919741.pdf

8/9/2019 919741.pdf

1/14

8/9/2019 919741.pdf

2/14

2 International Journal of Antennas and Propagation

Tx-Rx Tx-Rx Tx-Rx

Object

Acquisition B-scan of an object Image of an object

d 0 d nd n

Figure 1: Imaging process of GPR data.

T i m e

( n s )

0

0.2

0.4

0.6

0.8

1

1.2

X (m)1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2

T i m e

( n s )

0

0.2

0.4

0.6

0.8

1

1.2

X (m)1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2

Figure 2: B-scan of GPR data for basic circular and rectangular objects.

This research tries to interpret geometrical representa-tion from a B-scan of GPR data based on template matching.

The pattern of object under test will be interpreted by comparing its data with the training data using a simpledecision tree method.

2. The Proposed System

Proposed system for identication [ 16] is shown in Figure 3.Firstly an output GPR (a le a.out) is extracted to generatereected signal E(z ) (V/m) as function of time (ns). Thediff erence of the output GPR format and the input format of the C.45 [17] software tool, needs a program to convert thoseformats. The program is based on Java called DataCreator, NamesCreator, TstCreator, and AhCreator which generates ale .data, a le .names, a le .tst, and le .ah, respectively.The rst two les are needed to generate the training dataand the last twoles are needed to generate and to test thetest data.

The C.45 software tool needsa built-tree of training datato predict a decision. Generation of built-tree can be doneafter le .data and le .names have been generated. Valuesof training data are calculated to determine entropy and gainratio for each attribute. Entropy is used to measure how informative is a node in a tree, and gain ranks attribute tobuild a decision tree of the training data.

Interpretation of a test data can be done after the le .tstis generated.Using the built-tree, a testeddata record that has

unknown attribute values can be classied by estimating theprobability of the various possible results.

Finally the ResultShower.jar program converts the outputdecision of C.45 in text form to visual pattern form to show the visual of the tested unknown object. These steps of theprocess are summarized in algorithm 1.

As example on this experiment, training data consistof 30 samples of three types (circular, triangular, andrectangular) of object data. For each type object data has 10diff erent size samples ( r : 7, 10, 12, 13, 15, 17, 18, 19, 20, and23). Each sample has 10180 attributes of the le .data, so thetraining data has totally 305400 attributes. Figure 4 shows thebuilt-tree of the training data.

2.1. GPR Test Range Model. The model GPR test range is abox with 3 meter length, 3 meter wide, and 2 meter high(3 3 2 meter), containing a half full of sand. A 1 GHztransmitter and receiver of GPR, which has 10 centimetrex distance and located 20 centimetre above the sand, movesto x direction in 5 centimetre intervals until 1 meter length(20 scan positions). In each interval, the pulse signal with12 ns width is transmitted. The sands parameter relativepermittivity is 3.0, conductivity is 10 2 Siemens/meter, andrelative permeability is 1. A simple buried object (with threetypes of patterns: circular, triangular, or rectangular cross-section) is made from perfect conductor material and is liedin the sand at 50cm depth. Figure 5 shows the model of simulation.

8/9/2019 919741.pdf

3/14

International Journal of Antennas and Propagation 3

Training object Tested object

GPR simulator(GprMax2D ver. 2.0)

GPR simulator(GprMax2D ver. 2.0)

GPR test rangelaboratory

Preprocessing

Tree generator(C.45 software tool)

C.45 tested dataC.45 training data

C.45 built tree

Tree interpreter(C.45 software tool)

ResultShow.Jar(show the visual object output)

Procedure algorithm number 11. Create A-scan les from the training and the test data2. For all A-scan les do

2.2. Match the object feature to C4.5 input format2.3. Create data les, name les, ah les and test les2.4. Save the data le and the name le in a training folder,2.5. Save the test le and the ah le in a test folder

3. While there is a training folder do3.1. Create built-tree using the C4.5 tree generator tools

4. While there is a test folder do4.1. Interpret the test data using C4.5 tree interpreter tools

5. Show the visual of the tested objectEnd algorithm number 1

Extract reected signal Ez Format input conversion

PreprocessingExtract reected signal Ez Format input conversion

Ez training data

DataCreator .Java(x.data) NamesCreator .Java(x.name)

Ez tested data

TestCreator .Java(x.tst) AhCreator .Java(x.ah)

2.1. Extract the reected signal E(z )

Figure 3: Proposed system and algorithm number 1.

