资源描述：
《高分辨率卫星影像的道路提取》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、硕士学位论文高分辨率卫星影像的道路提取ROADEXTRACTIONFROMHIGHRESOLUTIONSATELLITEIMAGERYMAMHOUTRACHID(理查德)哈尔滨工业大学2018年6月国内图书分类号：TP751.1学校代码：10213国际图书分类号：621.3密级：公开工学硕士学位论文高分辨率卫星影像的道路提取硕士研究生：MAMHOUTRACHID(理查德)导师：张钧萍教授申请学位：工学硕士学科：信息与通信工程所在单位：电子与信息工程学院答辩日期：2018年6月授予学位单位：哈尔滨工业大学IIClassifiedIndex:TP751.1U.D.C.:621.
2、3DissertationfortheMasterDegreeinEngineeringROADEXTRACTIONFROMHIGHRESOLUTIONSATELLITEIMAGERYCandidate：MAMHOUTRACHIDSupervisor：Prof.ZhangJunpingAcademicDegreeAppliedfor：MasterofEngineeringSpeciality：InformationandCommunicationEngineeringAffiliation：SchoolofElectronicsandInformationEngineeri
3、ngDateofDefence：June,2018Degree-Conferring-Institution：HarbinInstituteofTechnologyIII哈尔滨工业大学工学硕士学位论文AbstractRoadnetworksplayanimportantroleinvariousapplicationssuchasurbanandruralplanning,infrastructureplanning,transportationmanagementandvehiclenavigation.Accurateroadnetworksextractionfrom
4、HighResolutionSatellite(HRS)imagesareofcentralimportanceinawidedomainofapplications.However,therearemanydifficultieswhichimpedestheextractionoftheroadnetworks,suchastheocclusionsoftreesandcars,shadowoftreesandhighbuilding,complexbackgrounds.Recently,deeplearningtechniquehasachievedimpressi
5、veperformanceonavarietyofproblems;suchasobjectdetection,speechrecognition,naturallanguageprocessing,visualrecognition,andimageclassification.TheConvolutionalNeuralNetwork(CNN)isoneofthemostimportanttypesofdeepneuralnetworks,andithasbeenwidlystudied.TheresearchonCNNhasbeenemergedquicklyanda
6、ccomplishedstate-of-the-artresultsonseveraltasks.ACNNisaspecialvariantofaneuralnetwork,whereconnectivitybetweenunitshavebeenconstrainedandparametersharingisemployed,whichreducetheparametersnumberinitsmodel.CNNcan,therefore,havemanyhiddenlayers,whichenablesthenetworktolearnahierarchicalrepr
7、esentationoftheinputdata.Byhavingalargedatasetandconductingabackpropagationprocess,thenetworkcanlearntoextractinformativefeaturesfromrawpixelvalues.Toaddresstheabovementionedcomplexissues,wesuggesttoemployaCNNalgorithm.Theproposedmethodisbasedonfourmainsteps.F
