of incomplete noisy fuzzy and random databasesAccording to the knowledge structure of data mining thecoastline extraction processes can be divided into four stagesas follows in Figure 1 image selection image registrationcorrection toningmosaic and other pretreatment processescoastline extraction and mining and the coastline resultanalysis and trend prediction

3 Study on the Remote Sensing ImageCoastline Extraction Technology

31 Data Preparation The remote sensing image data pre-sented and selected in this paper were collected at five stagesfrom 2007 to 2011 The space range covered the coastal zonesof the Chinese Mainland (including Hong Kong and Macao)and Hainan Island encompassing 290 scenic spots Details ofthe images are as in Table 1

32 Data Preprocessing

321 Automatic Registration of the Coastal-Zone Image Themultisource image matching algorithm of a SIFT operatorwas presented in this paper which provided different spectralband setting imaging model reflectivity dimension andother aspects of the multisource data RANSAC and theidentical point triangulation network construction methodwere used for eliminating the mismatching points in orderto achieve the automatic matching of multisource remote

Table 1 Image details in 2007ndash2011

Period Satellitedesignation

Spatialresolution (m)

Quantity(scenic spots)

2011 HJ-1A 30 652010 HJ-1A 30 422009 HJ-1A 30 342008 CBERS-01 20 862007 CBERS-01 20 63HJ-1A satellite is the abbreviation of environmental and disaster monitoringand forecasting CBERS-01 satellite was jointly invested in and developed byChina and Brazil on October 14 1999 Because HJ-1A images and CBERS-01images are low-resolution areas the results of shoreline extracted have smalldifferences but this does not affect this study

sensing images and control point image databases afterwhich the automatic image registration was completed

322 Geometric Correction of the Coastal-Zone Image Itwas difficult to find the control points due to the largewater area on the remote sensing image at the coastal zoneMeanwhile a small number of control points were used toachieve the geometric correction of the low-precision remotesensing images in order to reduce the workload of the fieldmeasurements Combined with satellite orbit parameters andthe geometric correction model with correction precisionin different satellite data sources and terrain conditions thegeometric correction ofmultiscene imageswas jointly carriedout to achieve the geometric correction of satellite images atseveral control points in the use of the correlation betweensubscene overlapping regions

323 Automatic Toning and Mosaic of Coastal-Zone ImageImage data toning processing was carried out by using thecolor-transferring algorithm to maintain the consistency oftypical surface features In the fuzzy classificationmethod theoverlapping region calculation method and image multiscalesegmentation method were used for the study on the colortransferring algorithm to maintain the consistency of typicalsurface features The coastal-zone remote sensing image wasspliced by using the automatic matching-line generationimage mosaic technology

33 Data Mining

331 Coastline Extraction Based on Edge Detection Edgedetection methods were widely used for the automaticextraction of the water sideline so as to significantly reducethe quantity of data remove the irrelevant information andretain the important structural properties of the imagesThe methods consisted of Roberts Cross operator Prewittoperator Sobel operator Canny operator Kirsch operatorand Compass operator in which the Canny operator canbe regarded as the most common edge detection methodIn order to compensate for the insufficient remote sensingimage detection of complex surface features edge featuresor broken linear characteristics by the conventional Canny

Abstract and Applied Analysis 3

operator the chromatic aberration Canny operator was usedinstead to extract the coastline

(1) Calculation of the Chromatic Aberration Amplitude byMeans of Chromatic Aberration Canny Operator In order toimprove on deficiencies of the traditional Canny operator animproved Canny operator was used based on the LAB colormodel as the main technical route in this paper that is thechromatic aberration Canny operator color difference canny(CDC) algorithm This algorithm was able to detect the edgethrough the color gradient (chromatic aberration) betweenpixels and determine the pixel amplitude as per the calcu-lation of the first-order partial derivative finite difference atthe 119909-direction 119910-direction 135∘ direction and 45∘ directionwithin the 8 adjacent zones of pixels Its main advantagesinclude high edge-positioning accuracy and effective noisesuppression It was able to accurately determine the colordifferences in accordance with the vision of human eye andgood results were achieved during practical application Themathematical expression of the chromatic aberration Cannyoperator (CDC) is as follows

