联合盲分解与稀疏表达的高光谱图像异常目标检测
jointblindunmixin gands p arsere p resentationforanomal y detectioninh yp ers p ectralima g e
-
摘要: 应用高光谱图像进行异常目标检测是高光谱遥感最重要的应用之一,而异常目标检测算法最为关键的是对背景的描述。rx 等经典算法受制于对背景分布的高斯假设,因而在复杂背景条件下不能有效地提取出感兴趣的异常目标。本文提出了一种新的异常目标检测算法,不仅能够有效地检测出亚像元的异常目 标，同时以新的方式描述背景。算法首先针对异常检测先验信息不足的问题,采用盲分解方法建立描述背景的冗余字典,该字典是根据像元的纯净性定义估计的背景类端元束构成;然后采用稀疏回归计算每个像元的重建误差,以误差特征作为异常指数,误差越大越可能是异常;为了增强对可能异常目标的描述能力,应用了局 部近邻分析来增强目标在图像邻域的离群表达,从而获得最终的异常检测特征。算法将字典构造的全局性与地物的局部连续性结合,提高了异常目标检测的可靠性。采用不同混合比例模拟的亚像元数据和两幅真实数据进行实验,结果表明,算法不仅仅获得了比 rx 等经典算法更高的精度,同时在不同信噪比条件下表现稳健且抗噪能力强。Abstract: anomal ytar g etdetectionisanimp ortantissueinh yp ers p ectralremotesensin g，however，howtomodeltheback g roundisamostdifficultp roblem.thetraditionalrxal g orithmisrestrictedb ynon-g aussiandistributionoftheback g round.theob j ectiveofourworkistodevelo panewrobustanomal ydetectional g orithm.thisnewal g orithmwasabletofindsub-p ixeltar g ets;wealsop resentanewback g roundre p resentationmodel.theback g roundwasmodeledb yadictionar ycomp osedofrel-ativel yp ureback g roundendmemberbundlesthatwereconstructedb yablindunmixin gal g orithm.ever yp ixel intheh yp ers p ectral ima g ewasmodeledb ys p arsere g ressionusin gthedictionar y.there-constructionerrorwasusedasanomal yfeature;thosep ixelsthathavelar g ere g ressionreconstructionerrorsarethep otentialanomal ytar g ets.finall y,adualwindowbasednearestnei g hboranal y siswasusedtoenhancetheanomal ylikefeatures.thisal g orithmj oinedg lobalandlocal informationtog uar-anteethereliabilit y.ascomp aredtotheclassicalrxal g orithms,thep ro p osedal g orithmp erformedver ywellwithsimulateddata,inwhichthesub-p ixel tar g etwasconstructedb ytar g etandback g roundsi g nalswithdifferentmixedfractionandp ollutedb ynoise.tworealdataex p erimentsalsoconfirmedtheeffectiveness.
下载:
