高分辨率遥感影像城区道路提取方法研究

华中科技大学硕士学位论文高分辨率遥感影像城区道路提取方法研究姓名：赵晓锋申请学位级别：硕士专业：模式识别与智能系统指导教师：谭毅华2010-01-20I,IIAbstractRoadisoneofthemostimportantartificialfeatures.Asthecity'sbackbone,itsextractionandupdatingisofgreatsignificanceforurbanplanning,trafficmanagementandsoon.Inthepastfewdecades,althoughalotofsolutionisputforwardtoextractroadfromremotesensingimagery,mostofstudieshavefocusedontheroadextractionlow-resolutionimagery.Becauseofthecomplexityofroadcharacteristics,roadextractioninhighresolutionimagesisquitedifferrentfromthatinlowresolutionimages.Therefore,thestudyoftheurbanroadnetworkextractionhasimportanttheoreticalandpracticalsignificance.Underthiscircumstance,thispaperresearchestheurbanroadextractionofthehighresolutionremotesensingimagesbasedonthedeepanalysisofthecharacteristiccsofroads.Themainelementsareasfollows:Firstly,thepaperanalyzesthecharacteristicofroadsindifferentresolutionimages.Incombinationwithpreviousstudy,thepapersummarysthecharacteristicofurbanroadofthehighresolutionremotesensingimages,whichistheoreticalbasisofroadextraction.Secondly,anewroadsegmentationmethodbasedontotalvariationsandregiongrowingispresented.Experimentsshowthenewalgorithmscanseperatetheroadregionfromthebackground.Afterapplyingtheroadsegmentationmathod,fourtypicalregiongeometricfactorsareselectedtodetectthetypicalroadsegmentsandremovethenon-targetsegments.Finally,basedontheextractionofroadarea,thepaperusesmorphology,skeletonextractionandcurvefittingtoextracttheskeletonoftheroadnetwork.Basedontheexperimentsofseveralhigh-resolutionimages,theexperimentresultsshowthatthemethodproposedcanextractaccuratelyurbanroadnetworkfromhighresolutionremotesensingimages.Keywords:roadextraction,totalvariations,regiongrowing,curvefitting,vectorization□_____□“√”111.1,,1.22Snake[1-4][5-8][9-11][12][13-15]1Park[16]2Descombes[17]3Renaud[18]SnakeDoubleSnake4Liang[19]SEOHFCMSEOH[20][21]Snake5Long[22]MeanShift6[23]37[24]8Jiuxiang[25]RPαpPαP()idθ()dθ()dθPP9[26]10[27][28]T41.3522.12.1.1[29]1232.1.21RobertssobelKirsehLaplacian6LogCannyLOGCanny2127[29]342.2[29]2.2.182.2.22.2.32.392.3.12.3.2Vosselman[30][30-31]1210LTY3122.41133.13.1.1MN00111(,)(,)mnijRxyfijmn===×∑∑3-1Rf3.1.2MNMN123-13.1.32D22221(,)2xyGxyeσπσ+−=3-2σ1σ=2D3-22D(1σ=)2r+1(3-2):132222(,)pyrkxrkxrpyrkpGxyeσ=+=+=−=−+−=∑∑3-33-355(=1)174111444444471616161672626262674627313-3(a)b14cd3-4553-33.23.2.1TotalVariations[32]1)[()]Yyx=()yx15J[()]JyxYY00,yYyyY∈+∆∈[]Jy0y00[][]JJyyJy∆=+∆−3-4000[][,][,]Jyyyyyαβ∆=∆+∆3-50[,]yyα∆y∆0[,]yyβ∆y∆0[,]yyα∆[]Jy0y2[,]fxy()fdefTDJffdxdy=∇∫3-622Xyfff∇=+ffDfff3[32](,)221((,))(,)rxydefLTxyDJfxyuvffdudvρα=++∫u3-7(,)uvρ(,)1uvρ≡1((,))LTJuxy0α416[32](,)1(,)((,))((,))rxyDLTuvdudvLSfxyJfxyρ=∫u3-8((,))LSfxy(,)fxy50255[32]T2rσσ3-8117ab3-502553-5256118ar=3br=6cr=93-63-6r36192aT=50bT=603-7T30603203.2.2[33]3-42.2112341(,)0(,)0fxypxyelse==3-9(,)0fxy=xy213333(,)(,)jiijnxypxiyj===−=−=−−∑∑3-1033233(,)((,))jiijxyfxiyjFε===−=−=−−−∑∑3-113333(,)jiijFfxiyj===−=−=−−∑∑1(,)1&&(,)2nxythresholdxythresholdε≥≤3-12(,)1nxythreshold≥39(,)2xythresholdε≤1002|(,)|fxyZthreshold−3-13Zthreshold30342step1fxy,xyffI22(,)(,)(,)xyIxyfxyfxyα=++22step2r61step3I1I1I1(,)IxyT1(,)Ixy0255T=40~60step41(,)255Ixy=(,)fxy(,)fxy0step5(,)fxy2-92-10step6step73a23bc3-83-8bc1234243.33.2[23]12MinimumEnclosingRectangleMERMER325(0,1]4ab3-92eopNN=+3-14eNoN1263-1023-113273-1245a28b3-113-113.42944.1[29]4.1.1ASAS{:}ASlSlAΘ=+⊂4-1SlAASAdAS{:}ASAddA⊕=+∈4-2304.1.2()ASASS=Θ⊕D4-3AS()ASASS•=⊕Θ4-44.2SkeletonBlum[34][35]HildichZhangOPTA3110104.2.1HiltitchHilditch[36]Hilditch113573pppp+++≤20()1Sp=3812iip=≥∑4331S()=1pp=5551S()=1pp=0p2-13S()=1p30p=3p128,,...ppp01p2p1p3p0p4p5p6p7p84-10p4.2.2ZhangZhang[37]3218126iip=≤≤∑20()1Sp=31370ppp××=41570ppp××=18126iip=≤≤∑20()1Sp=31350ppp××=43570ppp××=T4.2.3OPTAOPTA[38]4-2OPTAPNP334-2OPTA104-3OPTA4-4OPTA4-34-41034abZhangcHilditchdOPTA4-54-5OPTAZhangZhang4.34.3.135101P84-6()Ol8111()2niiiCppp+==−∑91()pp=4-5p2p1p3pp4p5p6p7p84-6Pip2P8-81()niiSpp==∑4-64-5[50]1()1nCp=2361()3nCp=2()4nCp=3()2S()6nnCpp=≥3Sn(p)=04.3.2[39][39]VV045V±090V±V1804-737Zhang1()00,xy243244-88Freeman84-8818145818183801,,,,,...nxynddd,xynkdkk+1kd18()()()0011,,,,...,,mmijijij4.3.31spurTspurT20stack24stack3spurTL2439ab4-94.44.340[40]xy4.4.1Douglas-PeukerDouglas-Peuker[41][42]Douglas-Peuker0D0DT0DT0D0DTT4-104-10aCDP10DTCP1CP24-10b0DT41CDP1P3aCDP1P3bCDP1P3p4p2p5p6p7c4-10Douglas-Peuker[43][29]TTT42DP4.4.2S:(){},1,2,iiSxyin==yaxb=+4-7()21,()niiifabaxby==+−∑4-8ab(4-7)ab(4-8)ab:1111nnniiiiiiianxyxyd====−∑∑∑4-911111nnnniiiiiiiiiibnxyxxxyd=====−∑∑∑∑4-102211nniiiidnxx===−∑∑()21niiiQyabx==−−∑4-1143()11,ab()22,ab(x,y)1221121221(,)ababbbaaaa−−−−4-114-11T1=5T2=104.5[33]44[44]4-12L1ABL2CDL1L2L2L1L2CL1L1CBABCDrL1L2ßL1L2abL2L1ADCBL1L2cd4-12454.64-134-144-134-144-14464-154-154-164-17474-164-174-184-194-19485,123412323[1]W.M.Neuenschwander,P.Fua,L.Iverson,etal.ZiplockSnakes[J].InternationalJournalofComputerVision.1997,25(3):191-201[2]P.Fua,Y.G.Leclerc.ModelDrivenEdgeDetection[J].MachineVisionandApplication.1999,3(1):45-56[3].[D]..2006[4]P.Agouris,A.Stefanidis,S.Gyftakis.Diffe