Chromatic aberration in the 119909-direction

119863119909[119894 119895] = CD (119868 [119894 + 1 119895] 119894 minus 1 119895) (1)

chromatic aberration in the 119910-direction

119863119910[119894 119895] = CD (119868 [119894 119895 + 1] 119894 119895 minus 1) (2)

chromatic aberration in the 135∘ direction

119863135

∘ [119894 119895] = CD (119868 [119894 + 1 119895 + 1] 119868 [119894 minus 1 119895 minus 1]) (3)


11986345

∘ [119894 119895] = CD (119868 [119894 minus 1 119895 + 1] 119868 [119894 + 1 119895 minus 1])

CD (119860 119861) = [(119871119860minus 119871119861)2

+ (119860119860minus 119860119861)2

+ (119861119860minus 119861119861)2

]

12

(4)

where CD(119860 119861) represents the chromatic aberration betweenthe pixel point 119860 and the pixel point 119861

The pixel chromatic aberration amplitude and directionwere calculated as per the coordinate conversion equationfrom the rectangular coordinates to the polar coordinatesThe chromatic aberration amplitude was calculated as per thesecond-order norm equation as follows

CDC [119894 119895]

= radic119863119909[119894 119895]2

+ 119863119910[119894 119895]2

+ 119863135

∘[119894 119895]2

+ 11986345

∘[119894 119895]2

(5)

Chromatic aberration direction

120579 [119894 119895] = arctan(119863119910[119894 119895]

119863119909[119894 119895]

) (6)

(2) Adaptive Calculated Dynamic Threshold A whole imagecan be divided into several subimages There might bea certain overlapping region between subimages so as to

achieve the continuous profile and calculate the parametersfor the proportion between the overlapping region and thesubimages The high and low threshold values of subimageswere adaptively set according to the nonmaximum suppres-sion results The calculation equation should be as follows

120591high = (1 minus 120573) 120591119867 + 120573120591ℎ

120591Low = (1 minus 120573) 120591119871 + 120573120591119897(7)

where 120591119867and 120591119871represent the global high and low threshold

values of the whole image respectively 120591ℎand 120591119897represent

the local high and low threshold values of subimage zonerespectively 0 lt 120573 lt 1 represents the threshold adjustmentrate If 120573 = 0 the whole image should not be adjusted If120573 = 1 the whole image should be divided fully in accordancewith the local features of the subimages

(3) Boundary Tracking Generated Coastline When the pointwith the chromatic aberration amplitude of a certain pixelin the whole image is greater than the high threshold valuewhich was used as the starting point for tracking theneighborhood pixel (other chromatic aberration amplitudeswithin the 8 pixel neighborhoods were greater than the highthreshold value) should be set as the edge and used as thestarting point for tracking If there were no pixels (chromaticaberration amplitude was greater than the high thresholdvalue) around the pixel point the pixel (chromatic aberrationamplitudewas greater than the low threshold value) should bediscovered within the 8 pixel neighborhoods and used as thestarting point for tracking up until the edge and the startingpoint cannot be found and thus determined as the contourendpoint The template was checked to determine whetherthe edge could be connected In the event of any noise at theisolated point it should be removed Finally the boundarywas extracted for fine processing

(4) Coastline Classification and Analysis Based on the Spec-trum Feature Database Based on the sample collection andfield spectrum measurement of the remote sensing image ofthe sea area the systematic research on the types and usagemethods for development and exploitation of coastal zonesas well as the spectral features and image features in thecommon star-source data the remote sensing classificationcriteria were presented for development and exploitation ofcoastal zones in this paper Coastlines can be divided intonatural coastlines artificial coastlines and estuary coastlinesThe natural coastlines can be divided into bedrock coastlinesilt-muddy coastline sandy coastline and biological coastline(including mangroves and coral reefs) The artificial coast-lines can be divided into cofferdam coastline marine recla-mation land coastline transportation engineering coastlineand protected coastline