Possible answer Possible answer

Object

RectangularNon

rectangular

Circular Triangular

Tx5time278 > 2.14551 Tx5time278 2.14551

Tx5time282 > 1.9039Tx5time282 1.9039

OptionsFile stem

Read 30 cases (10180 attributes) from training#.dataDecision Tree:Tx5Time278 > 2.14551: RECTANGULAR (10.0)Tx5Time278 C4.5.exe -f training#C4.5 [release 8] decision tree generator

Figure 4: Built-tree of the training data with algorithm number 1.

Table 1: The test result based on algorithm number 1.

DUT Sc no.Sc no. 0 Sc no. 1 Sc no. 2 Sc no. 3 Sc no. 4 Sc no. 5 Sc no. 6 Sc no. 7

TRI 8 True True True True True True False TrueTRI 9 True True True True True True False TrueTRI 21 True True True True True True False TrueTRI 22 True True True True True True False TrueCIR 8 True True False False False False False False

CIR 9 True True False False False False False False



REC 8 True True False False True False False False

REC 9 True True False False True False False False

REC 21 True True False False True False True False

REC 22 True True False False True False True False

8/9/2019 919741.pdf

4/14


Table 2: The test result based on algorithm number 2.

DUT Sc no.Sc no. 0 Sc no. 1 Sc no. 2 Sc no. 3 Sc no. 4 Sc no. 5 Sc no. 6 Sc no. 7

TRI 8 True True True True True True True TrueTRI 9 True True True True True True True True

TRI 21 True True True True True True True TrueTRI 22 True True True True True True True TrueCIR 8 True True True True True True True TrueCIR 9 True True True True True True True TrueCIR 21 True True True True True True True TrueCIR 22 True True True True True True True TrueREC 8 True True True True True True True TrueREC 9 True True True True True True True TrueREC 21 True True True True True True True TrueREC 22 True True True True True True True True

Antena dipole

Air

ObjectSand

Move

1000 mm

100 mm

500 mm 1 2 0 0 m m

2000 mm

3000 mm

3000 mm1500 mm1000 mm

Figure 5: GPR test range model.

Table 3: Correction factor for all scenarios.

Scenario no. 0 k =1Scenario no. 1 k =1.06Scenario no. 2 k =0.9Scenario no. 3 k =0.95Scenario no. 4 k =0.8Scenario no. 5 k

=1.13

Scenario no. 6 k =0.85Scenario no. 7 k =1.2

For training purpose, 30 samples of three types (circular,triangular, and rectangular) of object data are generated.Each type of object data has 10 diff erent size samples. Testdata consist of 12 samples of the three types of objects inwhich each type has 4 diff erent sample sizes.

Successful GPR modelling based on the nite-di ff erencetime-domain (FDTD) theory [ 18] has been reported by many authors [ 1922].

Giannopoulos [ 19] has developed GPR modelling facili-ties called GprMax2D/3D ver. 2.0 which are used to generatethe GPR data. The output simulator can be a le b.outand le a.out to represent hyperbolic image in binary leand ASCII le, respectively. Sampling experiment by usingreal GPR tool will be done in GPR test range ICTR-ITBlaboratory to validate and verify the GPR data simulator. Anexample of input and output GPR data simulator is shownin Figure 6. The research takes les a.out as input of theproposed system to generate training data and tests data les.

2.2. Testing Scenarios. There are scenarios to cover boththe same environment test condition and the di ff erentenvironment test condition. The test data for each scenarioconsist of 12 samples of the three types of object in whicheach type has 4 diff erent sample sizes ( r : 8, 9, 21, and 22).The total samples test data for all scenarios are 96 samples.These scenarios are as follows.

(1) Tested and training objects are in the same 50centimetre depth condition.

(i) Scenario number 0. The sand and the testedobject have the same parameters as in thetraining data (the sand: r = 3, r = 1, and theobject: perfect conductor). Tested and training

objects have the same environmental condition.(ii) Scenario number 1. The sand has the sameparameter as in training data and the testedobject has diff erent material (the sand: r = 3,r =1, and the object: fresh water, r =81).

