Bistatic landmine and IED detection combining vehicle and drone mounted GPR sensors

emo e sensing Le e Bis a ic Landmine and IED De ec ion Combining Vehicle and D one Moun ed GPR Senso s Ma ia Ga cia-Fe nandez 1,* , Ann Mo gen hale 2, Yu i Al a ez-Lopez 1, Fe nando Las He as 1 and Ca ey Rappapo 2 1G oup o Signal Theo y and Communica ions, Uni e si y o O iedo, 33203 Gijón, Spain; [email p o ec ed] (Y.A.-L.); [email p o ec ed] (F.L.H.) 2Depa men o Elec ical and Compu e Enginee ing, No heas e n Uni e si y, Bos on, MA 02115, USA; [email p o ec ed] (A.M.); [email p o ec ed] (C.R.) *Co espondence: ga [email p o ec ed] Recei ed: 31 July 2019; Accep ed: 29 Sep embe 2019; Published: 2 Oc obe 2019   Abs ac : This wo k p oposes a no el G ound Pene a ing Rada (GPR) sys em o de ec landmines and Imp o ised Explosi e De ices (IEDs). The sys em, which was nume ically e alua ed, is composed o a ansmi e placed on a ehicle and looking o wa d and a ecei e moun ed on a d one and looking downwa ds. This combina ion o e s bo h a good pene a ion and a high esolu ion, enabling he de ec ion o non-me allic a ge s and mi iga ing he clu e a he ai –soil in e ace. Fi s , a as ay acing simula o was de eloped o ind p ope con igu a ions o he sys em. Then, hese con igu a ions we e alida ed using a ull wa e simula o , conside ing a la and a ough su ace. All simula ions we e pos -p ocessed using a as and accu a e Syn he ic Ape u e Rada (SAR) algo i hm ha akes in o accoun he cons i u i e pa ame e s o he soil. The SAR images o all con igu a ions we e compa ed, concluding ha he p oposed con ibu ion g ea ly imp o es he a ge de ec ion and he su ace clu e educ ion o e con en ional o wa d-looking GPR sys ems. Keywo ds: landmine de ec ion; Imp o ised Explosi e De ice (IED); G ound Pene a ing Rada (GPR); d one; bis a ic ada 1. In oduc ion The non-in asi e de ec ion o hidden o bu ied objec s has a ac ed an inc easing in e es due o i s p ac ical applicabili y in se e al ields such as ci il enginee ing (s uc u al and oad inspec ion), secu i y and de ense (landmine de ec ion), and a cheology, among o he s [ 1 ]. These echniques a e able o de ec he concealed objec s wi hou physically in e ac ing wi h hem o he su ounding medium. Fu he mo e, hey can e en be used o image he inspec ed a ea. Elec omagne ic induc ion, he mal imaging, nuclea quad upole esonance o G ound Pene a ing Rada (GPR) a e some well-known examples o non-in asi e echniques. Among hese echniques, GPR has been widely used o subsu ace imaging applica ions [ 2 ]. I is based on ansmi ing an elec omagne ic wa e and de ec ing he sca e ed wa es a he ai –soil in e ace and om he bu ied a ge s, p o iding a ada image o he unde g ound. One o i s main ad an ages is ha i can de ec bo h me allic and dielec ic a ge s. Howe e , his echnique is qui e sensi i e o he soil he e ogenei y, he soil su ace oughness and he possible low con as be ween he soil and a non-me allic a ge [ 3 ]. As a esul , i equi es ca e ul con igu a ion and ad anced signal p ocessing echniques o o e come hese issues and imp o e he de ec abili y o he sys em. GPR sys ems can be classi ied using di e en c i e ia. Acco ding o he dis ance be ween he an ennas and he soil, hey can be classi ied as g ound-coupled o ai -launched sys ems. The o me usually allow a be e pene a ion in o he soil and