The object-oriented image feature analysis for the sea areawas used in this study Based on the main types and compo-nents of coastline development and exploitation all kinds oftarget surface features were detected and extracted includingthe spectrum shape texture shadow space location andrelated layout The standard curve spectrum was obtainedcorresponding to the componentThe spectral characteristics


Table 2 The analytical coastline length of China based on remote sensing from 2007 to 2011

Year Coastline length(m)

Natural coastline length(m)

Artificial coastline length(m)

Estuary coastline length(m)

2007 18501296 9383095 8980905 1372972008 18515830 9730865 8651800 1331652009 18599902 10167232 8301830 1308392010 18644768 10378360 8136654 1297542011 18946699 10733161 8085561 127977

of each component were analyzed the coastline categoryinformation extraction was achieved by using a fuzzy clas-sification algorithm and the length of Chinese coastlinewas obtained (as shown in Table 2) after the analysis of thespectrum feature database

332 Validation of Coastline Extraction Results After thecompletion of the coastline extraction algorithm model itwas necessary to test and validate the model extractionresultsThe validationmethods of themodel extraction resultshould include the point-by-point comparison method andthe overall comparison method Point-by-point inspectionrefers to the superposition of extracted results and actualconditions with point-by-point comparison as to its accu-racy Overall comparison should refer to the evaluation ofextracted results by means of the indexes for the overallspatial pattern as discussed byWu [23]TheKappa coefficientwas selected to quantitatively reflect the model operation forextracting the accuracy of coastline This was mainly basedon the different-stage coastline extraction results and thecorresponding remote sensing image data to carry out theKappa coefficient calculation of the adjacent grid map todetermine the average coefficient value of results at each stageand to obtain the Kappa coefficient for the accuracy of coast-line extraction results over the timeframe The calculationequation should be as follows

119870 =

(119875119900minus 119875119862)

(1 minus 119875119862)

(8)

where 119875119900represents the percentage of consistent types of

parts on the comparative grid map that is the observationconsistency ratio 119875

119862represents the expectation consistency

ratio in which 119875119900= 119904119899 119875

119862= (1198861 lowast 1198871 + 1198860 lowast 1198870)(119899 lowast 119899) (the

total pixel number of grid 119899 the pixel number of grid (1) 1198861the pixel number of grid (0) 1198860 the pixel number of extractedgrid (1) 1198871 the pixel number of extracted grid (0) 1198870 thepixel number of two grids with the equal corresponding pixelvalue 119904)The different Kappa coefficients showed consistencyto varying extents

Shoreline information from the National Marine Dataamp Information Service was used to test the accuracy of theextracted shoreline The calculation equation was as follows

119868 =

(119871 minus 1198711015840

)

119871

(9)

where 119868 is the accuracy coefficient 119871 is the actual length ofshoreline and 1198711015840 is the extracted length of shoreline

Table 3 The accuracy test of coastline extraction in China from2007 to 2011

Year 2007 2008 2009 2010 2011Kappa coefficient 069 072 078 081 086Accuracy coefficient 0018 0013 0023 0015 0012

It can be seen from Table 3 that all the Kappa coefficientsobtained in the previous five years were greater than 06 andthat the accuracy coefficients obtained in the previous fiveyears were less than 0023 Previous studies about Kappa byBlackman and Koval [24] and Landis and Koch [25] haveshown that the coastline extraction results achieved using thechromatic aberration Canny operator edge detectionmethodshould be consistent with the actual coastline heights and thecoastline extraction results in 2010 and 2011 should be almostcompletely consistent with the actual coastline heights

333 Validation of Coastline Classification Results In orderto ensure the accuracy of the coastline classification resultsthe supervised classification method was used to completethe automatic classification of the coastline remote sensingimages and to carry out sampling secondary interpretationand confirmation by means of the artificial visual inter-pretation method which ensured the accuracy of coastlineproperties and the distribution boundary of the classificationresults From the coastline extraction results over the time-frame 2000 samples were extracted for analysis As per theprobability calculation the ratio of conforming samples wasup to 95 or above which met the required standards

