GR

anexensuvemsimapmuazve

Raap

foth(Lre

ofclumrigde[3mthpe

verRe45CaanCOEu

SeUn(e-

VEala

T

RSL-00900-2

Abstract—Thind fast boats wxperience severncountered, inurveillance appessels is of greoving vessels inmulated resultpplying movinultichannel SA

zimuth domainessels.

Index Terms—adar (SBR), perture radar (

HE GROWcreated the

or moving-targis subject, Sp

LEO) offer thasonable operMedium and

f pixels in SAutter backgrou

may appear bugid-hulled infetected by SB]. In this pap

modeled to anais paper is to

erformance of

Manuscript receirsion published M

esearch Fellowsh90; by FI-AGA

atalunya, contractd Innovation

ONSOLIDER Curopean Commiss

A. Broquetas, Ensing Laboratory

niversitat Politècn-mail: broquetas@A. J. Beaton is w

EGA UK Ltd, Luasdhair.beaton@v

Mo

T

014

is paper analywith synthetic are defocusing nter alia, in plications whereat interest. Thn SAR imagests. A matched

ng-target indicAR images, ren and enhancin

—Moving-targradar detectio

(SAR), Matche

I. INTR

WING interest ineed to design

get indication pace-Borne Rahe advantage rating cost [1].

large vesselsAR images [2und. However

uried in clutteflatable boatsRs, except fo

per the kinemalyze the impao evaluate thef a processing

ived September 1March 22, 2015.ip Program, MinAUR Research t 2010FI EM051(MICINN) und

CSD2008-00068 sion under FP7-SPE. Makhoul,F. Cy, Department onica de Catalunya@tsc.upc.edu). with the satellite uton (Bedfordshi

vegaspace.com).

odeling

Francisco C

yses the probleaperture radarand/or smeari

ocean traffre the detectiohe target mod are presentedfilter bank (M

cation (MTI) efocusing the

ng the detectab

get indication on, ocean su

ed Filter Bank (

RODUCTION

in maritime tran new instrum(MTI) and de

adars (SBR) inof observing . typically app], which contr for small bor and noise.

s (RHIB), whor calm seas a

matics of smallact on SAR ime small and fachain includin

16, 2014; revised This work has

nisterio de EducaFellowship Pr

757; by the Spander projects Tand TIN2014-5

PACE Project SIMCeba and Y. Zhaof Signal Theorya-Barcelona Tech,

systems and appire) LU1 3LU, U

g Fast S

Ceba, Eduard

ems of imagingrs (SAR), whicing. This situafic monitoringon of small andeling and impd and discussedMFB) is used techniques ovtarget signal

bility of these ty

(MTI), Space-rveillance, syn(MFB).

affic monitorinments and technetection purpon Low Earth large areas w

pear as bright trast with the

oats the target This is the c

hich can hardand slow kineml and fast shimages. The g

fast vessel detng a MFB [9]

October 5, 2014. been supported bación, contract Arogram, Generalnish Ministry of TEC2011-28201-55413-C2-1-P; aMTISYS Ref. 263an are with the y and Communi, Barcelona 08034

plications BU, TelUnited Kingdom

Boat SAR M

do Makhoul, Yu Zhan

g small ch may ation is g and nd fast pact of d using before

ver the in the

ypes of

-Borne nthetic

ng has niques ses. In Orbits with a

group lower signal ase of dly be matics ips are goal of tection where

Current by FPU

AP2009-litat de Science C02-01, and by 3268. Remote

ications, 4, Spain

lespazio (e-mail:

eachbothmul

A.

ToRHIbeenscatcanodom(GTensuqua

To

outlleng(CAcano(inc

Duillumelemdisccalcallopuls

B.

Rfromvessthatwaychar

Tintoheavpitc

MotioMTI S

Member, IEEand Alasdha

Fig.

h filter coversh vessel veltichannel MT

Radar Cross-

o provide reprIB in this simn developed ttering canonionical shapes

minant scatteriTD) and physure solution sternion mathe

o model the Rline was consgth x 3m beamAD) model wonical eleme

cluding orientauring the rmination pulsments are projcretization. Thculated with towing the comse basis.

Vessel kinema

Realistic vessem trajectory vsel with the set allows a vypoints and a mracteristics an

The attitude ofo 6 elementaryve, surge and

ch and roll. A

on ImpSystem

EE, Antoni Bir J. Beaton.