a e less a ec ed by he e lec ions a he ai –soil Remo e Sens. 2019,11, 2299; doi:10.3390/ s11192299 www.mdpi.com/jou nal/ emo esensing Remo e Sens. 2019,11, 2299 2 o 14 in e ace (p o ided he an ennas a e well-ma ched o he soil impedance). Howe e , hey need o be in con ac wi h he soil, which also slows down he scanning speed and should be a oided when sea ching o dange ous a ge s such as landmines and Imp o ised Explosi e De ices (IEDs). The la e a oid he in e ac ion wi h he soil, bu he s ong obscu ing clu e (due o impedance misma ch a he ough ai –soil in e ace) g ea ly comp omises he de ec ion o bu ied a ge s. GPR sys ems can also be classi ied as Fo wa d-Looking GPR (FLGPR) [ 4 ] and Down-Looking GPR (DLGPR) [ 5 ]. In ehicle moun ed FLGPR sys ems, he an ennas look ahead o a ehicle, wi h an angle o incidence ha helps o maximize TM (T ans e se Magne ic) wa es pene a ion in o he soil and/o o minimize e lec ions om he ai –soil in e ace backsca e ed o he ecei e . Howe e , hey ha e lowe esolu ion (being di icul o dis inguish whe he he a ge s a e o e o unde he su ace) and sensi i i y (since much o a la - opped a ge ’s e lec ions a e in he o wa d opposi e di ec ion om he ansmi e ). Conce ning DLGPR sys ems, he an ennas a e pe pendicula o he soil su ace, which yields highe esolu ion a he expense o s onge clu e . In landmine de ec ion, he scanning sys em mus keep a sa e y dis ance om he inspec ed a ea in o de o a oid he h ea o explosion, which hus s ongly a o s FLGPR. To add ess his issue, a GPR sys em on boa d an Unmanned Ae ial Vehicle (UAV) o a d one has been ecen ly p esen ed o subsu ace imaging [ 6 , 7 ]. This sys em p o ides high esolu ion subsu ace images since i allows he cohe en combina ion o measu emen s using a Syn he ic Ape u e Rada (SAR) algo i hm. Howe e , he s ong clu e a he ai –soil in e ace clea ly deg ades he de ec ion capabili ies, especially when he con as be ween he a ge and he soil is low. I mus also be no iced ha , al hough bis a ic SAR sys ems ha e gained an inc easing in e es in he las yea s [ 8 ], mos GPR sys ems (bo h DLGPR and FLGPR) adop a monos a ic o a quasi-monos a ic con igu a ion. I would be desi able o combine he ad an ages o FLGPR and DLGPR sys ems in o de o ob ain bo h good pene a ion in o he soil and high esolu ion. This a icle is de o ed o analyzing his no el GPR con igu a ion. As shown in Figu e 1, a ansmi e is placed on a ehicle wi h he an enna looking ahead and a ecei e is placed on a UAV wi h he an enna poin ing s aigh down o he soil su ace. A as ay- acing me hod has been de eloped o ind easible con igu a ions o he sys em. Then, he esul ing con igu a ions we e accu a ely analyzed wi h a Fini e-Di e ence F equency-Domain (FDFD) me hod. These con igu a ions as well as he mul imonos a ic con igu a ion (co esponding o jus a DLGPR sys em on boa d a mo ing UAV) we e compa ed by pos -p ocessing he simula ed sca e ed ield wi h a SAR algo i hm. TX RX Ta ge Figu e 1. Scheme o he no el GPR sys em combining FLGPR and DLGPR. 