• Aucun résultat trouvé

Sub-pixel detection in hyperspectral imaging with elliptically contoured t-distributed background

N/A
N/A
Protected

Academic year: 2021

Partager "Sub-pixel detection in hyperspectral imaging with elliptically contoured t-distributed background"

Copied!
6
0
0

Texte intégral

(1)

an author's

https://oatao.univ-toulouse.fr/26072

https://doi.org/10.1016/j.sigpro.2020.107662

Besson, Olivier and Vincent, François Sub-pixel detection in hyperspectral imaging with elliptically contoured

t-distributed background. (2020) Signal Processing, 175. 107662-107667. ISSN 0165-1684

(2)

Short

communication

Sub-pixel

detection

in

hyperspectral

imaging

with

elliptically

contoure

d

t

-distribute

d

background

Olivier

Besson

a,∗

,

François

Vincent

a

University of Toulouse, ISAE-SUPAERO, 10 Avenue Edouard Belin, 31055 Toulouse, France

Keywords: Detection

Generalized likelihood ratio test Hyperspectral imaging Replacement model Student distribution

a

b

s

t

r

a

c

t

Detectionofatargetwithknownspectralsignaturewhenthistargetmayoccupyonlyafractionofthe pixelisanimportantissueinhyperspectralimaging.Werecentlyderivedthegeneralizedlikelihoodratio test(GLRT)forsuchsub-pixeltargets,eitherfortheso-calledreplacementmodelwherethepresenceof atargetinducesadecreaseofthebackground power,duetothesum ofabundancesequaltoone,or foramixedmodelwhichalleviatessomeofthelimitationsofthereplacementmodel.Inbothcases,the backgroundwasassumedtobeGaussiandistributed.Theaimofthisshortcommunicationistoextend thesedetectorstothebroaderclassofellipticallycontoureddistributions,morepreciselymatrix-variate

t-distributionswithunknownmeanandcovariancematrix.Weshowthatthegeneralizedlikelihood ra-tiotestsinthet-distributedcasecoincidewiththeirGaussiancounterparts,whichconfersthelatteran increased generalityfor application.The performanceas wellas the robustnessofthesedetectors are evaluatedthroughnumericalsimulations.

1. Problemstatement

Hyperspectralimaginghasbecomeanincreasinglypopulartool for remote sensing and scene information retrieval, whether for civil ormilitaryneeds andin a largenumber ofapplications, in-cludinganalysisofthespectralcontentofsoils,vegetationor min-erals,detectionofman-madematerials orvehicles,tonameafew [1,2].Oneofthechallengesofhyperspectralimagingistodetecta target-whosespectralsignatureisassumedtobeknown-withina backgroundwhosestatisticalpropertiesarenotfullyknown[3–5]. Depending onthe spatialresolution ofhyperspectral sensors and thesizeofthetarget,thelattermayoccupythetotalityoronlya fractionofthepixelundertest(PUT),inwhichcaseonespeaksof sub-pixeltargets.Inthelattercase,thetargetreplacespartofthe backgroundinthePUT,leadingtotheso-calledreplacementmodel [6,7].

Whateverthe case,full-pixel orsub-pixel targets,the problem can be formulated asa conventional composite hypothesis prob-lem[3–9]:givena vectory∈Rp-wherepdenotesthenumberof spectral bandsused-whichrepresentsthereflectance inthePUT, is there a component along t-the signature of interest (SoI)-in addition to the background? Since the background statistics de-pend onunknown parameters (for instancemeanandcovariance

Corresponding author.

E-mail addresses: olivier.besson@isae-supaero.fr (O. Besson), francois.vincent@isae-supaero.fr (F. Vincent).

matrix) a set of trainingsamples Z∈Rp×n, hopefully free of the

SoIt,is observedwhosestatisticsare assumedtomatchthoseof thebackgroundinthePUT.Thesetrainingsamplesaregatheredin thevicinityofthePUT(localdetection)oralongthewholeimage (globaldetection).

