无监督深高光谱

无监督深高光谱

ID:39590780

大小:2.78 MB

页数:21页

时间:2019-07-06

无监督深高光谱_第1页
无监督深高光谱_第2页
无监督深高光谱_第3页
无监督深高光谱_第4页
无监督深高光谱_第5页
资源描述:

《无监督深高光谱》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库

1、sensorsArticleAnUnsupervisedDeepHyperspectralAnomalyDetectorNingMa1,YuPeng1,ShaojunWang1,*andPhilipH.W.Leong2ID1DepartmentofAutomaticTestandControl,SchoolofElectricalEngineeringandAutomation,HarbinInstituteofTechnology,Harbin150080,China;maning@hit.edu.cn(N.M.);pengyu@hit.edu.cn(Y.P.)2SchoolofE

2、lectricalandInformationEngineering,TheUniversityofSydney,Sydney2006,Australia;philip.leong@sydney.edu.au*Correspondence:wangsj@hit.edu.cn;Tel.:+86-451-86413532(ext.439)Received:29November2017;Accepted:22February2018;Published:26February2018Abstract:Hyperspectralimage(HSI)baseddetectionhasattrac

3、tedconsiderableattentionrecentlyinagriculture,environmentalprotectionandmilitaryapplicationsasdifferentwavelengthsoflightcanbeadvantageouslyusedtodiscriminatedifferenttypesofobjects.Unfortunately,estimatingthebackgrounddistributionandthedetectionofinterestinglocalobjectsisnotstraightforward,and

4、anomalydetectorsmaygivefalsealarms.Inthispaper,aDeepBeliefNetwork(DBN)basedanomalydetectorisproposed.Thehigh-levelfeaturesandreconstructionerrorsarelearnedthroughthenetworkinamannerwhichisnotaffectedbypreviousbackgrounddistributionassumption.Toreducecontaminationbylocalanomalies,adaptiveweights

5、areconstructedfromreconstructionerrorsandstatisticalinformation.ByusingthecodeimagewhichisgeneratedduringtheinferenceofDBNandmodifiedbyadaptivelyupdatedweights,alocalEuclideandistancebetweenundertestpixelsandtheirneighboringpixelsisusedtodeterminetheanomalytargets.Experimentalresultsonsynthetica

6、ndrecordedHSIdatasetsshowtheperformanceofproposedmethodoutperformstheclassicglobalReed-Xiaolidetector(RXD),localRXdetector(LRXD)andthe-state-of-the-artCollaborativeRepresentationdetector(CRD).Keywords:hyperspectralimage;deeplearning;anomalydetection1.IntroductionAnHSIanomalytargetisgenerallydefi

7、nedasapixelorobjectwhichhasloweroccurrenceprobabilitythanthebackground[1].Examplesincludepixelsfromaburningtreeinaforestoroilspillsinthesea.Utilizinghundredsofverynarrowandcontinuousspectralbandsandspatialinformation,HSIcanbetterd

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。