Nondestructive imaging of shallow buried objects using acousticcomputed tomography
Waheed A. YounisDepartment of Electrical and Computer Engineering, University of Western Ontario, London,Ontario N6A 5B9, Canada
Stergios Stergiopoulosa)
Defence and Civil Institute of Environmental Medicine, Toronto, Ontario M3M 3B9, Canadaand Department of Electrical and Computer Engineering, University of Western Ontario, London,Ontario N6A 5B9, Canada
David HavelockInstitute for Microstructural Sciences, National Research Council, Ottawa, Ontario K1A 0R6, Canada
Julius GrodskiDefence and Civil Institute of Environmental Medicine, Toronto, Ontario M3M 3B9, Canada
~Received 11 June 2001; accepted for publication 19 February 2002!
Acoustical techniques have been used to do subsurimage reconstruction for imaging and classifying buried ojects for the past two decades. Most of these techniqueseither well-to-well tomography1 or surface-to-welltomography.2–4 However, their methodology is not approprate for detecting buried land mines or hazardous wasteterials, since they require ground disturbances that couldhazardous for demining and destructive for archeologicalplications.
A recent study5 investigates two-dimensional~2D! im-aging of shallow buried objects by using ultrasound B-sctomography.6 The system concept includes a linear arrayreceivers and transmitters operating in the frequency rang~1–5! kHz. In this case, three-dimensional~3D! imaging canbe accomplished by volume rendering of consecutiveimages.7 If the linear sensor array is replaced by a planarray, then 3D beamforming techniques6 could provide the
a!Author to whom correspondence should be addressed. [email protected]
J. Acoust. Soc. Am. 111 (5), Pt. 1, May 2002 0001-4966/2002/111(5)/2
ce-se
a-bep-
nfof
Dr
volume visualization of the underground area without cobining multiple 2D images.
Advancements in remote sensing have led to the deopment of nondestructive subsurface detection methodsas infrared ~IR! imaging,8,10 ground penetrating rada~GPR!,9,10 seismic refraction, electromagnetic sensing10 andelectrical conductivity. All these techniques use perturbatioin seismic or electromagnetic waves to detect, locate,identify the buried objects. Currently, GPR and IR imagiare widely used in demining applications. The performanof GPR for detecting landmines degrades considerably wincreased moisture in the ground and, like the electromnetic sensors, it fails to detect nonmetallic objects. Theimaging methods can only detect recently buried~1 monthperiod! objects. The poor performance characteristics ofexisting nondestructive imaging techniques can be improby using data and image fusion techniques in a multisensystem that has shown to have the potential for reliable mdetection and classification.10
Another challenge of international significance is tdisposal of buried hazardous waste, which requires reliaestimation of the location and nature of the buriil:
substances.11 For archeologists, subsurface imaging tecniques are useful for assessing the historical significancesite before beginning costly digging.12
The present paper considers an alternative approacnondestructive 3D imaging by using the concept of coputed tomography~CT!, which is successfully used for noninvasive medical imaging diagnostic applications.13 Briefly,medical CT imaging uses x rays to obtain cross-sectioimages, or ‘‘slices,’’ of the human body and image recostruction algorithms that are based on the Radtheorem.14,15 In this investigation we propose subsurface iage reconstruction of horizontal cross sections at differdepths. Once such images of sufficient quality are availaa next step could include implementation of pattern recogtion algorithms to identify and classify the mines. An advatage of our approach is the ability to detect mines thatburied beneath one another, as this is a common deploymprocedure to deceive the conventional mine detectiontems. The method is demonstrated with an experimesetup, which illustrates both the challenges and opportunfor imaging of shallow buried objects using acoustic coputed tomography.
II. PRINCIPLES OF OPERATION OF ACOUSTIC CT
The basic principles of a CT data acquisition proceand the relevant image reconstruction algorithms arecussed in detail in Refs. 3, 13–19. The data acquisition pcess consist of taking projection measurements of the obof interest defined by the imagef (x,y). The projection dataform a 2D array called sinogramp(r n ,u i) (n51,...,N) ( i51,...,M ) where u i is the projection angle andr n is thedistance ofnth detector from the center of the field of view.13
N and M are the total number of detectors in the array athe total number of projections, respectively. This set of pjections is used by CT image reconstruction algorithmsreconstruct the imagef (x,y). If the energy used to illumi-nate the object is nondiffracting~e.g., x rays!, the imagereconstruction algorithms are,iterative algebraic techniquesFourier slice theorem techniquesand filtered backprojection.13–15,17Whereas for diffracting energy~e.g., soundwaves!, these algorithms arefiltered back propagation andFourier diffraction projection techniques.3,16,18,19
In this investigation we propose to implement the Cconcept using acoustic energy to reconstruct subsurfaceages of horizontal cross sections of the entire volume oburied object at different depths. We assume that the recesignal carries information from different ground penetratidepths that are represented as different time delays, indicby the temporal indexj below,
p~r n ,u i ,t j ! ~n51,...,N!, ~ i 51,...,M !, ~ j 51,...,K !.~1!
As a result, for a given time delayt0 , image reconstructionof the projection data setp(r n ,u i ,t0) will provide horizontaltomography images of the buried object of interest corsponding to a specific depth in the ground. Then, repetiof the image reconstruction process for all the time del( j 51,...,K) would define a 3D volume consisting of horizotal underground tomography images. However, the co
2118 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
-f a
to-
al-n-nte,i--ent
s-ales-
ss-o-ct
d-o
m-aed
ted
-ns
-
spondence between time delayt0 and specific depth in theground requires stratification of the underground mediuFor a nonstratified medium, such as the experimental seof this study discussed next, the above correspondencetween time delay and depth is not valid due to the multipeffects of the acoustic waves in the homogenous undground.
Figure 1 depicts the experimental setup of the propoacoustic tomography concept for nondestructive imagingshallow buried objects. The acoustic sensors~microphonearray! and source array, shown in Fig. 1, do not have a ctact with the surface of the ground. Thus, the whole systcould be mounted on a platform that may be deployed byarmored vehicle or is remotely controlled. For each transmted acoustic pulse, the receiving array collects the sigwhich is reflected–refracted from the internal undergrouburied objects and the ground inhomogenouties.
Therefore, the proposed approach addresses the invproblem, which is the underground 2D and 3D imaging. Prameters related to the propagation problem~forward prob-lem!
the physics of acoustic/elastic wave propagating inthree-dimensional structure,
what waves are expected to generate in the geometrthe experimental setup,
what are their expected arrivals, their travel time atheir amplitude?
are not required by the proposed approach to addressimaging problem. Although, solutions to the forward prolem can provide the essential parameters for a better imreconstruction approach,20 they impose a practical implementation problem for a number of nondestructive imagunderground applications. In particular, to identify the 3propagation characteristics, it would require significaamount of information in terms of boundary conditions, desity estimates, and stratification characteristics. This kindinformation, although essential for an optimum approamakes the acoustic tomography concept impracticaldemining or archeological applications because it requground disturbances to collect the essential information. Ogoal and objectives of this experimental study is the devopment of an acoustic CT technique for nondestructiveaging of underground features that is practically realizabHowever, the theoretical and experimental aspects relewith the propagation problem~forward problem! are dis-cussed in Sec. IV B of this paper, since they are essentiaverify our experimental results.
III. PRACTICAL ISSUES AND PROPAGATIONCHARACTERISTICS
During the course of the present investigation we wconfronted with several challenges. First and foremost wthe issue that the propagation characteristics of the acousignals in this setting were not known. More specificalwhat is received by the receiving microphone array, depicin Fig. 1, is a superposition of direct waves~direct arrivalsthrough air!, surface waves, subsurface waves, reflecterefracted energy from buried objects, reflected energy fr
Younis et al.: Imaging of buried objects
. The circle
FIG. 1. ~a: Top! Experimental setup of the acoustic CT concept including the receiving 32-microphone array and the four-speaker source array.~b: Bottom!Top schematic view of the experimental setup. The dark features indicate buried objects. The triangular base was on the surface of the sandindicates the field of view of the 90 cm line array receiver during the CT data acquisition process.
no
o
tore-c-
fi-
various inhomogeneous ground layers and noise. Eachceived acoustic wave front has a different time delay depeing upon the propagation path it followed and the speedsound in various sections of that path. Because of the p
J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
re-d-f
or
air-to-ground-to-air coupling interface, another problem isreceive sufficient acoustic energy out of the soil by theceiving array5 that is essential for reliable image reconstrution. The soil is a highly attenuating medium and it is dif
2119Younis et al.: Imaging of buried objects
ne
FIG. 2. Experimental setup of microphones to examipropagation effects for signals shown in Fig. 3.
t
nntha
unwryinlo
o
-tn
hein
te
te
nesasy
tw
f
2
-de-
t ofircleingts,sentredby
a-
ralmi-
er-
ignorttia-eHz
alFM
thethe, ass of
nals
ro-, re-
e; 2
nes
cult for acoustic waves in the frequency range of 1–5 kHzpenetrate deep in the ground.21 Although high frequencyacoustic waves~2–10 MHz! produce good image resolutioin ultrasound imaging, they are not suitable for undergrouimaging applications due to their severe attenuation inground. Thus a lower frequency range of 0.2–3.0 kHz wselected due to its characteristics to penetrate in the grothough it produces poor image resolution. Furthermore,constructed a frequency-modulated FM pulse, linearly vaing from 200 Hz to 2000 Hz, cascaded with a Parzen wdow. In order to generate a plane wave acoustic signailluminate the underground area of interest, a linear arrayfour synchronized acoustic point sources was used, as shin Fig. 1.
