Nicoleta IFTIMIE, Adriana SAVIN. Nondestructive Testing Department,. National
Institute of Research and Development for Technical. Physics, Iasi, Romania. 1 ...
Raimond GRIMBERG, Rozina STEIGMANN, Nicoleta IFTIMIE, Adriana SAVIN Nondestructive Testing Department, National Institute of Research and Development for Technical Physics, Iasi, Romania
1
Outline
Introduction GPR principles The equipment and the studied samples Signal processing Results Conclusions Acknowledgements
2
Outline
Introduction GPR principles The equipment and the studied samples Signal processing Results Conclusions Acknowledgements
3
Introduction GPR is a non-intrusive technique used to obtain information about
the medium below the surface of the earth. The technique relies upon transmission of electromagnetic energy
into the earth, and subsequent reflection of the energy at interfaces of differing dielectric permittivity.
4
Introduction There are a wide range of domains that have used GPR such as Archaeology Geology Civil engineering, Military applications Forensic
5
Outline
Introduction GPR principles The equipment and the studied samples Signal processing Results Conclusions Acknowledgements
6
GPR principles GPR systems uses an impulse time domain waveform and
receives the reflected signal in a sampling receiver. Models of the GPR situation range from a simple single frequency evaluation of path losses to complete 3D time domain descriptions of the GPR and its environment. Modeling techniques include single frequency models, time domain models, ray tracing, integral techniques and discrete element methods. The Finite Difference Time Domain (FDTD) technique has become one of the popular techniques.
7
GPR principles
Block diagram of a generic GPR system 8
Outline
Introduction GPR principles The equipment and the studied samples Signal processing Results Conclusions Acknowledgements
9
The equipment and the studied samples
The GPR system in test site 10
Studied samples
Ballasted ammunition 11
Studied samples
Caliber 76 mm explosive projectiles ammunition
12
Studied samples
120mm HE mortar shell fitted with proximity fuze 13
Studied samples
100mm artillery shells 14
Studied samples
122mm CARGO 15
Studied samples
Anti-tank mine MAT 62B 16
Outline
Introduction GPR principles The equipment and the studied samples Signal processing Results Conclusions Acknowledgements
17
Signal processing
Scan of Cargo projectile- raw data
Simulation using FTDT software GPRMAX 3D
18
Signal processing A-scan processing The mean value of A scan must be near zero
1 A 'n = An − N
N
∑ An
n =1
Noise reduction
An − A 'n−1 A 'n = An + K
An -unprocessed data sample A’n - processed data sample n – the sample number N – total number of samples.
A’n -averaged value An - the current value
The general effect is to reduce the variance of the white noise.
19
Signal processing B-scan processing Imaging with GPR data is frequently hampered by clutters.
Clutters
20
Signal processing B-scan processing The principal problem for a correct interpretation of GPR
images consists in the extraction of unwanted signals as the ones due to transmission of forward wave from Tx to Tr and those due to reflection on the air-soil interface. We define a window of L pixels and subtract from all the pixels
in this window the mean of the pixels in it.
g ( x, y ) = f ( x, y ) −
1 L
i=
L 2
∑ f ( x + i, y ) L
i =−
2
g – filtered image f - raw data L - the window size
This method works well if the magnitude of the ground reflection is uniform across the GPR map. 21
Signal processing B-scan processing We propose a method in which spatial variations in
both the ground reflection magnitude and delay are simultaneously estimate using nonlinear optimization. Because these variations are relatively smooth functions of position, we model them using low-order polynomials.
22
Signal processing B-scan processing The measured data are split into Nseg spatial segments. Over each segment we approximate the spatial
resolution in both the amplitude and time delay of the ground reflection by weighted sums of Chebyshev polynomials, the domain of the segment was normalized to the interval [-1,1].
23
Signal processing B-scan processing Let Ai(x) denote the spatially ground reflection
amplitude and Bi(x) the time delay of the ground reflection peak over segment ith. We approximate Ai(x) and Bi(x) as a sum 4
4
n=0
n= 0
Ai ( x ) = ∑ ain Tn ( x ) ; Bi ( x ) = ∑ bin Tn ( x ) Tn(x) are the Chebyshev polynomials defined by the
recursive relation Tn+ 1 ( x ) = 2Tn ( x ) − Tn −1 ( x ) , T0 ( x ) = 1 ; T1 ( x ) = x
n>1 24
Signal processing B-scan processing Because the reflection coefficient at the air-soil
interface is typically complex, we can represent the coefficients as imag real imag ai = aireal , a , a , a , ... i0 i1 i1 0
bi = bi0 , bi1 , ...
25
Signal processing B-scan processing The frequency-domain ground reflection in spatial segment ith can be estimate as
Giest ( x , f ) = Ai ( x ) H ( f ) exp ( −2π jfBi ( x ) )
H ( f )- windowed frequency response of the radar H
H ( f ) = H ( f )W ( f )
( f )- average frequency spectrum of the raw GPR data within the segment
W ( f )- a window applied when creating the time-domain data
We have used a Hanning window.
26
Signal processing B-scan processing The time domain reflection estimated becomes giest ( x , t , ai , bi ) = Ai ( x ) h ( t − Bi ( x ) ) h ( t ) is the inverse Fourier transform of H ( f ) The measured ground reflection over segment ith can be expressed as gimeas ( x , t ) = giest ( x , t , ai , bi ) + ni ( x , t ) ni(x,t) - the modeling error
27
Signal processing B-scan processing The coefficients
ai and biare found by a least-square error
process, which requires a nonlinear optimization of
(
aˆ i , bˆi = arg min gimeas ( x , t ) − giest ( x , t , ai , bi ) a i
2
)
Once the weighting coefficients
are estimated, the amplitude term Ai(x) and the delay Bi(x) of segment ith are evaluated. Parameter estimation was performed using a nonlinear least square error minimization function, in the Matlab 2009b Optimization Toolbox
28
Outline
Introduction GPR principles The equipment and the studied samples Signal processing Results Conclusions Acknowledgements
29
Results GPR images of Cargo projectile
Original raw data After A-scan processing and ground removal
30
Results GPR images of Cargo projectile
Original raw data After migration 31
Results GPR images of Cargo projectile
Top view 32
Results GPR image of two 76mm caliber projectiles
Top view 33
Results GPR image MAT62B antitank mine
Top view 34
Outline
Introduction GPR principles The equipment and the studied samples Signal processing Results Conclusions Acknowledgements
35
Conclusions GPR has started to be used, with good results, in different types of application, from which, UXO must be nominated. This fact is more important for the countries that have been theatre of war. To can interpret correct the images delivered by GPR, due to the high level of noise and of clutters, it is very important to solve the forward problem using FDTD procedure.
36
Conclusions Applying specific procedures of signal and image processing, the shape of UXO become visible, the type, caliber, number and their coordinate can be determined from simple measurements on the results. These procedures allow the emphasizing of antitank mines from plastic, even if the weight of metallic parts is very small.
37
Special thanks to the Colonel Stefan Lazar, former Chief Inspector of Inspectorate for Emergency Situations “Mihail Grigore Sturdza” Iasi, Romania.
This paper is partially supported by the Romanian Ministry of Education, Research and Innovation under PN II Program – Project 32-134/2008 SysArr and Bilateral Project MENDE Ro-Sk/2011. 38
Thank you very much for your attention! Questions?
39