Recentlyin[10]weaddressedsub-pixeldetectionusingthe re-placement model under a Gaussian background, and we derived the plain generalized likelihood ratio test (GLRT) by maximizing thejointdistributionof(y,Z)withrespecttoallunknown param-eters.Moreover,motivatedbysomelimitationsofthereplacement model,especiallythefact thatthefillingfactorofasub-pixel tar-getmay not be inpracticeas largeas expected, we alsoderived theGLRT fora mixedmodel where presence ofa target induces a partial replacement of the background [11]. These two detec-tors assume a Gaussian background. However, evidence of non-Gaussianity of hyperspectral data has been brought [12,13] and thereforeitisofinteresttoextenddetectorsoriginallydevisedfor Gaussianbackgroundtothebroaderclassofellipticallycontoured (EC)distributions[14,15].Theaimofthiscommunicationisthusto extendourrecentGLRTsfromtheGaussiancasetothematrix vari-atet-distributedcase.Wewillshowthat theGLRTscoincidewith theirGaussian counterparts. Thepaperisorganizedasfollows.In Section 2, we consider each of the three models andderive the correspondingGLRTs.ThelatterareevaluatedinSection3on sim-ulateddatabutwherethetarget spectralsignature,themeanand covariancematrixofthebackgroundareobtainedfromreal hyper-spectralimages.

(3)

2. GLRTformatrixvariatet-distributedbackground

Asstatedbefore,letusassumethatwewishtodecidewhether agivenvector ycontainsasignatureofinteresttinthe presence of disturbance z whose mean value

μ

and covariance matrix



areunknown,andletusassume thatasetoftrainingsampleszi,

i=1,...,n are available which share the same distribution as z. ThesesamplescanbecollectedaroundthePUToralongthewhole image. We simply assume herethat n> p. Therefore, we would liketosolvethefollowingproblem:

H0:y=z; zi d

=z,i=1,...,n

H1:y=

α

t+

β

z; zi=d z,i=1,...,n (1) where =d means “has the same distribution as”. In (1), t corre-spondstotheassumedspectralsignatureofthetarget and

α

de-notesits unknown amplitude. When

β

=1 one obtains the con-ventionaladditivemodel.When

β

=1−

α

thereplacementmodel isrecovered,andthemixedmodelcorrespondstoanarbitrary

β

.

In orderto derive theGLRT,we need to specifythejoint dis-tributionofyandZwhereZ=



z1 z2 ... zn



.Assaidinthe introduction,we assume that



y Z



follows a matrix-variate t -distributionwith

ν

degreesoffreedom so that we need tosolve thefollowingcompositehypothesistestingproblem:

H0:



y Z



=d Tp,n+1

(

ν

,M0,

(

ν

− 2)



,In+1

)

H1:



y Z



=d Tp,n+1



ν

,M1,

(

ν

− 2

)



,



β

2 0T 0 In



(2) where M0=



μ μ

1T n



, M1=



α

t+

βμ μ

1T n



, 1n is a n × 1 vectorwithallelementsequaltoone,

μ

standsforthemeanvalue ofthe backgroundwhile



denotes its covariance matrix.In (2),

T

()

standsforthematrixvariatet-distribution[16,17]sothat the probabilitydensityfunction(p.d.f.)oftheobservationsundereach hypothesisisgivenby p0

(

y,Z

)

=C

|



|

n+1 2





I p+



−1

ν

− 2



y

μ

Z

μ

1T n



×



y

μ

Z

μ

1T n



T





ν+n+p 2 p1

(

y,Z

)

=C

β

−p

|



|

n+1 2





I p+



−1

ν

− 2



˜ y

μ

Z

μ

1T n



×



y˜−

μ

Z

μ

1T n



T





ν+n+p 2 (3) with y˜=

β

−1

(

y

α

t

)

andC= p((ν+n+p)/2) πp(n+1)/2p((ν+p−1)/2). It should be

observed that the columns of



y Z



are only uncorrelated but not independent, as p

(

y,z1,...,zn

)

cannot be factored as

p

(

y

)

ni=1p

(

zi

)

.