The poor signal-to-noise-ratio~SNR! in the received sig-nal, as mentioned above, was mainly due to the poor airground-to-air coupling interface that induced a severe atteation in the signal of interest. To minimize the impact of tvery poor SNR in the image reconstruction process, wetroduced an adaptive noise~interference! cancellation~AIC!6,22,23 processing scheme. Its impact is demonstrawith real data in Sec. IV D.
IV. EXPERIMENTS
A. Experimental setup
The proposed acoustic CT tomography imaging syswas implemented as shown in Fig. 1~a!. A minefield wassimulated by filling a commercial swimming pool of 4 mdiameter and 0.7 m depth with dry construction grade saA wooden tripod structure with three hollow wooden boxburied in the sand@Fig. 1~b!# was built to support the datacquisition system above the sand. The data acquisitiontem consisted of a 31-microphone receiving line array~Pa-nasonic broad band capsule model WM063P! with 3 cmspacing and a four-speaker line array, mounted at thesides of a horizontal metal bar~3.5 meter long!. Two objects~simulating landmines! were buried in the sand@Fig. 1~b!# atdifferent depths in the range of 10 cm to 50 cm:~1! an inertantivehicle plastic landmine 30 cm in diameter and~2! ametal flat box 5320330 cm3. A set of 32 preamplifiers(amplification540 dB) was placed in the close vicinity othe microphones@Fig. 1~a!#. A 4 kHz, 32-channel antialiasingfilter was used to condition the received signal for the 3
2120 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
o
desd,e--tof
wn
o-u-
-
d
m
d.
s-
o
-
channel 12-bit A/DC~analog-to-digital converter with 8 kHzsampling frequency!. The digitized time series from the 32channel were multiplexed and stored in a SCSI storagevice. The CT data acquisition process included 360 semeasurements at 1-degree interval around a horizontal cdefined by the rotated metal bar with the mounted receivand source arrays@Fig. 1~a!#. For each set of measurementhe four-speaker line array generated a plane wave pulsein the ground; and the data acquisition system was triggesimultaneously to acquire the reflected acoustic signalsthe 31-microphone receiving array.
To align all the received time series from the 360 mesurements, a reference microphone~32nd sensor! was placednear the four-speaker array. This procedure of tempoalignment, using the received signal from the referencecrophone, will be discussed later in Sec. IV C.
The transmitted acoustic signal was a FM pulse genated by a PC through a D/AC~digital-to-analog converter!and the audio amplifiers of the four-speaker array~Fig. 1!.The FM pulse’s characteristics were another major desissue, since they affect the image resolution. A very shpulse results in better image resolution and better differention of signal arrivals from different directions. However, thconstraints imposed by the frequency range of 200–2000did not allow for the design of a very short FM pulse. A triand error process provided a practical choice of a 0.1 mspulse with 100 kHz sampling for the D/AC unit.
B. Propagation characteristics of the acoustic signalsin the underground medium
A set of experiments was conducted to determinepropagation speed, time of arrival, amplitude and SNR ofpropagating acoustic signals in the air and in the groundthese were important parameters to determine the sectionthe received microphone time series that included the sigof interest.
The sections~a!–~f! of Fig. 2 and~a!–~f! of Fig. 3 showthe various experimental arrangements in terms of micphone positions and the corresponding received signalsspectively. More specifically, in Fig. 2, the microphone~a! isthe reference microphone placed near the acoustic sourcmeters away is the microphone~b!, which is one of the mi-crophones of the 31 sensor receiving array. The micropho
Younis et al.: Imaging of buried objects
-g.
e
n
FIG. 3. Received time series for microphone arrangements shown in Fi2. ~a!, ~b!, ~d!, and ~e! show receivedsignals by microphone above thground while ~c! and ~f! shows re-ceived signal by microphones buried ithe sand.~a! and ~c! are signals atacoustic source while~b!, ~d!, ~e!, and~f! are signals at detector side.
al
thnt
s
icsth
y.-pa-nd
c-e
am-
des
~c! and ~f!, shown in Fig. 2, are buried in the sand~5 cmdeep!, and just below the microphones~a! and ~b!, respec-tively. The time delay between the signal wavefront arrivof microphones~a! and ~b! and between~c! and ~f! corre-sponds to the time taken by the signal to travel betweencorresponding microphones through the air and in the sarespectively. These time delays can be estimated fromresults of Fig. 3 that shows the corresponding time seriethe microphones~a!–~f! of Fig. 2.