4 Analysis of Coastline LengthChanges and Trends

41 Variation of Coastline Length over the Years Variationsin coastline length reflect the overall impact of coastlineresource utilization natural erosion and siltation oceanicdynamics and other factors Changes and trends in thecoastline length depend on the development speed and trendsbetween natural coastline artificial coastline and estuarycoastline

It can be seen from Table 2 that over the past five yearsexcept for when the coastline length decreased slightly in2009 the total length of Chinarsquos coastline increased From2007-2008 and 2009ndash2011 the rate of increase of Chinarsquos


artificial coastline was greater than the rate of decrease of nat-ural coastline and estuary coastline so the overall coastlinelength increased In 2009 the artificial coastline increased by436368m less than the rate of increase in 2008 The naturalcoastline and the estuary coastline decreased by 349970mmore than the rate of decrease in 2008 Analysis showed thatthe main reason for this was the formal implementation ofChinarsquos Sea Reclamation Scheme and Management Systemin 2009 The area of sea reclamation (mostly including thecutoff type of reclamation) was increased by 688738 hectaresover 2008 levels while the coastline length slightly decreasedin 2009 After 2010 when the coastal marine administrativedepartments at all levels began making considerable effortson sea reclamation management and paid more attentionto sea reclamation methods which resulted in larger pro-portions of offshore type (island type) and convex barriertype reclamations and combined with the enhancement ofenvironmental protection and ecological awareness the rateof decrease of the natural coastline and estuary coastlineslowed while the coastline length increased

42 Coastline Changes and Trends in the Future

421 PredictionMethods There was a big difference betweenthe method of endpoint rate and the average rate as cal-culating the coastline change rates It was known from thecoastline change rate calculation method and the impactfactor analysis made by Jingfu et al [26] that a changein coastline length should be regarded as a periodic andoscillating change and that the rate calculation methodshould not be used for prediction of future variation trendsHaving only five stages of data the total change in lengthof the coastline should be predicted using the grey sequenceprediction method

422 Future Variation Trend Using the grey predictionmethod the coastline length values extracted from 2007ndash2011 were included into the grey sequence prediction modelThe development factor (119886 = minus001) and the grey action(119887 =1835774) should be calculated and the grey predictionmodel GM (1 1) of coastline should be obtained

119883(1)(119905+1)

= [119883(0)(1)

minus 263735624] 119890(001119905)

minus 263735624

(10)

where 119883(1)(119905+1) represents the predicted accumulated valueof 119905 + 1 time 119883(0)(1) represents the original value of startingtime The accumulative length of coastline (in 2008ndash2015)should be calculated as per (10) The former term shouldbe subtracted by the later term and the coastline predictionresults were shown in Table 4 The variance ratio was 0391and the small error probability was 1 which was biggerthan 095 It was shown that the accuracy of predictedresults was consistent with the requirements Therefore thepredicted value of the mainland coastline length obtainedby the grey prediction method was in accordance with theactual conditions The predicted results showed that thefuture mainland coastline length would increase and Chinarsquos

Table 4 Chinarsquos coastline length prediction table

Year Coastlinelength (m)

Predictionlength (m)

Residual(m)

Relativeerror

2007 18501296 185012962008 18515830 18549380 minus3355033 0002009 18599902 18678455 minus7855297 0002010 18644768 18808427 minus16365973 0012011 18946699 18939304 739430 0002012 185012962013 1132504562014 193374252015 19471983

mainland coastline length would be up to about 19471983mby 2015

5 Conclusions and Discussions

Through the data mining algorithm the preprocessing ofChinarsquos low-precision remote sensing images taken in theprevious five years the extraction of coastline lengths andtypes from 2007ndash2011 and the analysis of coastline changesand trends the following conclusions can be made