1. RHIB Referen

s a range of celocity and TI-SAR techniq

II. TARGET

-Section

resentative radmulation, a 3D

based on a cal shapes. Th [4] is mappeng terms fromsical optics (Pstability over ematics has rep

RHIB RCS, astructed to mm). From this was created an

nts, their diation) from theadar observase, the positjected into rahe complex the parametric

mputation of th

atics

el kinematics velocity changea. This was m

vessel to altemodel of the i

nd the vessel dyf a vessel in y oscillatory sway together

As an initial e

pact onms

Broquetas, M

ce and CAD mod

chirp rates, taacceleration,

ques [7].

T MODELING

dar cross sectD radar reflemesh of pa

he backscattered by parame

m geometric thPO) solutionsall possible vplaced Euler T

a frame of theeet specified frame, a compnd used to imensions ande body frame cation time, tions of the adar grid usinscattering of

c models and he target raw

includes acceges and by themodeled in a er course andinteraction betynamical propa seaway canmotions, 3 trr with 3 rotatiestimate, thes

n Sate

Member, IEEE

del.

aking into acccombined

tion (RCS) foectivity modelrameterized rring response etric models uheory of diffracs. Additionall

viewing directTransforms.

e vessel watedimensions (

puter aided deidentify 15 Rd offset poscentroid (Fig. for every rRHIB scatte

ng the SAR imf each elemen

coherently adata in a puls

elerations resue interaction o

simulation end speed betwtween the locaperties. n be broken dranslation motonal motions ye motions ca

ellite

E,

1

count with

r the l has radar of 6

using ction ly to ions,

erline (10m esign

RHIB sition 1).

radar ering mage nt is dded se to

ulting f the

ngine ween al sea

down tions yaw,

an be

GRSL-00900-2014 2

treated as independent modes and then coupled using the forcing provided by the incoming wave and its direction. The roll and pitch periods for a vessel can be estimated using empirical equations that base upon the vessel dimensions. Then, using wind vector parameters passed to the simulation [5], a significant wave height and period can be determined using a sea spectrum formulation [6].

With the sea and vessel dynamics characterized, a novel 6-degree of freedom kinematic propagator facilitates solution of the kinematic state of the RHIB centre of gravity (CoG) for a set of discrete time steps during the SAR image acquisition. This propagator resolves the RHIB-wave encounter frequency, the amplitudes of the 6 elementary motions and then scales these amplitudes to the RHIB using a Lorentzian function based on the modal natural frequencies. The wave elevation discrepancy between hull and CoG is solved using a 2D intersection model utilising the waterline outline and injects an additional phase offset into the final motion.

The resultant amplitude of the CoG motion can be solved as function of the maximum amplitude, the encounter frequency (ωe) and the wave elevation phase offset (ε), for a set of time steps corresponding to the pulse repetition frequency of the SBR during the acquisition

( ) = cos ( ) + . (1) A geometry engine provided by Satellite Tool Kit (STK)

allows solution of the scattering elements motion relative to the CoG using the element position parameters. Higher order kinematic terms (velocity and acceleration) are generated utilising a 3rd order Hermite polynomial interpolation scheme. Once the kinematics are generated, the motion of each scattering element is projected into the SBR coordinates.

III. SIGNAL MODEL

For a monostatic SAR in X-Band working in stripmap mode, the signal captured by the antenna can be expressed using the vector notation [13-14]

= ( ) + ( ) + , ∈ (2) as the addition of target signal ( ), sea clutter ( ) and noise ; where and are the moving target and clutter parameter vectors respectively (range, velocities and accelerations) and M is the number of channels.

A. Moving Target

Let us first consider a point target with uniform accelerated rectilinear motion in the slant range-azimuth plane. Assuming a flat Earth geometry with the platform moving at with respect to ground, the range history equation during the observation time can be written as

( )

( )

( )

2 22 2

0

2

20

0

2 2

1,

2

sr azsr p az

p az

sr sr

R

a aR v v v

v vR v a

R

η

η η η η

η η

=

+ + + − + ≈

− + + +

(3)

being η the azimuth time, R0 the slant range at η=0, v the target velocity, a the target acceleration, and the subscripts az and sr refer to azimuth and slant range components, respectively. The right-hand part of equation (3) corresponds to a second order Taylor series expansion around η=0. The acquired low-pass equivalent range compressed signal in time domain with a zero-squint angle acquisition can be expressed as follows:

( ) ( ) [ ]

( )