2. Me hodology 2.1. Scena io Ray- acing and FDFD me hods a e used o simula e a 2D GPR scena io, such as he one shown in Figu e 2, wi h a a ge bu ied in he soil. In his scena io, he e a e T ansmi e s (TX) placed a posi ions ( = 1, ..., T ) and, o each ansmi e , he e a e R ecei e s (RX) a posi ions (whe e Remo e Sens. 2019,11, 2299 3 o 14 subindex = 1, ..., R deno es he ecei e and supe index he ansmi e ). The soil is cha ac e ized by i s ela i e pe mi i i y e s and i s conduc i i y σs . The a ge is assumed o ha e a ci cula shape, wi h adius δ g and cen e ed a coo dina es ( x g , y g ). I is cha ac e ized by i s cons i u i e pa ame e s e g and σ g . The simula ion is pe o med a N equencies, assuming ei he TE (T ans e se Elec ic) o TM (T ans e se Magne ic) pola iza ion. ... ... ... ... ... R x T x Ta ge Soil Figu e 2. 2D GPR modeling scena io. 2.2. Ray-T acing Ray- acing (RT) is a geome ical op ics me hod ha models p opaga ion by ollowing s aigh ays [ 9 ]. Al hough i is less accu a e han con en ional ull-wa e me hods, i equi es a much lowe compu a ional e o . Thus, i is use ul o as modeling o la ge scena ios a se e al equencies and o se e al ansmi e and ecei e posi ions. 2.2.1. Field Compu a ion The con ibu ion o he elec ic ield o a ay impinging a gi en ecei e in ai is calcula ed acco ding o Equa ion (1) o Equa ion (2), depending on whe he he ay comes om he e lec ion a he soil in e ace o om he a ge . In hese equa ions Einc is he inciden ield ampli ude; A and A a e used o ake in o accoun he ansmi e and ecei e an enna beamwid h; Gin and Gou a e he in-plane and ou -o -plane geome ical sp eading ac o s [ 10 ]; Γ and τ deno e he e lec ion and ansmission coe icien s; Rm , αm and βm a e he o al ay pa h-leng h and he a enua ion and phase cons an s in medium m (whe e m= 0 is ai and m=s is soil). The ay- acing me hod implemen ed in his con ibu ion calcula es he pa h leng h in each medium ( Rm ), which is hen mul iplied by he p opaga ion cons an s so as o pe o m a mul i equency simula ion. Esoil =EincA A GinGou Γai –soil exp(−jβ0R0)(1) E a ge =Einc A A GinGou τai –soilΓsoil- a ge τsoil-ai exp(−αsRs)exp(−j(β0R0+βsRs)) (2) Fo hese ay- acing simula ions, each ansmi e and ecei e is cha ac e ized by i s angle o incidence wi h espec o he soil ( θi and θi , espec i ely) and by i s 3-dB beamwid h. The e ms A and A model he an enna assuming a cosq pa e n (whe e q is calcula ed acco ding o he an enna beamwid h). Remo e Sens. 2019,11, 2299 4 o 14 Assuming a mode a ely lossy soil, mul iple e lec ions a e no conside ed and he e lec ion and ansmission angles a e calcula ed using Snell’s law wi hou aking in o accoun he conduc i i y o he soil. These angles a e hen used o compu e he e lec ion and ansmission coe icien s ( Γ and τ ), which do inco po a e he conduc i i y o he soil. 2.2.2. Implemen a ion Usually, many ays a e launched om each ansmi e o p ope illumina ion o he scena io [ 11 ]. Howe e , o educe he compu a ional ime equi ed o mul iple ays, only he ays ha come om he specula e lec ion a he ai –soil in e ace and he ays coming om he a ge a e used. The o me can be calcula ed di ec ly using simple geome ical ela ions. Fo he la e , we es ima e he angles o he inciden ays ha hi he le and he igh sides o he a ge (i.e., he poin s (x g −δ g , y g) and (x g +δ g , y g) ). These es ima ions equi e i s calcula ing he e ac ion poin s a he ai –soil in e ace o hose poin s on he a ge . Then, many ays a e launched be ween he compu ed angles. I one o hese ays (a e i s e lec ion a he a ge ) is close han a gi en h eshold ( h ) o a ecei e , i is assumed ha ay hi s ha ecei e and i is used o compu e E a ge . A de ailed lowcha o he implemen ed app oach is shown in Figu e 3. Calcula e specula ay a ai -soil in e ace Es ima e e ac ion poin s a he ai -soil in e ace o poin s (x g-δ g, y g) and (x g+δ g, y g) Launch N ays be ween hese poin s Find in e sec ion wi h soil su ace (i.e. e ac ion poin ) Calcula e ansmi ed ay ( om ai o soil) In e sec ion wi h a ge ? Calcula e e lec ed ay ( om a ge o soil) In e sec ion wi h soil su ace? Calcula e ansmi ed ay ( om soil o ai ) Calcula e dis ance om he closes ay d d < h Fo each ay Fo each pai TX-RX Fo each RX Fo each TX This ay hi s he RX Compu e E a ge Compu e Esoil Yes Yes Yes Figu e 3. Flowcha o he ay- acing implemen a ion. Remo e Sens. 2019,11, 2299 5 o 14 2.3. FDFD The 2D Fini e Di e ence F equency Domain (FDFD) algo i hm is well sui ed o nea ield analysis o dielec ic o me al a ge s om abou 0.1 o 30 wa eleng hs in size placed in lossy, ough dielec ic backg ounds [ 12 ]. This algo i hm simula es nea ield sca e ing om objec s ha a e o elec ical sizes ha a e pa icula ly di icul o model (less han abou 30 wa eleng hs), illing a desi able niche be ween geome ic op ics me hods (high equency o elec ically la ge sca e e s) [ 13 ] and Bo n app oxima ion me hods (low equency o elec ically small sca e e s) [ 14 ]. The sca e ing objec s and backg ounds can bo h be lossy dielec ics o any con as and any loss angen . The 2DFDFD algo i hm subdi ides space in o uni o m Yee cells and applies simple ini e di e ences o desc ibe he 2D pa ial di e en ial Helmhol z wa e equa ion o which one dimension, ypically z, is in a ian . Te mina ion o he space is done by using a pe ec ly ma ched laye (PML) o minimize sca e ing om he compu a ional bounda ies [ 15 ]. Compa ed wi h 3DFDFD me hods which equi e i e a i e (and slow) GMRES (Gene alized Minimal Residual Me hod) o LGMRES (“loose” GMRES) sol e s, he simple 2DFDFD algo i hm uses di ec ma ix in e sion o bo h he TM and TE subclasses o p oblems. Compu a ional ime is ela i ely as and complex geome ies a e easy o model. 2.4. In e sion To compa e he esul s o he di e en con igu a ions, he simula ed ield is ep esen ed in he ime-domain (B-scan) and pos -p ocessed wi h a SAR algo i hm. SAR e lec i i y a poin 0 o he in es iga ion domain is gi en by Equa ion (3), whe e R is he pa h leng h be ween he - h ansmi e (loca ed a ), he poin whe e he e lec i i y is calcula ed 0and he - h ecei e . ρ( 0) = N ∑ n=1 T ∑ =1 R ∑ =1 E( n, , )exp(+jβ0R )(3) Assuming ee-space p opaga ion, R would be equal o k − 0k+k − 0k . This assump ion p o ides good esul s when he incidence angle is close o no mal incidence and he pe mi i i y and conduc i i y o he soil a e low. When hese condi ions a e ul illed, i is possible o de ec he objec in he SAR image a app oxima ely √e sd dep h (being d he ue dep h o he bu ied a ge ). To ob ain be e esul s and o de ec he objec a i s eal dep h, he cons i u i e pa ame e s o he soil mus be aken in o accoun . The common app oach consis s o calcula ing he e ac ion poin a he ai –soil in e ace ( o each poin 0 in he in es iga ion domain, and each combina ion o ansmi e and ecei e posi ions). This equi es sol ing a ou h-o de equa ion de i ed o Snell’s Law. Howe e , ins ead o calcula ing he e ac ion poin , R is modi ied so as o conside he pe mi i i y o he soil [ 16 , 17 ]. Thus, R is gi en by Equa ion (4), whe e ns=√ε s−1−√ε s and he o he pa ame e s a e de ined acco ding o he scheme shown in Figu e 4. R =2dpε s−1+d (d −dnscos(2φ )) d +dnssin2(2φ )+d (d −dnscos(2φ )) d +dnssin2(2φ )(4) Remo e Sens. 2019,11, 2299 6 o 14 Figu e 4. Scheme o es ima ing he pa h leng h ( he g een dashed line ep esen s he ue ay pa h). 