(1) The multisource imaging matching algorithm of aSIFT operator the geometric correction of a fewcontrol point satellite images the color-transferringalgorithm for maintaining the consistency of typ-ical surface features toning processing technologyand the automatic matching-line generation imagemosaic technology were used to preprocess the low-precision environmental disaster reduction satelliteand CBERS-01 satellite images Combined these areall effective methods

(2) The coastline can be extracted using the chromaticaberration Canny operator edge detection methodThe Kappa coefficient should be greater than 06throughout the calculation which shows that theextraction results are consistent with the actual mea-sured results By analysing the coastline type usingthe spectrum feature database trends identify adecreased natural coastline and estuary coastline butan increased artificial coastline were found Thesetrends are consistent with the current utilization ofthe coastline area in China

(3) The total length of Chinarsquos coastline increased overallThis is despite the coastline decreasing slightly in2009 due to the formal implementation of ChinarsquosSea Reclamation Scheme By using the grey predic-tion method Chinarsquos coastline length is predicted toincrease in the future with the total length reaching19471983m by 2015

With the continuous increase of artificial coastline lengthand related economic benefits the balance of the marine


ecosystem will be disturbed affecting the harmony betweenhumans and nature In order to ensure the sustainableexploitation and utilization of coastal resources and thereasonable adjustment of coastal development methods theimpact of coastline change trends on the marine ecologicalenvironment and the changes of spatial position should bestudied further

Acknowledgments

This work was financially supported by the Public Scienceand Technology Research Funds Projects of Ocean (no201005011) and Foundation of Key Laboratory of Sea AreasManagement Technology SOA (201107)

References

[1] Z Li and W Jigang ldquoCoastline GPS real-time dynamic mea-surement technology and error impactrdquo Surveying andMappingSciences vol 3 pp 9ndash12 2008

[2] J S Lee I Jurkevich P Dewaele P Wambacq and A Oost-erlinck ldquoSpeckle filtering of synthetic aperture radar images areviewrdquo Remote Sensing Reviews vol 8 no 4 pp 313ndash340 1994

[3] P Manavalan P Sathyanath and G L Rajegowda ldquoDigitalimage analysis techniques to estimate waterspread for capacityevaluations of reservoirsrdquo Photogrammetric Engineering andRemote Sensing vol 59 no 9 pp 1389ndash1395 1993

[4] R Holman J Stanley and T Ozkan-Haller ldquoApplying videosensor networks to nearshore environment monitoringrdquo IEEEPervasive Computing vol 2 no 4 pp 14ndash21 2003

[5] L I Rudin S Osher and E Fatemi ldquoNonlinear total variationbased noise removal algorithmsrdquo Physica D vol 60 no 1ndash4 pp259ndash268 1992

[6] T F Chan S Osher and J Shen ldquoThe digital TV filter andnonlinear denoisingrdquo IEEE Transactions on Image Processingvol 10 no 2 pp 231ndash241 2001

[7] D L Donoho ldquoOrthonormal ridgelets and linear singularitiesrdquoSIAM Journal onMathematical Analysis vol 31 no 5 pp 1062ndash1099 2000

[8] M Xiaofeng Z Dongzhi Z Fengshou et al ldquoStudy on thecoastline satellite remote sensing extraction methodrdquo RemoteSensing Technology and Application vol 4 pp 575ndash580 2007

[9] Z Chaoyang Study on the Remote Sensing Image CoastlineExtraction and Its Change Detection Technology Surveying andMapping Institute PLA Information Engineering UniversityZhengzhou China 2006

[10] D Jiang Y Huang D Zhuang Y Zhu X Xu et al ldquoA simplesemi-automatic approach for land cover classification frommultispectral remote sensing imageryrdquo PLoS ONE vol 7 no 9Article ID e45889 2012

[11] D G Stavrakoudis J B Theocharis and G C Zalidis ldquoAboosted genetic fuzzy classifier for land cover classification ofremote sensing imageryrdquo ISPRS Journal of Photogrammetry andRemote Sensing vol 66 no 4 pp 529ndash544 2011