00

0 00

0 0

2

20

0 0

2,

4 4exp exp

2exp ,

RC r a c

sr

p az

sr

RS A p w

c

f f vj R j

c c

v vfj a

c R

ητ η τ η η

π π η

π η

≈ ⋅ − ⋅ −

⋅ − ⋅ − − ⋅ − +

(4)

with τ representing the range time (or fast time), pr the range compressed pulse envelope (i.e: a sinc-like function) and wa the antenna weighting function (i.e: squared azimuth pattern). The second exponential term in (4) represents a shift in the Doppler spectrum of the target due to the across-track velocity component [7], which results into an azimuth displacement in the SAR image. In the case of high radial velocities the target spectrum can either fall out of the Doppler processing band or re-enter in the back-folded spectrum as an ambiguity. Furthermore they produce a high range walk not compensated by the azimuthal filters which, for high speed values, can impact in the slant range resolution. The third exponential term in (4) represents the Doppler frequency rate (ka), exploited for the azimuth compression [8]. As can be seen, the Doppler rate is mainly affected by vaz and asr [7,9], producing a mismatch with the stationary world matched filter (SWMF). Therefore, the Doppler rate error can be defined as the Doppler frequency rate difference between the moving and a static target resulting in azimuth defocusing [7]:

2

00

22,az p az

a sr

v v vk k k a

Rε λ −

= − = +

(5)

B. Clutter and noise

The undesirable captured components are the background sea scattering ( ) and the thermal noise . In the simulation the sea clutter is modeled as a zero mean complex Gaussian process, c(ϑc)∈C~N(0, σ , ), where represents the correlation matrix between channels. The clutter power has been obtained using the model presented in [10], which provides the normalized RCS σ as a function of the frequency, polarization, incidence angle and sea state. Assuming this value equal for all the channels the clutter power σ is then proportional to the area of the resolution

GR

cedi

ecdeefm

chaswa GanmEqse

traslirepamcousdecoin

[1jois (S

Iusto anDocobe

Figtarg

RSL-00900-2

ell. As in threction correcOwing to th

choes backscecorrelate betwffect has been

matrix structure

hannels - andssumed, whicavelength, theThe noise h

aussian proc∈C~N(0, σmatrix. The no

quivalent Sigmensitivity [15].

SBR systemsack direction ightly differengistered, the c

arameters canmission has beeonfiguration wsing external etection capabompared to thstrument has bThe Extende4], has been intly with a mperformed to

SCNR) at the pIn order to imsed prior to M establish a cr

nd their tuningoppler rates

ompression losetween the rec

g. 3. Scenario lagets are moving a

014

he TSC backction term has he internal clcattered fromween the spatimodelled inc

e, giving a cor

d . A Gaussih depends o

e resolution anhas been modcess uncorre), wheroise power σma Zero (NE

IV. MTI

s with multiplprovide imagnt positions. clutter can be cn be estimateen presented with non-unif

deployable abilities for slohe current statbeen used throed DPCA (Eimplemented.

matched beammaximize the

processor’s oumprove the deTI processing riterion to defg. The proposconsists in msses as a funcceived signal a

ayover depicting at 37.8 knots.

kscattering mbeen includedlutter motion m the sameially separated

corporating a trrelation coeff

ian correlationon the baselinnd the wind spdeled as a zlated betweere repreσ is determESZ), a meas

I PROCESSING

le channels ales of the samOnce the imacancelled and d. A new opin [13], basedformly displaantennas. It owly and smte-of-the-art Soughout the simDPCA) algor An adaptive

m-former (to the signal to clu

utput. etection perfor [7, 14]. There

fine the numbeed method to

measuring emction of the quand the station

the targets locat

odel [11], a d.

(ICM) the e spatial pod receivers, antemporal covaficient , be

n function hasne time, the eed [12]. zero-mean coen channels,esents the id

mined by the surement of s

ligned in the me scene acquiages have beethe targets dy

ptimized SARd on a multichaced phase ce

provides impmall moving vSAR missionsmulations. rithm, describclutter cance

he target of inutter-plus-nois

rmance, the Mefore, it is necer of required calculate the

mpirically the uadratic phasenary world ma

tion and direction

wind

clutter osition nd this ariance etween

s been radar

omplex , i.e., dentity Noise

system

along-ired at en co-

ynamic R-MTI hannel enters, proved vessels s. This

bed in ellation nterest) e ratio

MFB is cessary

filters filters pulse

e error atched

filtemetthe one

Thrad.Figufuncacce

Tof (formtargThefrommulfrompurpfromcritetypirangrangnumbothphenthe

Fevenrangposithis wasdiffdivinumimp

Figlin

ns. All

er (SWMF) at thod we can pronly knowledof those filter

he points of i. which proviure 2 shows ction of the seleration of 0.