To maximize the microphones receiving characterista number of experiments were conducted to determine
J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
s
ed,
heof
,e
optimum position configuration for the microphone arraFor example, microphone~d!, shown in Fig. 2, was acoustically isolated from the surrounding acoustic signals progating through the air by using thick foam to cover the sasurface. In another case, the microphone~e!, shown in Fig. 2,was set in a plastic trumpet to maximize its receiving diretivity pattern of the acoustic signals arriving from thground.
From these experiments it became obvious that theplitude of the received signals by microphone~d! @e.g., Fig.3~d!# were much smaller than the corresponding amplitu
2121Younis et al.: Imaging of buried objects
is ca
FIG. 4. Temporal alignment of received signals for the 12th microphone of the receiving array and for all the 360 projections. This signal displaylled‘‘gram.’’
th-
t oe-to
rotth
el o
e
hn
ls
tinte
onroei
inio
hess
wason-or-sig-tinetion
en-edsti-to-
Fig.de-
y
eac-
iseof
of the signals of microphones~b! and ~e!, @e.g., Figs. 3~b!,~e!#. In particular, the differences in amplitude betweenreceived signals~b! and ~e! in Fig. 3, show that a very significant portion of the energy collected by~b! is receiveddirectly from the air and hence the SNR of the componenthe acoustic signal propagating through the ground is vsmall, as expected because of the very poor air-to-groundair coupling. In another experimental set up, the amplitudethe acoustic signal received by microphone~e!, @e.g., Fig.3~e!# has smaller amplitude than that of the signal of micphone~b! @e.g., Fig. 3~b!#. This confirms that the dominansignal components of the received acoustic signals bymicrophone~e! @e.g., Fig. 3~e!# are signals propagating in thground. This observation is confirmed also by the signathe shallow buried microphone~f!, since the time arrival ofthe received signals for microphones~e! and ~f! are nearlyidentical. Furthermore, it appears that the frequency sptrum for the times series~c! and~f! is different from those forthe microphones in the air~d! and ~e!. This was expectedsince the ground acts as a low pass filter and the higfrequencies that do not penetrate the ground they areincluded in~c! and ~f!.
Thus, the time delay of the signal wavefront arrivabetween the reference microphone~a! @e.g., Fig. 3~a!# andmicrophones~e! and ~f!, @e.g., Figs. 3~e!, ~f!#, provides esti-mates of the speed of the acoustic signals propagathrough the sand, which are in the range of 80–100 mesecond, which agrees with a similar experiment reportedRef. 25.
In summary, the segments of the received microphtime series in the range of 180–500 samples for the micphones~b!, ~e!, and~f! include the signal components of thpulses propagating through the air-to-ground-to-air. We wuse this segment of the received signal to define the sgrams for the purpose of underground image reconstruct
2122 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
e
fryo-f
-
e
f
c-
erot
gr/
in
e-
llo-n.
It is worth noting that although the signal loss due to tpoor air-to-ground-to-air coupling was great, this signal lowas tolerable in the image reconstruction process. Thisconfirmed also from the image results that were recstructed using these very low SNR signals. Another imptant observation is the increased duration of the receivednal by microphone~f!, shown in Fig. 3~f! as compared to thain Fig. 3~a!. This is an indication that the received signalFig. 3~f! includes multipath propagation effects in thground. This, however, cannot be considered as justificaof our claim that time delayt j in the projection datap(r n ,u i ,t j ) can be equated with depth, since the experimtal setup with the tank filled with sand is a nonstratifimedium. Therefore, the reconstructed images of this invegation would represent near the surface 2D horizontalmography sections of the underground medium.
C. Experimental results
Each experiment, which was based on the setup of1, generated a three-dimensional data set of projectionsfined by
p~r n ,u i ,t j ! ~n51,...,N!, ~ i 51,...,M !, ~ j 51,...,K !,~2!
whereN532 microphones~31 sensors in the receiving arraplus the reference microphone!, M5360 projections arounda circle, andK58192 samples for each microphone timseries. These 360 sets of time series had to be alignedcording to their reference microphone’s signal@Fig. 3~a!#.Figure 4, shows the aligned received signals for the 12th~arandomly selected! microphone. Such a signal displaycalled ‘‘gram.’’ The alignment of the time series from th360 different projections is clearly evident by the resultsFig. 4.