[12] Z Xiaocheng and W Xiaoqin ldquoStudy on the remote sensingimage data miningrdquo Remote Sensing Information vol 3 pp 58ndash62 2005

[13] G Dengke Study on the GIS-Based Spatial Data Method XirsquoanUniversity of Electronic Science and Technology 2010

[14] L Deren W Shuliang and L Deyi Spatial Data MiningTheoryand Application Science and Technology Press Beijing China2006

[15] W Zhanquan Study on the Key Spatial Data Mining Tech-nologies Based on the Geographic Information System ZhejiangUniversity 2005

[16] D Kaichang L Deren and L Deyi ldquoCloud theory and itsapplication in the spatial datamining and knowledge discoveryrdquoChinese Journal of Image and Graphics vol 4 no 11 pp 924ndash929 1999

[17] L Longshu N Zhiwei and L Cheng ldquoResearch and practice onthe spatial attribute datamining based on the rough setrdquo Journalof System Simulation vol 14 no 12 pp 1702ndash1705 2002

[18] L Aihua Y Jianwei M Tushu et al ldquoResearch and systemimplementation of coastline change trend prediction methodrdquoSurveying andMapping Sciences vol 34 no 4 pp 109ndash110 2009

[19] L Deren and S Juliang ldquoThe wavelet and its application inimage edge detectionrdquo ISPRS Journal of Photogrammetry andRemote Sensing vol 22 no 2 pp 4ndash111120 1993

[20] Z Guobao Y Hua and CWeinan ldquoMulti-scale edge extractionbased on the biorthogonal wavelet transformrdquo Chinese Journalof Image and Graphics vol 3 no 8 pp 651ndash654 1998

[21] D Tao and Z Bin ldquoStudy on the coastline positions determinedby remote sensing image based on the analysis of wavelettechnologyrdquoMarine Science vol 4 pp 19ndash21 1999

[22] M Kantardzic S Siqing C Yin et al Data Mining ConceptModel Method and Algorithm Tsinghua University Press 2003

[23] F Wu ldquoCalibration of stochastic cellular automata the appli-cation to rural-urban land conversionsrdquo International Journalof Geographical Information Science vol 16 no 8 pp 795ndash8182002

[24] N Blackman and J Koval ldquoInterval estimation for CohenrsquosKappa as a measure of agreementrdquo Statistics in Medicine vol19 no 5 pp 723ndash741 2000

[25] J R Landis and G G Koch ldquoThe measurement of observeragreement for categorical datardquo Biometrics vol 33 no 4 pp671ndash679 1977

[26] L Jingfu Y Ping R Jingco et al ldquoSelection of methodsfor calculating shoreline change rate and analysis of affectingfactorrdquo Advances in Marine Science vol 21 no 1 pp 52ndash53

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MathematicsJournal of


Mathematical Problems in Engineering

Hindawi Publishing Corporationhttpwwwhindawicom

Differential EquationsInternational Journal of

Volume 2014

Applied MathematicsJournal of


Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of


CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of


Operations ResearchAdvances in

Journal of


Function Spaces

Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of Mathematics and Mathematical Sciences


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Algebra

Discrete Dynamics in Nature and Society



Decision SciencesAdvances in

Discrete MathematicsJournal of


Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Stochastic AnalysisInternational Journal of


Remote sensing data selection

Data preprocessing

Data preparation

Data mining

Result interpretation

Image registration correction toning mosaic and other pretreatment processes

Coastline extraction and verification based on the chromatic aberration Canny operator

Coastline result analysisand trend prediction

Figure 1 Coastline extraction flow chart based on data mining

the future variation trend of coastline length bymeans of greyprediction algorithm to achieve the resulting conclusions andto propose feasible recommendations for the exploitation ofthe future coastal zone

2 Coastline Extraction Process

Data mining refers to the process as discussed by Kantardzicet al [22] in which implicit unknown potential and usefulinformation and knowledge can be extracted from a largenumber of incomplete noisy fuzzy and random databasesAccording to the knowledge structure of data mining thecoastline extraction processes can be divided into four stagesas follows in Figure 1 image selection image registrationcorrection toningmosaic and other pretreatment processescoastline extraction and mining and the coastline resultanalysis and trend prediction