The necessary 5). For this, thmula (i.e: ϕ=πget observatione resulting kεm zero to -3dltiplying the pm the center. Iposes, three fm -30 to 30erion and conical SBR numge from -22.5 ge acceleration

mber of filtersh kinematic nomenon (depacross-track a

From the resuln forcing thege accelerationitive and 2) nvalue, jointly

s used in (5)ference with reided by the f

mber of 21 plemented in th

g. 2. Losses in ae) as a function o

the edges of redict the lossdge of the obrs to a certain interest are ϕε ide losses of the losses an

slant range ac22 m/s2 produnumber of fil

he ϕε -3dB point πkη2), replacinn time (namely-3dB value is

dB point. Theprevious valueIn [9] it is mefilters are nee0 m/s. Nonetnsidering only

mbers in X-Bato 22.5 m/s isns margin froms increases to

parameters pending on thacceleration halting modellede seaworthinesn within two never exceedsy with the obs) to find the espect to the Sfilter chirp ra

filters, whiche MFB for th

amplitude (solid lof slant range acce

the captured wses for a certaibservation timDoppler rate i-3dB = 2.72 ra-3 dB and -6

nd broadeningcceleration. Auces a loss of 3ters can be cais substituted

ng the time η by the edge of the chirp rateen the -3dB e by 2 for spaentioned that feded to covertheless, applyy the azimuthand, with thres covered. If adm -2 to 2 m/s2

seventeen. Thcan increase

he signs of vaz as a stronger imd kinematics iss of all the consecutive s

s the 9 m/s2 (Fervation time

relative maxSWMF. The reate bandwidth ch is the nhe simulation r

line) and azimutheleration asr.

window. Within phase error

me and adapt interval. ad. and ϕε -6dB

6 dB respectivg in azimuth s can be seen3 dB. alculated by md in the chirp pby one half othe time wind

e distance in width is obta

anning either sfor target detecr a velocity rying the prevhal velocities ee filters onlydditionally, a 2 is consideredhe combinatioe the defocuand asr), altho

mpact. it is observed vessels, the pikes 1) is alwFig. 5). Thereshown in Tab

ximum chirp esulting value

to obtain anumber of firesults.

hal broadening (d

3

h this with each

B = 5 vely. as a

n, an

means phase f the

dow). Hz/s

ained sides ction range vious

and y the slant

d, the on of using ough

that, slant ways efore ble I,

rate e was

total filters

dashed

GR

deat inimdimucluthwiwico

shthbukinallRCstrthacreloexthT2a T3wim/

FPEC

FigcooSA

RSL-00900-2

The structureepicted in Fig.

SAR image verse SWMF

mpact of limitefferent chirp

multichannel imutter or noise)e above (mulith three linesith multiple

ombined to forThe simulatio

hown in Tableem moving a

ut with differenematics for el the RHIBs hCS. As can berong impacts e slant range

cceleration peasults into a sess of -10 dB w

xcluding thesee simulations 2 is moving wfilter tuned to3 with vaz=-25ith a filter tun

m/s and asr=7.5

Fig. 4. SimulatoProcessing and MEDPCA processoCFAR image and

g. 5. Acceleratordinate. The sha

AR is acquiring da

014

V. SIMULA

e of the SAR. 4. Despite olevel, their r SAR procesed SAR resolrates in the

mages for a ) while the whltichannel ands are image slayers (the

rm the EDPCAons have beene I. The scenaat 37.8 knots ent directions each vessel. Inhave been mo

e seen in Fig. 5against the waacceleration

aks with a Meverely defocuwas obtained)e strong accele

for all boats (with a vaz=8.17o 5343.8 Hz/s5 m/s and asr=ned to 5343.8 5 m/s2 and it is

or breakdown sMTI stages. The or, which for eaca report of the PD

tions profiles foradowed area repreata.