3. Resul s 3.1. Scena io Con igu a ion The scena io simula ed wi h hese me hods consis s o a low mois u e sandy soil (wi h e s= 2.5 and σs= 0.0125 S/m) whe e a a ge o 2 cm adius is bu ied a 25 cm dep h ( x g = 0 m and y g =− 0.25 m). Bo h me allic and dielec ic a ge s a e conside ed. I he a ge is dielec ic, i is modeled as ini o oluene (TNT) wi h e g = 2.9 and σ g = 0 S/m. Simula ions a e pe o med be ween 3.5 and 5.5 GHz a 10 MHz s eps conside ing TE pola iza ion. Al hough hese equencies a e highe han hose commonly used in GPR, hey ha e been chosen so ha he ada could be ligh enough o be moun ed on boa d a UAV (as i has been al eady p o ed in he p o o ype shown in [ 6 ]). In he ay- acing simula ion, he an enna beamwid h is 30 ◦ and he esul s a e con amina ed wi h whi e Gaussian noise, esul ing in a signal o noise a io o 30 dB. I mus be no ed ha he di ec signal be ween he TX and RX has no been included in he simula ions, since i is expec ed o be emo ed om he ecei ed signal in a eal implemen a ion o he sys em ( hanks o he ac ha i will a i e ea lie han he signal coming om he soil e lec ion and i will likely o be s onge ). Th ee di e en con igu a ions ha e been simula ed: • Mul imonos a ic, whe e he TX–RX (d one moun ed anscei e ) is placed a 65 di e en posi ions be ween down- ack posi ions x=− 0.8 m and x= 0.8 m a y= 1 m heigh . The angle o incidence is 0 ◦ (i.e., he an ennas a e aligned pe pendicula o he soil su ace, wi h main beam poin ing s aigh down). • Mul is a ic, whe e he TX is placed a a ixed posi ion (on a ehicle, a down- ack posi ion x=− 20 m and heigh y= 2.5 m) wi h main beam poin ing a an angle o incidence o 83 ◦ wi h he nominal g ound su ace, and he d one-moun ed RX is looking downwa d and is mo ed o he same posi ions as in he mul imonos a ic case. • Mul ibis a ic, whe e he ehicle-moun ed TX is placed a y= 2.5 m heigh and is mo ed be ween down- ack posi ions x=− 20.8 m and x=− 19.2 m and he d one-moun ed RX is mo ed be ween he same posi ions as in he mul imonos a ic case. Thus, bo h TX and RX a e mo ed cohe en ly. The angles o incidence a e 83◦ o he TX and 0◦ o he RX. The posi ions o he TX–RX in he mul imonos a ic con igu a ion we e se acco ding o hose al eady used in p e ious expe imen al wo k. The posi ions o he TX–RX and he angle o incidence in he mul is a ic and mul ibis a ic con igu a ions we e ound using ay acing simula ions, so as o be able o de ec dielec ic a ge s. The pe o mance o all con igu a ions we e hen e i ied wi h FDFD. Remo e Sens. 2019,11, 2299 7 o 14 3.2. Ini ial Compa ison: Sca e ed Field and B-Scan Be o e applying he in e sion algo i hm, he simula ed sca e ed ields ob ained wi h each me hod we e compa ed, in bo h he equency and ime domains. This compa ison is shown in Figu es 5–7 o he mul imonos a ic scena io wi h a me allic a ge bu ied in he soil. The no malized sca e ed ield in he equency domain is shown o wo obse a ion domain posi ions: x=− 0.8 m (Figu e 5) and x= 0 m (Figu e 6). The in e se Fou ie T ans o m is used o