Younis et al.: Imaging of buried objects
e
FIG. 5. Simulation results using a Shepp–Logan phantom for diffraction image reconstruction integrated with interpolation techniques. Left imageshows thesimulated Shepp–Logan phantom. Upper-middle shows the 2563256 reconstructed image from a sinogram (N531,M5360), without interpolation. Lower-middle presents the 2563256 reconstructed image from the sinogram (N531,M5360) that has been interpolated to (N5256,M5360). Upper-right showsthe 2563256 reconstructed image from a sinogram (N562,M5360), without interpolation. Lower-right presents the 2563256 reconstructed image from thsinogram (N562,M5360) that has been interpolated to (N5256,M5360).
n
enavese36o
esc
erpafanlan)/
ytinthr
jec.n,tela
5e pa-
no-
htthe
ourc-
age-tictheR,
ailshytheing
-
6
1. Interpolation for improved image resolution andsimulations
The image resolution characteristics of a CT recostructed image are related to the number of sensorsN, thesensor spacing and the number of projectionsM @Eq. ~2!#. Atypical x-ray CT scanner deploys approximately 1400 ssors with 1 mm pitch and acquires 1100 projections to hsufficient image resolution for medical diagnostic purposIn our case, however, due to limited resources, the presetup included only 31 sensors and the acquisition ofprojections; hence the image resolution was anticipated tvery poor. Furthermore, because of our sensor spacing~e.g.,pitch 3 cm! the spatial sampling frequency was 33 samplmeter, which indicated that the maximum spatial frequenthat could be handled by this pitch was 16.5 samples/met18
Another complication in terms of the image resolution cabilities of the current experimental setup was due to thethat our experimental observations in Fig. 3 show that oacoustic frequencies below 1 kHz penetrated the smedium @i.e., wavelength in the ground, (100 m/s(1000 Hz to 500 Hz)510 cm to 20 cm#. Thus, the imageresolution was very limited in our diffraction tomographexperiments since the wavelengths of the interrogaacoustic waves were in the range of 10 cm to 20 cm andspatial sampling frequency was 33 samples/meter. As asult, spatial details higher than 16 sample/meter and obdimensions comparable to the wavelength of 10 cm to 20cannot be detected with our current system configuration
To minimize the impact of the poor image resolutioinherent in our system, we used interpolation techniquesdefined in Ref. 18. To assess the performance of this inpolation technique we used diffraction tomography simutions for 800 Hz and the Shepp–Logan phantom14,17,18~left
J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
-
-e
s.nt0
be
/y.-ctyd
gee-ctm
asr--
image in Fig. 5!. The upper-middle image results of Fig.have been generated from a sinogram that has the samrameters as in the [email protected]., M5360, N531 ~500–1000! Hz#. Image reconstruction results using the same sigram and with interpolation parameters (M5360,N5256)are shown by the lower-middle image of Fig. 5. The rigupper and lower reconstructed images of Fig. 5 indicateanticipated improved image resolution performance ofexperimental setup forN562 and with smaller sensor spaing of 1.5 cm~e.g., pitch 1.5 cm!. It is apparent from thesesimulations that the experimental configuration ofN531sensors would not be sufficient to provide reasonable imresolution even for high SNR~which is the case in the simulations!, since the wavelength of the interrogating acouspulses is comparable to the dimensions of the objects andsensor spacing is not sufficiently small. Thus, for poor SNimage resolution would be even worse. For more detabout the image resolution limits of diffraction tomograpin terms of the wavelength of the interrogating energy,size of the object and the spatial sampling of the receivsensor array, the reader may review Ref. 26.
2. Real results
Based on Eq.~2! and Fig. 4, a 2D sinogram for a predefined temporal sample is given by
p~r n ,u i ,t j 5predefined!
~n51,...,N531!, ~ i 51,...,M5360!, ~ j 5predefined!.~3!
Figure 6~a! displays a sinogram for the time delayj 5376.The correspondence between the data sets of Figs. 4 and~a!is that a vertical line for the sensorn512 in Fig. 6~a! repre-
2123Younis et al.: Imaging of buried objects
-
si-
if-
isindri-e
cker-
ndors
FIG. 6. ~a! Sinogram from the data set of Fig. 4 corresponding to the temporal samplej 5376, for the 31microphones and 360 projections.~b! Image reconstruc-tion based on filtered back projection and using thenogram data set shown in~a! with no interpolation~wetsand!. ~c! Image reconstruction based on Fourier dfraction algorithm and using an interpolated version~31to 256 sensors! of the same sinogram data set~a!. Theimage of the buried mine is obvious at the bottom. Thlocation coincides with the expected mine locationFig. 1~b!. ~d! Image reconstruction based on filtereback projection of a data set from a previous expement reported in Ref. 24. This image is for the samburied mine of Fig. 1~b!. The mine location here isslightly different from~b! and~c!. In this case the sandwas dry.~e! Image reconstruction based on filtered baprojection and using a sinogram data set with no intpolation ~temporal samplej 5186!. In this experimentthe sand was wet.~f! Image reconstruction based oFourier diffraction algorithm and using an interpolateversion of the sinogram data set from 31 to 256 sensand for the same data set with those of image~e!. Thelocation of the white image spot in~e! and~f! coincideswith that of the buried metallic object in Fig. 1~b!.
im
e
o
tioti
Fialu
ow
u
inr
oncar-n
r-llic
us-
n-in
ieflythetws:ar-
sents the same data as the vertical line of Fig. 5 for the tdelay j 5376.