WenowderivetheGLRTfortheproblemin(2).Letusstartby consideringthe following function f(



) where S issome positive definitematrix: f

(

)

=

|



|

n+1 2



I p+

(

ν

− 2)−1



−1S



ν+n+p 2 =

|



|

ν+p−1 2





+

(

ν

− 2)−1S



ν+n+p 2 (4)

Differentiationoflogf(



)yields

logf

(

)



=

ν

+ p− 1 2



−1

ν

+n+p 2

(



+

(

ν

− 2

)

−1S

)

−1. (5)

Settingthisderivativeoftozero,wecanseethatf(



)achievesits maximumat



∗=

(

ν

(

ν

− 2+p

)(

− 1)n+1S

)

=

γ

S (6) Itfollowsthat max  p0

(

y,Z

)

=C 







y

μ

Z

μ

1T n



(

y

μ

)

T

(

Z

μ

1T n

)

T





n+1 2 max  p1

(

y,Z

)

=C 

β

−p







y˜

μ

Z

μ

1T n



(

y˜−

μ

)

T

(

Z

μ

1T n

)

T





n+1 2 (7)

withC=C

γ

−p(n+1)/2[1+

(

ν

− 2

)

−1

γ

−1]−p(ν+n+p)/2.SinceCisthe

sameunderH0andH1 itwillcanceloutintheGLRandtherefore theGLRdoesnotdependon

ν

.Thismeansthat

ν

doesnotneedto beknownsinceitsestimationisnotactuallyrequiredtoobtainthe GLR.

Now,foranyvectorx,

M

(

μ

)

=



x

μ

Z

μ

1T n



x

μ

Z

μ

1T n



T =

(

x

μ

)(

x

μ

)

T+

(

Z

μ

1T n

)(

Z

μ

1Tn

)

T =xxT

μ

xT− x

μ

T+

μμ

T +ZZT

μ

1T nZT− Z1n

μ

T+n

μμ

T =

(

n+1

)

μμ

T

μ

(

x+Z1 n

)

T

(

x+Z1n

)

μ

T +xxT+ZZT =

(

n+1

)

μ

x+Z1n n+1

μ

x+Z1n n+1

T +xxT+ZZT

(

x+Z1n

)(

x+Z1n

)

T n+1 =

(

n+1

)

μ

x+Z1n n+1

μ

x+Z1n n+1

T +



x Z



In+1− 1n+11Tn+1 n+1



x Z



T (8) Consequently, min μ

|

M

(

μ

)

|

=







x Z



Pn+1



x Z



T





(9)

with Pn+1 the orthogonal projector on the null space of 1n+1.

Hence,wearriveat max μ, p0

(

y,Z

)

=C 







y Z



Pn+1



y Z



T





n+1 2 max μ, p1

(

y,Z

)

=C

β

−p







y˜ Z



Pn+1



˜ y Z



T





n+1 2 (10)

Next,foranyvectorxandmatrixQ(notnecessarilyPn+1),



x Z



Q



x Z



T=



x Z

Q

11 Q12 Q21 Q22



x Z



T =Q11xxT+ZQ21xT+xQ12ZT+ZQ22ZT =Q11



x+Q11−1ZQ21



x+Q11−1ZQ21



T +ZQ2.1ZT (11) whereQ2.1=Q22− Q21Q11−1Q12.Therefore,







x Z



Q



x Z



T





=



ZQ2.1ZT



×

1+Q11

x+Q11−1ZQ21

T

ZQ2.1ZT

−1

x+Q11−1ZQ21

(12)

Comingback tothecaseQ=Pn+1=In+1−

(

n+1

)

−11n+11Tn+1,we

have Q=



1 0T 0 In



(

n+1

)

−1



1 1T n 1n 1n1Tn



(13) sothat

(4)

Q11=1−

(

n+1

)

−1=n

(

n+1

)

−1 Q21=−(n+1

)

−11n

Q22=In

(

n+1

)

−11n1Tn

Q2.1=In− n−11n1Tn=Pn (14) It follows that Q11−1ZQ21=−n−1Z1n=−¯z and ZQ2.1ZT=ZPnZT= ZZT− n¯z¯zT=S.Hence,theGLRisgivenby

GLR=



1+Q11

(

y− ¯z

)

TS−1

(

y− ¯z

)



(n+1)/2 minα,β

β

p[1+Q 11

(

y˜− ¯z

)

TS−1

(

y˜− ¯z

)