ATION RESULT

R processor af generating traw data havesing in order lution and azi

MFB. Blacksingle layer

hite arrows ared multilayer). streams of mumultichannel

A image). n carried out wario contains tin the slant r(see Fig. 3), r

n order to makodelled with t5 the hull of thater which proprofile. The

MFB using conused image (w; thus a SAR eration peaks h(shadowed are7 m/s and asr=s is expected t=0.87 m/s2, is Hz/s. Target Ts foreseen to b

tructure, showinoutput of the az

ch filter generateD for each target.

r the three vessesents the time wi

TS

and MTI blocthe clutter ande been obtainr to account fmuth focusing

k arrows repr(i.e: target s

e the combinatThe arrows s

ultiple filter odomain has

with the paramthree vessels,

range-azimuthresulting in difke a fair compathe same maxhe boats experoduces the spiprocessing of

nstant Dopplerwith the best fobservation inhas been selecea of Fig. 5).

=1.58 m/s2, sucto refocus it. Tbest refocuse

T1 moves at vbe properly fo

ng separately thzimuthal MFB fees a single MTI

els in the slant indow within whi

ck are d noise ned by for the g with resents signal, tion of striped outputs been

meters all of

h plane fferent arison, ximum riences ikes in f these r rates filter a nterval cted in Target ch that Target

ed also vaz= 35 ocused

by afiltefilteThis5321.7 filte9.2 attenradinomrangrangprodthe

Fconsuffand (RChas targprovstati

Wwerof tarchprocprobFiguthe EDPthe adapveloT3.

Sym

R0

Vp

λ PRFBWBWTobs

vw

θw

H1/

Tw

NEσ0

τc

Pol

RC

a Rb F

he SAReeds theI image,

rangeich the

a MFB set to ers may provider, owing to ts is the case o

21.3 Hz/s and dB respectiv

ers. CompareddB for vess

ntion has beenial velocities

minal range cge defocusingge velocity caducing its brespectral doma

For a duly asssidered in ofered. Through

d without theCMC) have be

been appliedget T2 and T3 vided a gain oionary RCMC

With the SWMre at -10 dBsmthem with thhitecture, obtacessing thatbability of faure 6 shows thoutput of a coPCA responsecorrespondin

pted target procity, has redu

S

mbol

Slant ranPlatform Radar wa

F Pulse repWr Range prWa Azimuth

s ObservatWind velWind dir

/3 SignificaWave en

ESZ Noise EqClutter siClutter co

l Instrume

CS Maximum

Relative to radar lFor all the targets

5703.4 Hz/s. de better respothe fact that k

of targets T1 an5680.9 Hz/s g

vely, compared with the SWMel T3 and 13n paid to targproduce high

cell migrationg [7]. Moreovauses a back-foeakup into twoain. essment high

order to corrhout the simule tailored raneen compared

d since the difadapting the R

of 4 dB and 9.1C reconstructioMF, the maxi

m. A minimumhe best matchaining a probab

exceeds the alse alarm (Pf

he EDPCA imonstant false ae is velocity d

ng processor’srocessing, i.euced SCNR c

TABCENARIO AND RA

Parameter

nge to the target atvelocity

avelength petition frequencyrocessing bandwidprocessing bandw

tion time locity rection a

ant wave height ncountered period quivalent Sigma Zigma-zero orrelation time

ent polarization

m RCS b

look direction. s.

In practice oonse than the kinematics arend T3, for whgive an incremed with the eMF the improv3.5 dB for v

gets T2 and T3h mismatch wn curvature, ver for target olding in the Do halves locat

across-track rect the ranglations, the resnge cell mig

d. For target Tfferences can RCMC to each1 dB respectivon. imum amplitu

m gain of 8 dBhed filter usibility of detec

90% in allPfa) of 10-5 hamages and the alarm rate (CFdependent [13s outputs is d., T1, with thconditions com

BLE I ADAR PARAMETER

t η=0

y dth width

Zero

ne of the adjaexpected mate time depend

hich filters tunement of 2.3 dBexpected matvement factor

vessel T1. Sp3, since their

with respect toresulting in T2 the high Doppler spectted at the edge

velocities shage cell migrasults obtained gration correcT1 a static RC

be neglected.h filter of the bvely referred to

udes of the b is obtained foing the presection after the l three casesas been assumdetection map

FAR) detector.3] and that is different for he smallest rmpared to T2

RS

Value

6·105 m 7000 m/s0.031 m 4183 Hz

62.7 MHz 2.6 KHz 0.49 s

8.25 m/s -90º

1.802 m 2.66 s -23 dB

-17.5 dB 11 ms HH

2 dBsm

4

acent tched dent. ed to

B and tched rs are pecial

high o the slant slant trum, es of

all be ation with ction CMC . For bank o the

boats or all ented MTI s. A med. ps at . The why each adial

2 and

GRSL-00900-2014 5

VI. CONCLUSIONS

In this paper the radar scattering and kinematics for a small and fast boat sailing with a strong breeze has been modelled, considering realistic accelerations. To increase the SCNR and SAR detection performance a MFB has been included before the MTI processing stage. This architecture combined with the proposed SAR-GMTI mission [13] has been shown to be able to detect a RHIB experiencing moderate accelerations with three different heading directions.

