高分辨率遥感影像分割及地物轮廓提取方法研究

《高分辨率遥感影像分割及地物轮廓提取方法研究》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、北京交通大学硕士学位论文高分辨率遥感影像分割及地物轮廓提取方法研究姓名：沈伟申请学位级别：硕士专业：摄影测量与遥感指导教师：谭衢霖20081201ABSTRACTABSTRACT：Withtherapiddevelopmentofspacetechniquesandinformationtechniques，thehigh-resolutionremotesensingsatellitec姐supplyhumanwiththehighaccuracyandplentifuldigitalimages．High-resolutionremotesensingi

2、magesmakeUSobserveearthchangesinasmallerspacescales．Theimportanceofhigh-resolutionremotesensinginformationextractionisinincreasing．Now,forstudyofhigh-resolutionremotesensingimagesegmentationandobjectcontourextractionhasbecomethekeyresearchsubjectofhigh-resolutionremotesensingapplic

3、ation，onthissubject，thispaperfocusonfollowingaspects：(1)Thispaperanalysesthecharacteristicsofhigh-resolutionremotesensingimagesandthekeyofhigh-resolutionremotesensingimagesegmentationandinformationextraction．Andproofthelimitationsofpixel-basedspectralsegmentationwimexperiments．(2)T

4、hispaperstudyontheobject-orientedmulti—scalesegmentationalgorithm，useeCognition,experimentalconductingmulti—scaleimagesegmentationonbuildingsandroads，andgetthebest-scaleparametersabouttheimage．．(3)StudyonthealgorithmofMeanshiflsegmentationbasedonconfidence，ACMalgorithm，thetwoalgori

5、thmsUSedinhigh-resolutionremotesensingimageanalysisapplication．Combinatorialexperimentsweremadewithdifferentsortsobjectforextractingobjectcontour,analyzethecharacteristicsofalgorithms．(4)Withanumberofsegmentationmethods，manyexperiments，comparedtheresultofmulti-scalesegmentation,bas

6、edontheMeanshiflsegmentationbasedonconfidenceandbinarysegmentation，afterstatisticalanalyzeextractionofbuildingsandroadscontourinformation；carryouttheevaluationofthestatisticalresults．(5)Aftercomparisonthispapershow：ObjectOrientedMulti-scalesegmentationcouldgetbetterresultsonrelativ

7、elylargebuildingscontourextraction；Meanshiffsegmentationbasedonconfidencecoulddobetteronrelativelysmallbuildingscontourextraction；withregardtotheroadscontourinformation，binarysegmentationisquitesuitable。KEYWORDS：High-resolutionremoteSenSeimages；Objectcontour；Multi—scaleSegmentation

8、；CLASSNO：TP75l：P237独创性声明本人