3 Study on the Remote Sensing ImageCoastline Extraction Technology

31 Data Preparation The remote sensing image data pre-sented and selected in this paper were collected at five stagesfrom 2007 to 2011 The space range covered the coastal zonesof the Chinese Mainland (including Hong Kong and Macao)and Hainan Island encompassing 290 scenic spots Details ofthe images are as in Table 1

32 Data Preprocessing

321 Automatic Registration of the Coastal-Zone Image Themultisource image matching algorithm of a SIFT operatorwas presented in this paper which provided different spectralband setting imaging model reflectivity dimension andother aspects of the multisource data RANSAC and theidentical point triangulation network construction methodwere used for eliminating the mismatching points in orderto achieve the automatic matching of multisource remote

Table 1 Image details in 2007ndash2011

Period Satellitedesignation

Spatialresolution (m)

Quantity(scenic spots)

2011 HJ-1A 30 652010 HJ-1A 30 422009 HJ-1A 30 342008 CBERS-01 20 862007 CBERS-01 20 63HJ-1A satellite is the abbreviation of environmental and disaster monitoringand forecasting CBERS-01 satellite was jointly invested in and developed byChina and Brazil on October 14 1999 Because HJ-1A images and CBERS-01images are low-resolution areas the results of shoreline extracted have smalldifferences but this does not affect this study

sensing images and control point image databases afterwhich the automatic image registration was completed

322 Geometric Correction of the Coastal-Zone Image Itwas difficult to find the control points due to the largewater area on the remote sensing image at the coastal zoneMeanwhile a small number of control points were used toachieve the geometric correction of the low-precision remotesensing images in order to reduce the workload of the fieldmeasurements Combined with satellite orbit parameters andthe geometric correction model with correction precisionin different satellite data sources and terrain conditions thegeometric correction ofmultiscene imageswas jointly carriedout to achieve the geometric correction of satellite images atseveral control points in the use of the correlation betweensubscene overlapping regions

323 Automatic Toning and Mosaic of Coastal-Zone ImageImage data toning processing was carried out by using thecolor-transferring algorithm to maintain the consistency oftypical surface features In the fuzzy classificationmethod theoverlapping region calculation method and image multiscalesegmentation method were used for the study on the colortransferring algorithm to maintain the consistency of typicalsurface features The coastal-zone remote sensing image wasspliced by using the automatic matching-line generationimage mosaic technology

33 Data Mining

331 Coastline Extraction Based on Edge Detection Edgedetection methods were widely used for the automaticextraction of the water sideline so as to significantly reducethe quantity of data remove the irrelevant information andretain the important structural properties of the imagesThe methods consisted of Roberts Cross operator Prewittoperator Sobel operator Canny operator Kirsch operatorand Compass operator in which the Canny operator canbe regarded as the most common edge detection methodIn order to compensate for the insufficient remote sensingimage detection of complex surface features edge featuresor broken linear characteristics by the conventional Canny





119863119909[119894 119895] = CD (119868 [119894 + 1 119895] 119894 minus 1 119895) (1)


119863119910[119894 119895] = CD (119868 [119894 119895 + 1] 119894 119895 minus 1) (2)


119863135

∘ [119894 119895] = CD (119868 [119894 + 1 119895 + 1] 119868 [119894 minus 1 119895 minus 1]) (3)


11986345

∘ [119894 119895] = CD (119868 [119894 minus 1 119895 + 1] 119868 [119894 + 1 119895 minus 1])

CD (119860 119861) = [(119871119860minus 119871119861)2

+ (119860119860minus 119860119861)2

+ (119861119860minus 119861119861)2

]

12

(4)



CDC [119894 119895]

= radic119863119909[119894 119895]2

+ 119863119910[119894 119895]2

+ 119863135

∘[119894 119895]2

+ 11986345

∘[119894 119895]2

(5)