](n+1)/2 = [1+ n n+1

(

y− ¯z

)

TS−1

(

y− ¯z

)

](n+1)/2 minα,β

β

p[1+ n n+1

(

yβαt− ¯z

)

TS−1

(

yβαt− ¯z

)

](n+1)/2 (15)

A few important observationscan be maderegarding this re-sult. First,forallthreemodels,theGLRs in(15)coincide withtheir Gaussian counterparts. Forthe additive modela proof is givenin AppendixA.Amoreintuitive waytofigureoutthisequivalenceis torealizethat theexpressionoftheGLRin(15)doesnotdepend on

ν

andthat, letting

ν

grow to infinity, one should recoverthe GLRforGaussiandistributeddata.Asforthereplacementandthe mixedmodels,theexpressionin(15)isexactlythatoftheACUTE andSPADEdetectorsof[10]and[11]respectively,wheretheGLRTs forthereplacementmodelandthemixedmodelarederivedunder theGaussianassumption.Therefore,thelatterarestillGLRTsfora muchbroaderclassofdistributionsthaninitiallyexpected.

LetusalsobrieflycommentontheimplementationoftheGLRT. For the additive model

β

=1, and the minimization problemin (15) isa simple linearleast-squares problem forwhicha closed-formsolutioncanbeobtained.Thisyields

GLR2AM/(n+1)= 1+ n n+1

(

y− ¯z

)

TS−1

(

y− ¯z

)

minα1+ n n+1

(

y− ¯z

α

t

)

TS−1

(

y− ¯z

α

t

)

= 1+ n n+1

(

y− ¯z

)

TS−1

(

y− ¯z

)

1+ n n+1

(

y− ¯z

)

TS−1

(

y− ¯z

)

n+1n [(y−¯z) TS−1t]2 tTS−1tn n+1[

(

y− ¯z

)

TS−1t]2 [1+ n n+1

(

y− ¯z

)

TS−1

(

y− ¯z

)

][tTS−1t] (16)

The GLRin(16)generalizesKelly’s detectortothe caseof a non-centered Student distributed background. Note that (16) differs fromKelly’sdetectorbythe n+1n factor.

Asforthereplacementmodel,

β

=1

α

andtheminimization shouldbeconductedwithrespectto

α

only,i.e.,

GLRRM=

1+ n n+1



S−1/2

(

y− ¯z

)



2

(n+1)/2 minα

(

1−

α

)

p

1+ n n+1 S−1/2(yαt(1α)¯z)2 (1−α)2

(n+1)/2 (17)

Asshownin[10],thissimplyamountstofindingthe(unique) pos-itiverootofa2nd-orderpolynomial.Finally,forthemixedmodel where

β

isarbitrary,one hasa2-D minimization problem. How-ever,minimizationover

α

canbedoneanalytically, leavingonlya minimizationover

β

: GLRMM=

1+ n n+1



S−1/2

(

y− ¯z

)



2

(n+1)/2 minβ

β

p



1+ n n+1  PS−1/2tS −1/2(yβ¯z) 2 β2



(n+1)/2 (18)

Again the solution is obtained as the unique positive root of a second-order polynomialequation [11].Thereforeforboththe re-placement modeland themixedmodel, all unknown parameters canbeobtainedinclosed-form.

3. Numericalsimulations

In the present section, we will compare the detectors devel-oped above.The additive model GLRAM in(16)will be referred to

as mKELLY in the sequel as it consists in a slight modification of Kelly’s original GLRT [18]. The replacement model GLRRM in

10-4 10-3 10-2 10-1 100

Probability of false alarm

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Probabi li ty of detecti o n

Student background (ν =5), replacement model, α = 0.05

GLRAM(mKELLY)

GLRRM(ACUTE)

GLRMM(SPADE)

10-4 10-3 10-2 10-1 100

Probability of false alarm

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 P ro b a b ilit y o f d etectio n

Student background (ν =20), replacement model, α = 0.05 GLRAM(mKELLY)

GLRRM(ACUTE)

GLRMM(SPADE)

10-4 10-3 10-2 10-1 100

Probability of false alarm

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Probabi li ty of detecti o n

Student background (ν =10000), replacement model, α = 0.05 GLRAM(mKELLY)

GLRRM(ACUTE)

GLRMM(SPADE)

(5)

0.95 0.96 0.97 0.98 0.99 1

Value of

β

-15 -10 -5 0 5 10 15 20

P

fa

gain

w

ith

resp

ect

to

m

K

E

L

L

Y

(decib

els)

Student background (

ν =20), α = 0.05

GLRRM(ACUTE) GLRMM(SPADE)

Fig. 2. P fa gain versus βfor P d = 0 . 5 .

