45258.高光譜遙感影像中小目標探測技術研究

1、中國科學院合肥物質(zhì)科學研究院博士學位論文高光譜遙感影像中小目標探測技術研究姓名：尋麗娜申請學位級別：博士專業(yè)：光學指導教師：方勇華20080501AbstractTheResearchofSmallTargetDetectionTechniquesforHyperspectralRemoteSensingImageryXunLina(Optics)DirectedbyProfFangYonghuaABSTRACTHyperspectra

2、lremotesensingimageryprovidesawealthofspatial，radiativeandspectralinformationwhichhelptoimprovetheeffectsoftargetdetectionandclassificationThisdissertationexploresthetheoriesandmethodsforsmalltargetdetectioninhyperspectr

3、alimageryAutomatictargetdetectionalgorithmsinblindenvironmentsforpurepixelsmalltargetandsubpixeltargetarediscussedAsaresult，theycanprovidethefoundationforbig—areamonitoringFirstly，thecharacteristicsofhyperspectralimagery

4、anditsinfluenceontargetdetectionaresystematicallysummarizedandanalyzedThen，threedifferentalgorithmstoautomatictargetdetectionforpurepixelsmalltargetsareproposedAccordingtothefactthathyperspectraldatahasdisplayedconvexcon

5、einitsfeaturespace，thealgorithmbasedonprojectionpursuit(PP)searchesforinterestinglow。dimensionalprojections，whichcanseparatesmalltargetsfromthebackground，viaaprojectionindexwhichissusceptibletothedistributionofsmalltarge

6、tsThesecondalgorithmisbasedonendmemberextractionandmatchedfilterInordertosuppressthebackgroundinformation，thehyperspectraldatacubeisprojectedintotheorthogonalsubspaceofbackgroundspectralsignatureswhichcanbeobtainedbyprin

7、cipalcomponentanalysis(PCA)becauseoftargets’lowprobabilitiesExtractingtheendmemberofsmalltargetsanddetectingthesmalltargetsarebothimplementedinthisprojectedsubspaceTheproposedtwoalgorithmsareapplicabletodetectsmalltarget