欢迎来到天天文库
浏览记录
ID:53768679
大小:2.64 MB
页数:8页
时间:2020-04-25
《一键式多特征遥感图像采样器-论文.pdf》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、第29卷第2期遥感信息V01.29,No.22014年4月RemoteSensingInformationApr.,2014一键式多特征遥感图像采样器金叶蒙1,虞思逸1,周坚华2(1.华东师范大学地理系,上海200241;2.华东师范大学地理信息科学教育部重点实验室,上海200241)摘要:本文提出了一键式多特征遥感图像采样器(MultipleFeaturePointSampler,MFPS)的思想,它以多特征串行搜索为基础、通过设计适合特定特征的搜索算法,实现一键式多特征采样。该算法允许用户定制采样特征空间和窗口尺寸,该采样器能够响应用户的一键式点选、再以采样点为中心的窗口内计算
2、特征阈值、并生长出与点选位置具有多特征相似性的采样区域。相较于传统的单特征采样,该算法能比较充分地利用用户提供的目标物和背景之间的多特征差异,因此在提高采样效率的同时,进一步提高了样本对用户判读知识的代表性。整个算法经过MATLAB仿真测试,测试结果表明MFPS能够快速获得具有较好代表性的样本,运算开销适中,并具有较好的普适性。与仅依据波谱信息的传统点采样方法比较,以MFPS采样做监督分类的全局精度平均提高了6.34%;但采样工作量只有传统点采样方法的五分之一。关键词:图像采样;多特征空间;串行搜索doi:10.3969/j.issn.1000—3177.2014.02.010中图
3、分类号:TP75文献标识码:A文章编号:1000一3177(2014)132一0056一07阳脚IlgI嘞寥S踟叫erBased蚰MIlltipleFeatur鹪儿NYe_men91,YUSi—yil,ZHOUJian-hua2(1.(为og,。吐声^ySfiP竹c已,。EⅡsf(_i挖nJVo,,,zn£L7矗i副Pr5打y,S,l口砣g^ni200241;2.KPyLn60rn£D7≯o,GPog,—n户^ic口ZJ咒/b,‘,以口£io挖Sci已咒fP,Mi行i5fryo厂巨H“f口£io行,Eas£C^i刀口No,‘mnzLki口Prsify,S^口粗g^ni200241)
4、Abstract:ThispaperproposedoneMultipleFeaturePointSampler(MFPS).Basedontheserialsearch,MFPSputsanewsamplingaIgorithmintopractice,whichischaracterizedbyone—buttonsamplingandmulti—featuredistinguishing.ThisalgorithmallowsuseTstocust。mizethemulti—featurespaceandthesizeofthesamplingwindow.1nrespons
5、etothepointsampllng,themaximumandtheminimumofeachfeaturecanbeobtainedfromthestatisticsoftheDixelvalueswithinthewindowwhichhasthesamplepointasitscenter.Andthenanirregularregioncanbeextendedfromthewindow.Allthepixelsintheregionhavemulti-featuredsimilaritywiththosewithinthewindow.Comparedwiththet
6、raditionalsingle-featureandone.buttonsampler,MFPScanfullytakeadvantageofthemulti—featuredifferencesbetweentheobjectandthebackground,andthusincreasessamplingefficiencyandfurtherimprovestherepresentativenessofsamplesontheuserinterpretationknowledge.ThroughthesimulationtestswithMATLAB,ithasbeende
7、monstratedthatMFPSisnotonlyanefficienttooltoobtainsampleswithbetterrepresentatiVenessanduniVersality.ComparingwiththetTaditionalpointsamplingrelyingononlythespectralfeatures,thereis5.06%increasebyusingMFPSintheaverageglobalaccuracyofima
此文档下载收益归作者所有