(17)and the modified model GLRMM in (18) will be referred as

toACUTE [10]and SPADE[11]. Theselast two detectorshave al-readybeenassessedagainstrealdatadrawnfromtheRIT[19]and Viareggio[20]experiments.Herein,weevaluatetheirperformance aswell as their robustness on simulated yetrealistic data. More preciselyweconsideranimageacquiredinViareggio(Italy)inMay 2013froman aircraftflyingat1200meters.Theoriginal dataisa [450× 375]pixels map intheVisible NearInfraRed(VINR)band

(

400− 1000nm

)

withaspatialresolutionofabout0.6meters.The sceneiscomposedofparkinglots,roads,buildings,sportfieldsand pinewoods. Different kindsof vehicles aswell ascolored panels servedasknowntargets.Foreach ofthesetargets,a spectral sig-natureobtained fromgroundspectroradiometermeasurements is available.Herein,weconsiderttobethe signatureoftheV5 tar-get,

μ

and



arerespectivelythesamplemeanandsample covari-ancematrixobtainedfromthewholeViareggioimage.Ithastobe noticedthattheserawradiancedatahavefirstbeenconvertedto reflectancemeasurementsusinganELMmethod[21,22].The num-berof spectral bands used is p=32 andthe number oftraining samplesisn=60.Thebackgroundissimulatedusingat distribu-tionwith

ν

degreesoffreedom.

Fig.1plotstheReceiverOperationCurve(ROC)obtainedforthe replacementmodel[

β

=1−

α

]with

α

=0.05,fordifferentvalues of

ν

ranging from

ν

=5(heavy-taileddistribution) to

ν

=10,000 (nearlyGaussiandistributedbackground).Ascouldbe anticipated, ACUTEexhibits thebest performance since it corresponds tothe GLRT for the specific case

β

=1−

α

. However, SPADE is shown to incur a small degradation compared to this optimal detector. Onthe contrary, mKELLY exhibitsa significant performance loss, mostlybecause E

{

z

}

=E

{

zk

}

, a fact that is not accounted for in theadditivemodel,contrarytotheothertwodetectors.

WenowassesstherobustnessofACUTEandSPADE.More pre-cisely, we study their performance when

β

varies between

β

= 1−

α

(replacement model) and

β

=1(additive model). InFig. 2, wedisplay the probabilityof falsealarm (Pfa) gain ofACUTEand SPADEwithrespect tomKELLY, i.e., 10log10

Pf a(GLRAM)

Pf a(GLRRM/MM).This Pfa gain allows tomeasure the improvement(if positive)or theloss (whennegative)ofbothACUTEandSPADEwithrespecttomKELLY.

Inthisfigure

α

isfixedto0.05andtheprobabilityofdetectionis

Pd=0.5. Fig.2confirmsthatACUTE isslightlybetterthan SPADE whentheactualvalueof

β

iscloseto1−

α

.However,assoonas

β

departsfrom1−

α

SPADEshowsbetterperformancethanACUTE. Moreover,SPADEperformsbetterthanmKELLYformostvaluesof

β

unless

β

comesverycloseto1,butinthiscasethelossofSPADE withrespecttomKELLYismarginal.Therefore,SPADEprovidesthe bestrobustness,withsmallperformancelosscomparedtothe op-timalsolutionwhateverthevalueof

β

.