The image in Fig. 6~b! has been reconstructed from thsinogram of Fig. 6~a! and for the temporal samplej 5376. Itshows a circular pattern~indicated by the arrow! that its lo-cation coincides with the location and time delay depththe buried plastic mine@Fig. 1~b!#. In this experiment thesand was wet in order to improve the acoustic propagaand compare results with dry sand. The image reconstrucalgorithm for the results of Fig. 6~b! was the Fourier diffrac-tion algorithm18 using interpolated data~from 31 to 256 sen-sors! for the same temporal samples. Thus, the image in6~c! has been derived from the same sinogram as the imin Fig. 6~b!. Their differences represent better image resotion characteristics as those in the simulation results shin Fig. 5.
As discussed earlier, the temporal samples forj5180,...,500, are expected to provide information about bied objects. Although the images of the buried mine~Fig. 6!have very poor image resolution, their spatial location cocides with that of the buried mine as in our previous expements reported in Ref. 24 and shown by Fig. 6~d!. The image
2124 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
e
f
non
g.ge-n
r-
-i-
in Fig. 6~c! was reconstructed using filtered back projectialgorithm on noninterpolated data. The experiment wasried out in dry sand and for slightly different mine locatioand depth with those of Figs. 6~b! and ~c!. Figures 6~e! and~f! show reconstructed images for temporal samplej 5186using filter back projection and Fourier diffraction~interpo-lated data!, respectively. The white spot in both figures coresponds to the expected location of the buried metamine, shown in Fig. 1~b!.
D. Adaptive interference cancellation „AIC… inacoustic CT
The results in Fig. 7 demonstrate the advantages ofing adaptive interference cancellation~AIC! processing tominimize the impact of the poor SNR in the image recostruction process. The implementation of the AIC processour acoustic CT concept has already been discussed brin Sec. II and for more details the reader may reviewpapers.13,22,23 The AIC implementation effort requires thathe 31-microphone array should be rearranged as follo16-microphones should be positioned along the receiving
Younis et al.: Imaging of buried objects
ein the
FIG. 7. Time series from the adaptive acoustic CT experiment.~a! is the signal from the reference microphone.~g! is the signal from the upward microphonthat is considered as the noise~interference! in the AIC process.~b! is the signal from the downward microphone that is considered as the noisy signalAIC process.~i! is the output signal of the AIC process and~f! is the signal from the buried microphone.
roethlsro
-s--e
veFi
mentmees
ew-im-the
ray downwards to receive the noise signal for the AIC pcess. 15-microphones should be positioned along the recing array upwards to receive the interference, which ispropagating in the air component of the transmitted puThe spacing for both the upward and downward micphones was the same.
The time series in Fig. 7~a!, refer to the reference microphone~a! ~see Fig. 2!, located near the array of the acoutic sources. Figures 7~g! and ~b! show the received time series for the upward and the downward facing microphonthat represent the interference and noisy signals, respectithat are essential in an AIC process. The time series in
J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
-iv-ee.-
sly,g.
7~i! show the output of the AIC process and Fig. 7~f! showsthe signal of the buried microphone that defines the tiposition of the wave front arrival in the ground. At this poiit is important to note that a comparison between the tiseries~i! and~f! of Fig. 7 indicates that these two time seriare not similar since AIC filtering process~because of thevery low SNR! has not been very effective to remove all thcomponents of air contributions in the received signal. Hoever, the AIC process has been sufficient to enhance theage resolution for this experimental setup as shown byimage reconstruction results of Fig. 8.
Thus, for the temporal samplej 555 the image 8~a! and
2125Younis et al.: Imaging of buried objects
nt
lar
FIG. 8. Sinograms from the acoustic CT experimeand for the temporal samplej 555. ~a! For the 15-upward microphones.~b! For the 16-downward micro-phones.~c! Output of the AIC process.~d! Image recon-struction using the sinogram shown in~b!. ~e! Imagereconstruction using the AIC sinogram shown in~c!.The AIC image output shows the expected triangubase, shown in Fig. 1.
.Tigre
am
innigio
es
-. Ie
veeroaththth
row
ss inin
us-ssfulon-ess
64-
ofthe
.signn-cy
tly
ele-
nti-res
nddis-og-
~b! are the sinograms for the 15-upward~interference! andthe 16-downward~noisy signal! microphones, respectivelyThese sinograms define the inputs for the AIC processor.output of the AIC processor is the sinogram shown in F8~c!, which is supposed to have improved SNR as compawith the SNR of Fig. 8~b!.