120579 [119894 119895] = arctan(119863119910[119894 119895]

119863119909[119894 119895]

) (6)



120591high = (1 minus 120573) 120591119867 + 120573120591ℎ

120591Low = (1 minus 120573) 120591119871 + 120573120591119897(7)













2007 18501296 9383095 8980905 1372972008 18515830 9730865 8651800 1331652009 18599902 10167232 8301830 1308392010 18644768 10378360 8136654 1297542011 18946699 10733161 8085561 127977



119870 =

(119875119900minus 119875119862)

(1 minus 119875119862)

(8)




ratio in which 119875119900= 119904119899 119875




119868 =

(119871 minus 1198711015840

)

119871

(9)














119883(1)(119905+1)

= [119883(0)(1)

minus 263735624] 119890(001119905)

minus 263735624

(10)





Residual(m)

Relativeerror











Acknowledgments


References


































Volume 2014




Journal of











Journal of


Function Spaces






Algebra













119863119909[119894 119895] = CD (119868 [119894 + 1 119895] 119894 minus 1 119895) (1)


119863119910[119894 119895] = CD (119868 [119894 119895 + 1] 119894 119895 minus 1) (2)


119863135

∘ [119894 119895] = CD (119868 [119894 + 1 119895 + 1] 119868 [119894 minus 1 119895 minus 1]) (3)


11986345

∘ [119894 119895] = CD (119868 [119894 minus 1 119895 + 1] 119868 [119894 + 1 119895 minus 1])

CD (119860 119861) = [(119871119860minus 119871119861)2

+ (119860119860minus 119860119861)2

+ (119861119860minus 119861119861)2

]

12

(4)



CDC [119894 119895]

= radic119863119909[119894 119895]2

+ 119863119910[119894 119895]2

+ 119863135

∘[119894 119895]2

+ 11986345

∘[119894 119895]2

(5)


120579 [119894 119895] = arctan(119863119910[119894 119895]

119863119909[119894 119895]

) (6)



120591high = (1 minus 120573) 120591119867 + 120573120591ℎ

120591Low = (1 minus 120573) 120591119871 + 120573120591119897(7)













2007 18501296 9383095 8980905 1372972008 18515830 9730865 8651800 1331652009 18599902 10167232 8301830 1308392010 18644768 10378360 8136654 1297542011 18946699 10733161 8085561 127977



119870 =

(119875119900minus 119875119862)

(1 minus 119875119862)

(8)




ratio in which 119875119900= 119904119899 119875




119868 =

(119871 minus 1198711015840

)

119871

(9)














119883(1)(119905+1)

= [119883(0)(1)

minus 263735624] 119890(001119905)

minus 263735624

(10)





Residual(m)

Relativeerror











Acknowledgments


References


































Volume 2014




Journal of











Journal of


Function Spaces






Algebra















2007 18501296 9383095 8980905 1372972008 18515830 9730865 8651800 1331652009 18599902 10167232 8301830 1308392010 18644768 10378360 8136654 1297542011 18946699 10733161 8085561 127977



119870 =

(119875119900minus 119875119862)

(1 minus 119875119862)

(8)




ratio in which 119875119900= 119904119899 119875




119868 =

(119871 minus 1198711015840

)

119871

(9)














119883(1)(119905+1)

= [119883(0)(1)

minus 263735624] 119890(001119905)

minus 263735624

(10)





Residual(m)

Relativeerror











Acknowledgments


References


































Volume 2014




Journal of











Journal of


Function Spaces






Algebra














119883(1)(119905+1)

= [119883(0)(1)

minus 263735624] 119890(001119905)

minus 263735624

(10)





Residual(m)

Relativeerror











Acknowledgments


References


































Volume 2014




Journal of











Journal of


Function Spaces






Algebra











Acknowledgments


References


































Volume 2014




Journal of











Journal of


Function Spaces






Algebra
















Volume 2014




Journal of











Journal of


Function Spaces






Algebra









research article a study on coastline extraction and its

Documents