4. Concludingremarks

Inthis communication, we considered thedetectionof a sub-pixel targetin hyperspectral imagingwhenthe backgroundis no longer Gaussian butt-distributed.The GLRTs fora generalmixed model,including the standardadditive caseandthe replacement one, were derived, generalizingthe Gaussian versions previously derived.Forthethreespecific valuesof

β

consideredinthe liter-ature, it wasshown that the GLRTs remain the same andhence thedetectorsinitiallyproposed underaGaussian frameworkhave moregenerality thanexpected. Moreover, theydo notdepend on the unknown degree of freedom ofthe t-distribution. Numerical simulations showedthatSPADEprovides averygood trade-off as itisalwaysclosetoorbetterthanKellyandACUTEwhichare op-timalonlyforspecificvaluesof

β

.

DeclarationofCompetingInterest

The authors acknowledge that there is no conflict of interest regardingthepaperentitled“Sub-pixel detectioninhyperspectral imagingwithellipticallycontouredt-distributedback-ground”. AppendixA. GLRfortheadditivemodelandGaussian distributedbackground

Inthis appendix,we derive the GLRTforGaussian distributed backgroundandfortheadditivemodel.Wethusconsiderthe fol-lowingdetectionproblem

H0:



y Z



=d Np,n+1



μ μ

1Tn



(6)

H1:



y Z



=d Np,n+1



α

t+

μ μ

1Tn



,



In+1

(A.1)

Thep.d.f.of(y,Z)isinthiscase

p0

(

y,Z

)

=

(

2

π

)

(n+1)p 2

|



|

n+1 2 × etr



−1 2



−1



y

μ

Z

μ

1T n



(

y

μ

)

T

(

Z

μ

1T n

)

T



p1

(

y,Z

)

=

(

2

π

)

(n+1)p 2

|



|

n+1 2 × etr



−1 2



−1



y

μ

Z

μ

1T n



(

y

μ

)

T

(

Z

μ

1T n

)

T



(A.2)

It iswell-knownthat

|



|

n+12 etr

{

−1

2



−1S

}

achievesitsmaximum

at



=

(

n+1

)

−1S,andhence max 

|



|

n+1 2 etr



−1 2



−1S



=



e n+1

−

n+1 2

|

S

|

n+12 (A.3)

Itfollowsthatmaxp0(y,Z)andmaxp1(y,Z)areproportionalto

(7) whichholds for Studentdistributions. From there, everything followsandtheGLRsforStudentorGaussiandistributionsarethe sameandaregivenby(15).

CRediTauthorshipcontributionstatement

Olivier Besson: Conceptualization, Methodology, Validation, Writing -original draft, Writing -review & editing,Visualization. FrançoisVincent:Conceptualization,Methodology,Software, Writ-ing-originaldraft,Writing-review&editing.

References

[1] M.T. Eismann , Hyperspectral remote sensing, SPIE, 2012 .

[2] D.G. Manolakis , R.B. Lockwood , T.W. Cooley , Hyperspectral Imaging Remote Sensing, Cambridge University Press, 2016 .

[3] D. Manolakis , G. Shaw , Detection algorithms for hyperspectral imaging appli- cations, IEEE Signal Process. Mag. 19 (1) (2002) 29–43 .

[4] D. Manolakis , E. Truslow , M. Pieper , T. Cooley , M. Brueggeman , Detection algo- rithms in hyperspectral imaging systems: an overview of practical algorithms, IEEE Signal Process. Mag. 31 (1) (2014) 24–33 .

[5] N.M. Nasrabadi , Hyperspectral target detection : an overview of current and future challenges, IEEE Signal Process. Mag. 31 (1) (2014) 34–44 .

[6] D. Manolakis , G. Siracusa , G. Shaw , Hyperspectral subpixel detection using the linear mixing model, IEEE Trans. Geosci. Remote Sens. 39 (7) (2001) 1392–1409 .

[7] D. Manolakis , Hyperspectral signal models and implications to material detec- tion algorithms, in: Proceedings ICASSP, volume 3, 2004, pp. 117–120 . [8] R. DiPietro , D. Manolakis , R. Lockwood , T. Cooley , J. Jacobson , Performance

evaluation of hyperspectral detection algorithms for sub-pixel objects, in: Pro- ceedings of SPIE - The International Society for Optical Engineering, 2010 . [9] J. Frontera-Pons , F. Pascal , J.-P. Ovarlez , Adaptive nonzero-mean Gaussian de-

tection, IEEE Trans. Geosci. Remote Sens. 55 (2) (2017) 1117–1124 .