The image reconstructed from the AIC output sinogrof Fig. 8~c!, is shown in Fig. 8~e!. This image [email protected]~e!# demonstrate the efficiency of the AIC processorminimizing the impact of the air-reverberations in the dowward microphones, when compared with the image of F8~d!, which has been derived from the image reconstructof the noisy sinogram of Fig. 8~b! of the downward facingmicrophones.
The features in Fig. 8~e! represent objects and structurabove the ground~the triangular base shown in Fig. 1!, asthis was expected sincej 555, which indicates the early arrival of the signal wave fronts propagating through the airis interesting to note that these triangular base features wcompletely masked in the image results of Fig. 8~d!, andwere only revealed by the AIC process in Fig. 8~e!. Al-though, the number of sensors in Fig. 8~e! were halved fromthose in Fig. 6~b!, the AIC process appeared to be effectiin improving image quality by minimizing the impact of thsurrounding interference noise. However, since the micphone spacing in the AIC process was 6 cm, the spatial spling frequency was 16 samples/meter, which indicatesthe maximum spatial frequency that could be handled bypitch would be insufficient to sample the dimensions ofburied mines, as indicated also by the simulation resultsFig. 5. Therefore, because of the limited number of micphones that were available in this experimental study,
2126 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
he.d
-.n
tre
-m-atiseof-e
were unable to assess the effectiveness of the AIC proceproviding better image resolution for the mines shownFig. 6.
E. Key issues and future directions
The results of this experimental study and the discsions in the preceding sections suggest that a succeimplementation of the proposed acoustic tomography ccept as a nondestructive underground imaging procshould address the following issues.
The receiving microphone array should have at leastdownward-microphone~preferably 128-microphones! withsmaller array pitch to improve the image resolution.
The AIC process is essential to minimize the effectsthe signal arrivals propagating through the air. Thus,number of sensors should be increased to 128~preferably256! to include the 64~preferably 128! upward microphones
The pulse design characteristics should guide the deefforts for the acoustic source array to avoid the nolinearities discussed in Sec. IV A and to include a frequenrange that would allow ground penetration with sufficiensmall wavelengths for improved image resolution.
Finally, if the AIC process would be effective to improvthe SNR, the replica correlation process should be impmented on the microphone time series in order to differeate the signal arrivals from various underground structuand from the air.
Although, the available resources were very limited anot sufficient to address the above issues, the resultscussed in Sec. IV, suggest that the proposed acoustic tom
Younis et al.: Imaging of buried objects
ct
eg
ueo-
taoub
ime
CT
ndan
ba
yu
coimtc
atmheruhsthhe
e
pro
of
ie
c
ofn
s.
c-
nd
ical
of
ical
and
es-
, Al-
r
ic
anro-
a-
ralad.
-
o-a-
nd
-
a
ure
s
3D’’ J.
t.
-at-
raphy concept can be used successfully as a nondestruunderground imaging technique.
At this point it is important to note that the poor imagresults of our investigation are in agreement with the findinof another investigation by Crawford and Kak20 that pro-posed a technique for breast tomography imaging usingtrasonic CT concepts. Although their physical paramet~forward problem! were more favorable than the poor air-tground-to-air coupling~and multipath! imposed in our ex-perimental design, the image quality of their experimenresults was very poor and similar with those reported inpaper. In summary, investigators should anticipate thatcause of the very low frequencies~e.g., long wavelengths!used in acoustic CT imaging applications the associatedage resolution would be poor and that they should not expthe high image resolution that is available by the x-rayscanners.
V. CONCLUSION
The present paper describes a new approach to nostructive subsurface imaging for detecting landminesother buried objects. Subsurface imaging is performedsending acoustic energy into the soil around a circular ptern and collected at the diametrically opposite end of it blinear microphone array. From these collected signals, ssurface images are reconstructed employing image restruction algorithms, in the same manner as done in CTaging. It is assumed that acoustic waves that propagate inground carry information about the subsurface buried objeto the sensors of the receiving array. It is also assumed ththe received signal, the signal amplitude at different tidelays carries information from the different depths of tstratified soil. Consequently, we would be able to reconstthe cross sectional images at different underground dept
Among several challenges, the most important waslimited knowledge of the propagation characteristics. Tvery weak signals of interest, received by the microphonresulted in poor SNR that was another major problem duthe poor air-to-ground-to-air acoustic coupling.
In summary, the results of this investigation provide suporting arguments that the implementation of the AIC pcess in combination with the deployment of large numbersensors in the receiving array could be a valuable method3D subsurface imaging for the detection of shallow burobjects using acoustic diffraction CT.
