資源描述:
《一種基于單形體正化的高光譜數(shù)據(jù)全約束線性解混方法》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在行業(yè)資料-天天文庫。
1、第35卷第5期紅外與毫米波學(xué)報Vol.35?No.52016年10月J.InfraredMillim.WavesOctober?2016文章編號:1001-9014(2016)05-0592-08DOI:10.11972/j.issn.1001-9014.2016.05.014一種基于單形體正化的高光譜數(shù)據(jù)全約束線性解混方法1?2?3?1?21?24許寧?耿修瑞?尤紅建?曹銀貴(1.中國科學(xué)院空間信息處理與應(yīng)用系統(tǒng)技術(shù)重點實驗室?北京100190?2.中國科學(xué)院電子學(xué)研究所?北京100190?3.中國科學(xué)院大學(xué)?北京100049?4.中國地質(zhì)大學(xué)?北京1000
2、83)摘要:在端元已知情況下?線性混合模型的非負(fù)約束最小二乘無閉式解?需要多次迭代得收斂最優(yōu)解?時間復(fù)雜度高.通過高光譜數(shù)據(jù)凸面幾何特性分析?指出當(dāng)數(shù)據(jù)為正單形體時?可經(jīng)有限步驟快速得線性混合模型最優(yōu)解.據(jù)此提出一種單形體正化的高光譜數(shù)據(jù)全約束線性解混方法?據(jù)已知端元進(jìn)行單形體正化?采用和為一約束求解豐度系數(shù)?最后迭代剔除豐度負(fù)值端元得全約束解.實驗結(jié)果表明該方法可獲得傳統(tǒng)全約束解一致的豐度估計?且效率大大提升.關(guān)鍵詞:高光譜數(shù)據(jù)?光譜解混?端元白化?單形體正化?全約束最小二乘中圖分類號:TP394.1文獻(xiàn)標(biāo)識碼:AAfullyconstrainedlinea
3、runmixingmethod:Simplexregularizationforhyperspectralimagery1?2?3?1?21?24XUNing?GENGXiu ̄Rui?YOUHong ̄Jian?CAOYin ̄Gui(1.KeyLaboratoryofTechnologyinGeo ̄spatialInformationProcessingandApplicationSystem?IECAS?Beijing100190?China?2.InstituteofElectronics?ChineseAcademyofSciences?Beijing10
4、0190?China?3.UniversityofChineseAcademyofSciences?Beijing100049?China?4.SchoolofLandScienceandTechnology?ChinaUniversityofGeosciences?Beijing100083?China)Abstract:Withaprioriinformationoftheknownendmembersinhyperspectralimage?thereisnoclosed ̄formsolutionofLeastSquare(LS)methodforlin
5、earmixingmodelundertheAbundanceNon ̄negativityConstraint(ANC).SomanyiterationswhichmayresultinbigcomputationalcomplexityareneededinthetraditionalFullyConstrainedLS(FCLS)methodstoobtaintheoptimalsolution.Inthispaper?ananalysisofimpactsonabundanceestimationofhyperspectalimageindifferen
6、tsimplexshapeswasimplementedandafullyconstrainedlinearunmixingmethodbasedonsimplexregulariza ̄tionwasproposedwhichcouldgetoptimalsolutionunderlimitediterationwhenthehyperspectralimagewasspannedintoaregularsimplex.Theproposedmethodwascarriedoutbythreesteps.Firstly?thesimplexofhyperspe
7、ctralimagewasregularizedbytheknownendmembers’whiteningmatrix.Second ̄ly?theanalyticalsolutionofabundancecoefficientswasobtainedunderAbundanceSum ̄to ̄oneCon ̄straint(ASC).Thenforeverypixel?theFCLSsolutionwasachievedbyeliminatingtheendmemberswithnegativeabundancecoefficientsandsolvingthe
8、ASCequationiterativ