[10] F. Vincent, O. Besson, Generalized likelihood ratio test for subpixel target detection in hyperspectral imaging, IEEE Trans. Geosci. Remote Sens. 58 (6) (2020) 4 479–4 489, doi: 10.1109/TGRS.2020.2965212 .

[11] F. Vincent, O. Besson, Generalized likelihood ratio test for modified replace- ment model in hyperspectral imaging detection, Signal Process. 174 (2020) pa- per 107643, doi: 10.1016/j.sigpro.2020.107643 .

[12] D. Manolakis , D. Marden , G. Shaw , Hyperspectral image processing for auto- matic target detection applications, Lincoln Lab. J. 14 (1) (2003) 79–116 . [13] S. Matteoli , M. Diani , G. Corsini , A tutorial overview of anomaly detection in

hyperspectral images, IEEE Aerosp. Electron.Syst. Mag. 25 (7) (2010) 5–27 . [14] T.W. Anderson , K.-T. Fang , Theory and Applications of Elliptically Contoured

and Related Distributions, Technical Report, Department of Statistics, Stanford University, 1990 .

[15] F. Kai-Tai , Z. Yao-Ting , Generalized Multivariate Analysis, Springer Verlag, Berlin, 1990 .

[16] A.K. Gupta , D.K. Nagar , Matrix Variate Distributions, Chapman & Hall/CRC, Boca Raton, FL, 20 0 0 .

[17] S. Kotz , S. Nadarajah , Multivariate T Distributions and their Applications, 2004 . [18] E. Kelly , An adaptive detection algorithm, IEEE Trans. Aerosp.Electron. Syst. 22

(2) (1986) 115–127 .

[19] D. Snyder , J. Kerekes , I. Fairweather , R. Crabtree , J. Shive , S. Hager , Develop- ment of a web-based application to evaluate target finding algorithms, in: Pro- ceedings IGARSS 2008, volume 2, Boston, MA, 2008 .

[20] N. Acito , S. Matteoli , A. Rossi , M. Diani , G. Corsini , Hyperspectral airborne “Viareggio 2013 trial” data collection for detection algorithm assessment, IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 9 (6) (2016) 2356–2376 .

[21] G. Ferrier , Evaluation of apparent surface reflectance estimation methodolo- gies, Int. J. Remote Sens. 16 (1995) 2291–2297 .

[22] G.M. Smith , E.J. Milton , The use of the empirical line method to calibrate re- motely sensed data to reflectance, Int. J. Remote Sens. 20 (1999) 2653–2662 .

Figure

Fig.  1. ROC for the replacement model  β =  ( 1 − α )  .
Fig.  2. P  fa  gain versus  β for P  d  = 0 . 5 .

Références

Documents relatifs

• l a notion de durée chronologique (l'archevêque J. Remontons ensemble le temps afin de découvrir le terreau qui fertilisa l'esprit de Charles Darwin au milieu

Figure 7 Comparison between the proposed detector (c) and the state-of-the-art detector (b) for one GRD image with linear structures highlighted by red

Although non-Gaussian distributions have already been as- sumed for background modeling, the parameters estimation is still performed using classical Gaussian based estimators;.. as

Enfin, le null astrophysique et la phase rms semblent être des paramètres indépendants (en bas). Cette coupe correspond à 20 répétitions de la méthode dans le cas d’un histogramme

Crem azie Bélang er Du P arc Lege ndre Rosem ont Saint-J oseph Sa int- De nis La Petite Patrie Villeray Parc Extension Du Plateau Mont- Royal Nord de l'île Ahuntsic Saint-

Vous devez savoir comment les périphériques sont physiquement connectés pour pouvoir résoudre des problèmes au niveau de la couche physique, notamment les problèmes de câblage ou

Given that SBS is swollen by the saturates and aromatics fractions from bitumen [12] it is likely that the maltenes portion that blends with PB is rich in alkanes and

Il constitue une représentativité théorique forte et un potentiel de découverte réel (Hlady-Rispal, 2002).. Lorsque nous avons commencé nos investigations sur le terrain, le