1K. A. Dines and J. Lytle, ‘‘Computerized geophysical tomography,’’ ProIEEE 67, 1065–1073~1979!.
2W. C. King, A. J. Witten, and G. D. Reed, ‘‘Detection and imagingburied wastes using seismic wave propagation,’’ ASCE J. Environ. E115, 527–540~1989!.
J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002
4A. J. Witten and E. Long, ‘‘Shallow applications of geophysical diffration tomography,’’ IEEE Trans. Geosci. Remote Sens.GE-24, 654–662~1986!.
5Reporting on Science and Technology, ‘‘Archaeology—UndergrouNoises,’’ The Economist, pp. 83, issue of July 2000.
6S. Stergiopoulos, ‘‘Advanced beamformers,’’Advanced Signal ProcessingHandbook: Theory and Implementation for Radar, Sonar and MedImaging Systems~CRC, Boca Raton, 2000!, Chap. 6.
7G. Sakas, G. Karangelis, and A. Pommert, ‘‘Advanced applicationsvolume visualization methods in medicine,’’Advanced Signal ProcessingHandbook: Theory and Implementation for Radar, Sonar and MedImaging Systems~CRC, Boca Raton, 2000!, Chap. 7.
8J. Haystead, ‘‘Portable systems apply infrared technology to detectclassify mines,’’ Vision Systems Design, 36–40, 1997.
9C. J. Vaughan, ‘‘Ground penetrating radar survey in archeological invtigations,’’ Geophysics51, 595–604~1986!.
10J. McFee et al., ‘‘CRAD Countermine R&D Study—Final Report,’’DRES-SSP-174, Defense Research Establishment Suffield, Rlastonberta, Canada, 1994.
11R. C. Benson, R. P. Glaccum, and M. R. Noel,Geophysical Techniques foSensing Buried Wastes and Waste Migration~U.S. Envir. Prot. Agency,Washington, DC, 1982!.
12C. H. Frazier, N. Cadalli, D. C. Munson, and W. D. O’Brien, ‘‘Acoustimaging of objects buried in soil,’’ J. Acoust. Soc. Am.108, 147–156~2000!.
13A. Dhanantwari, S. Stergiopoulos, and I. Iakovides, ‘‘Correcting orgmotion artifacts in x-ray CT medical imaging systems by adaptive pcessing~Part I: Theory!,’’ Med. Phys.28, 1562–1576~2001!.
14A. C. Kak, ‘‘Computerized tomography with x-ray, emission, and ultrsound sources,’’ Proc. IEEE67, 1245–1272~1979!.
15J. Radon, ‘‘Uber die Bestimmung von Funktionen durch ihre Integwerte langs gewisser Mannigfaltigketiten,’’ Ber. Verh. Saechs. AkWisse. Leipzig, Math.-Phys. Kl.69, 262–279~1917!.
16A. J. Devaney, ‘‘A filtered back projection algorithm for diffraction tomography,’’ Ultrason. Imaging4, 336–350~1982!.
17A. C. Kak, Digital Image Processing Techniques~Academic, New York,1984!.
18S. X. Pan and A. C. Kak, ‘‘A computational study of reconstruction algrithms for diffraction tomography: Interpolation versus filtered backpropgation,’’ IEEE Trans. Acoust., Speech, Signal Process.ASSP-31, 1262–1275 ~1983!.
19R. K. Mueller, M. Kaveh, and G. Wade, ‘‘Reconstructive tomography aapplications to ultrasonics,’’ Proc. IEEE67, 567–587~1979!.
20C. R. Crawford and A. C. Kak, ‘‘Multipath artifact corrections in ultrasonic transmission tomography,’’ Ultrason. Imaging4, 234–266~1982!.
21K. Attenboroughet al., ‘‘The acoustic transfer function at the surface oflayered poroelastic soil,’’ J. Acoust. Soc. Am.79, 1353–1358~1986!.
22S. Stergiopoulos, ‘‘Implementation of adaptive and synthetic apertbeamformers in sonar systems,’’ Proc. IEEE86, 358–396~1998!.
23B. Widrow and M. E. Hoff, Jr., ‘‘Adaptive noise cancellation: Principleand applications,’’ Proc. IEEE63, 1692–1716~1975!.
24S. Stergiopoulos, R. Alterson, D. Havelock, and J. Grodski, ‘‘Acousticcomputed tomography for demining and archeological applications,Acoust. Soc. Am.107, 2806~2000!.
25G. D. Larsonet al., ‘‘Air acoustic sensing of seismic waves,’’ J. AcousSoc. Am.107, 2896~2000!.
26A. C. Kak and M. Slaney,Principles of Computerized Tomographic Imaging ~IEEE Press, 1988 or Society of Industrial and Applied Mathemics, 2001!.