compu e he sca e ed ield in he ime domain (B-scan), as shown in Figu e 7. The ag eemen be ween he sca e ed ield simula ed wi h RT and TE FDFD modeling Ez (o TM z , ela i e o z-axis) is good. The main di e ence is ha in FDFD he ampli ude a he ai –soil in e ace is la ge han he ampli ude a he a ge , whe eas in RT bo h ampli udes a e simila . This migh be due o he ac ha RT only conside s he specula e lec ion a he ai –soil in e ace. This ac also explains ha he sca e ed ields in he equency domain a e mo e simila a x=− 0.8 m (le side o he obse a ion domain) han a x= 0 m (cen e o he obse a ion domain, exac ly o e he a ge ). F equency [GHz] 3.5 4 4.5 5 5.5 Real Pa -1 -0.5 0 0.5 1 RT FDFD (a) F equency [GHz] 3.5 4 4.5 5 5.5 Imagina y Pa -1 -0.5 0 0.5 1 RT FDFD (b) Figu e 5. No malized sca e ed ield a he i s ansmi e - ecei e posi ion ( x=− 0.8 m): mul imonos a ic scena io wi h a me allic a ge . Real pa (a) and imagina y pa (b). F equency [GHz] 3.5 4 4.5 5 5.5 Real Pa -1 -0.5 0 0.5 1 RT FDFD (a) F equency [GHz] 3.5 4 4.5 5 5.5 Imagina y Pa -1 -0.5 0 0.5 1 RT FDFD (b) Figu e 6. No malized sca e ed ield in he middle o he obse a ion domain ( x= 0 m): mul imonos a ic scena io wi h a me allic a ge . Real pa (a) and imagina y pa (b). Remo e Sens. 2019,11, 2299 8 o 14 (a) (b) Figu e 7. B-Scan compa ison om RT (a) and FDFD (b) simula ions: mul imonos a ic scena io wi h a me allic a ge . 3.3. SAR Image Compa ison The inal goal was o compa e he SAR images o each con igu a ion (mul imonos a ic, mul is a ic and mul ibis a ic) o de e mine he bes con igu a ion. Fi s , he SAR images we e ob ained om he RT simula ions and hen he esul s we e e i ied wi h he FDFD simula ions. 3.3.1. Mul imonos a ic Simula ions The SAR image o he mul imonos a ic scena io wi h a bu ied me allic a ge is shown in Figu e 8. Bo h he in e ace and he objec a e clea ly de ec ed. The e lec i i y a he in e ace is la ge in he FDFD simula ion, as expec ed om he p e ious discussion. (a) (b) Figu e 8. SAR image om RT ( a ) and FDFD ( b ) simula ions: mul imonos a ic scena io wi h a me allic a ge . When he bu ied a ge is dielec ic (TNT), i is ha dly de ec ed in he SAR image (as shown in Figu e 9). Thus, in acco dance wi h he ini ial hypo hesis, he an enna con igu a ion mus be imp o ed o be able o de ec non-me allic a ge s. Remo e Sens. 2019,11, 2299 9 o 14 (a) (b) Figu e 9. SAR image om RT ( a ) and FDFD ( b ) simula ions: mul imonos a ic scena io wi h a dielec ic a ge . 3.3.2. Mul is a ic and Mul ibis a ic Simula ions Since he goal is o de ec non-me allic a ge s, he mul is a ic and mul ibis a ic simula ions compa ison was pe o med when he bu ied a ge is dielec ic. SAR images a e shown in Figu es 10 and 11 o he mul is a ic and mul ibis a ic scena ios, espec i ely. In RT simula ions, he specula e lec ions om he soil su ace do no each he ecei e . The e o e, in FDFD simula ions, he known la g ound backg ound is emo ed, hus showing only he a ge -sca e ed esponse. The esul s a e almos he same o bo h con igu a ions, whe e he dielec ic objec is clea ly dis inguishable. The e is also a good ag eemen be ween he RT and FDFD simula ions, hus i can be concluded ha RT is a use ul ool o designing new GPR con igu a ions. (a) (b) Figu e 10. SAR image om RT ( a ) and FDFD ( b ) simula ions: mul is a ic scena io wi h a dielec ic a ge .