When the high acceleration spike in slant range is encountered within the radar observation time, the proposed processing chain is not able to detect the vessels. For the simulated scenario, the three analyzed vessels experienced different peak acceleration periods due to differences in their kinematics and headings with respect to wind direction. Assuming a uniformly distributed random location of the time observation window with respect to each vessel acceleration profile in Fig. 5 the probability of observing the target free of acceleration peaks can be obtained. Averaging the 3 considered cases the probability of observing the fast boats in the simulated sea state without the acceleration peak is estimated to be 43%. To overcome this limitation, higher order terms could be considered in the range history in equation (3), to model and compensate the high target dynamics. However this would result into a higher number of degrees of freedom and larger number of filters in the MFB.

REFERENCES [1] Skolnik M.,"Spaceborne radar for the global surveillance of ships over

the ocean”, Proceedings of the 1997 National Radar Conference, pp.120 -125, 1997.

[2] Jackson, Christopher R., et al. “Synthetic aperture radar marine user's

manual” US Department of Commerce, 2004. [3] Blake, T. M. “The detection and tracking of small fast boats using HF

surface wave radar” HF Radio Systems and Techniques, IEE Conference Publication, pp. 219-224, 2000.

[4] Jackson, Julie A. “Three-Dimensional Feature Models for Synthetic Aperture Radar and Experiments in Feature Extraction” Electronic Thesis. Ohio State University, 2009.

[5] Doerry, A. W. “Ship Dynamics for Maritime ISAR Imaging” Sandia National Laboratories, U.S. 2008

[6] Walter H. Michel, “Sea spectra simplified, “Marine Technology, 1968, Vol. 5, No. 1, pp. 17-30.

[7] Cristallini, D., Pastina, D., Colone, F., Lombardo, P. “Efficient detection and imaging of moving targets in SAR images based on chirp scaling,”, IEEE Transactions on Geoscience and Remote Sensing, 2013, vol. 51, no 4, pp. 2403-2416

[8] Ian G. Cumming, Frank H. Wong, “Digital Processing of Synthetic Aperture Radar Data: Algorithms and Implementations”, Artech House, Norwood (MA), Jan. 2005.

[9] Charles E. Livingstone, Alan A. Thompson,“The moving object detection experiment on RADARSAT-2,”Can. J, Remote Sensing,vol. 30, N o . 3 , pp.355-368, 2004.

[10] V. Gregers-Hansen, R. Mital,“An improved empirical model for radar sea clutter reflectivity, “IEEE Trans. Aerosp. Elect. Sys.,vol. 48, N o . 4 , pp.3512-3524, 2012.

[11] I. Antipov,“Simulation of Sea Clutter Returns,” DSTO Electronic and Surveillance Research Laboratory,Salisbury,DSTO-TR-0679, Jun. 1998.

[12] S. Frasier, A. Camps,“Dual-Beam Interferometry for Ocean Surface Current Vector Mapping, “IEEE Trans. Aerosp. Elect.Sys.,vol. 39, N o . 2 , pp.401-414, Feb. 2001.

[13] E. Makhoul, A. Broquetas, J. Ruiz-Rodon, Y. Zhan, and F. Ceba, “A Performance Evaluation of SAR-GMTI Missions for Maritime Applications,” IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 5, pp. 2496–2509, May 2015.

[14] D. Cerutti-Maori, I. Sikaneta, “A Generalization of DPCA Processing for Multichannel SAR/GMTI Radars,” IEEE Trans. Geosc. Remote Sens., vol. 51, No. 1, pp. 560-572, Jan. 2013.

[15] Curlander, John C., McDonough, Robert N., “Synthetic aperture radar”, New York, Wiley, 1991.

(a) (b) (c)

Fig. 6. EDPCA images (first row) and CFAR products (second row) applying a Pfa=10-5. Each column represents the results for (a) target 1, (b) target 2 and (c)target 3, obtained with their respective best-matched filters.