欢迎来到天天文库
浏览记录
ID:31941525
大小:6.93 MB
页数:70页
时间:2019-01-29
《高光谱图像亚像元级目标检测的非线性方法-.研究》由会员上传分享,免费在线阅读,更多相关内容在应用文档-天天文库。
1、杭州电子科技大学硕士学位论文和背景光谱未知时,端元的选择十分重要,直接影响后续的目标检测、分类、识别等结果。于是本文提出了一种核非监督正交子空间投影(KUOSP)方法,用该方法来实现端元自动提取,实验结果表明该方法自动提取端元合理,且效果佳。(4)前面的研究工作主要用MATLAB仿真实现,而这部分用VC++开发了一个高光谱遥感图像亚像元目标检测软件系统,该系统有打开GeoTiff格式图像、区域选择、OBS端元提取、丰度估计(包括一般的目标检测方法,如SSP、OBSP;还有结合了核函数的相应检测方法,如KSSP、KOBSP;以及运用组合核将空间信息与光谱信息综合
2、利用起来的新方法一CKSSP)等功能,其中目标检测是本系统的突出应用部分。关键词:高光谱图像目标检测,非线性方法,亚像元,扩展数学形态学,空间信息._蒸一≯‘鍪瓣,ABSTRACTTargetdetectionforhyperspectralimageryplaysanimportantroleintheresea。cho±remotesensingtheoryanditsapplication.Forhyperspectralimageryoftenha、scharactersasfollows:highdimension,nonlinearrelation
3、shipbetweenbands,mixedpixels,s’hemateralwithdifferentspectrum,etc,thentheperformanceoftraditionallinearhyperspectraltargetdetectionalgorithmsaretendtobeinfluenced.Usingkernelmethodisaneffectivewaytoprocessthenonline‘arinformation,butstillfaceschallenges.Besides,owingtothelimitationo
4、fspatialresolutionforhyperspectralimagery,aswellasthecomplexityanddefferenceamongearthmaterials,someinterestedtargetscouldonlyexistassubpixeIs,whichmeansthatthereareonlymixedpixelsconsistedofinterestedtargetsandothermateralsintheimagery.Hence,howtoeffectivelyraisetheabilityoftargetd
5、etectiononsubpixelsiSacriticalproblemforfurtherdevelopmentofhyperspectralremotesensingimagery.Inaddition,anhyperspectralimageryisusuallynotamerecollectionofindependentandidenticallydistributedpixels.asassumedinpixel—basedimageclassification,butitconstitutesastructureddomain:spatiall
6、yclosepixelsshouldintuitivelybelongtothesameclass.especiallywiththeadventofVHRsensors,whererobusttechniquesareneededtodealwiththehighspatialcorrelationbetweenspectralresponsesofneighboringpixels.Andithasbeenpaidhighattentiontoeffectivelyusespacialinformationofagivenhyperspectralimag
7、ery.Thisdissertationmainlyresearchonhowtoextendalgorithmsbasedonlinearspectralmixturemodeltononlinearfeaturespacebykerneltrickforresolvingthedifficultiesinnonlinearunmixing.Andmainlystudyspatialinformationextractioninagivenhyperspectralimagerybyusingextendedmathematicalmorphological
8、methods,aswellasinc
此文档下载收益归作者所有