(iii) Scenario number 2. The tested object has thesame parameter as training data and the sandhas diff erent permittivity (sand: r = 4, r = 1,and the object: perfect conductor).

(iv) Scenario number 3. The sand medium andthe tested object have diff erent parameterscompared to training data (sand: r =4, r =1,and the object: fresh water, r =81).

8/9/2019 919741.pdf

5/14


+0.00000e+00

+2.35865e 02

+4.71731e 02

+7.07596e 02

+9.43462e 02

+1.17933e 01

+1.41519e 01

+1.65106e 01

+1.88692e 01

+2.12279e 01

+2.35865e 01

+2.59452e 01

+2.83038e 01

+3.06625e 01

+3.30212e 01

+3.53798e 01+3.77385e 01

+4.00971e 01

+4.24558e 01

+4.48144e 01

+4.71731e 01

+4.95317e 01

+5.18904e 01

+5.42490e 01

+5.66077e 01

+5.89664e 01

+6.13250e 01

+6.36837e 01

+6.60423e 01

+6.84010e 01

+7.07596e 01

+7.31183 e 01

+7.54770e 01

+7.78356e 01

+8.01943e 01+8.25529e 01

+8.49116 e 01

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00 1.92590e+00 3.80957e+00 5.18314e+01

9.87165e+01 3.17852e+02 5.30018e+02 7.51751e+02 9.57002e+02

9.07277e+02 8.37664e+02 8.55610e+02 8.54798e+02 7.92935e+02 7.13688e+02 5.77958e+02 4.29555e+02 3.33900e+02 2.30923e+02 3.31163e+01

+1.65418e+02

+2.75217e+02

+3.78985e+02

+5.64460e+02+7.37562e+02

+8.16947e+02

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+3.61483e 03 9.11636e 03

+4.88417e 02

2.35021e 02+1.02810e 01

+1.11528e 01

+4.22328e 02

+2.01777e 01

+1.47463e 01

+6.72124e 02

+1.61598e 01

+1.91622e 01

+7.72128e 02

+9.24497e 02

+1.60353e 01

+9.53088e 02

+1.54310e 02

+6.67077e 02

+7.95314e 02 1.44620e 02 5.34592e 02

+2.43086e 04

2.34835e 02 1.12051e 01 1.17961e 01

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+0.00000e+00

+7.22966e 03

+1.79156e 02

+1.70995e 01

+3.76354e 01

+1.00544e+00+1.71708e+00

+2.34036e+00

+2.82184e+00

+2.78803e+00

+2.67887e+00

+2.78650e+00

+2.90974e+00

+2.79286e+00

+2.51783e+00

+2.21036e+00

+1.95383e+00

+1.68660e+00

+1.29783e+00

+7.83453e 01

+3.23215e 01 3.61834e 02 4.46616e 01 9.77479e 01

1.44818e+00 1.76413e+00 2.06395e+00

#GprMax2D, Ver 1.5 Rev.2#title: circular object model in sand#iterations: 509#DX: 0.01 metre #DY: 0.01 metre#DT: 2.35865e 11 secs

#Number of traces: 20#scan: 20 100 120 5 0 110 120 5 0 1.000000 1 e+09 sine

TX at: 00105 00120, RX at : 00115 00120TIME(NS) EZ(V/m) HX(A/m) HY(A/m)

Input: a circular object

#medium: 3.0 0.0 0.0 0.01 1.0 0.0sand

#domain: 3.0 2.0

#box: 0.0 0.0 3.0 1.0 sand

#cylinder: 1.5 0.4 0.1 pec

#scan: 20 1.0 1.2 0.05 0.0 1.1 1.2

0.05 0.0 1.0 1.0e9 sine

#title: circular object model#messages: y

Each le has10180 rows data

20406080

100120140160180200

D e p

t h ( c

m )

50 100 150 200 250 300Distance (cm)

#dx dy: 0.01 0.01#time window: 12 e 9

CIR#0 10b.out b

#geometry le: CIR#0 10b.geo

Output: a le a.outA le b.out. Ez scans and plots

using MATLAB

Scanned le b.out

Time (ns)

A m p l

i t u

d e

( V / m

)

Plotted le b.out

Figure 6: An input and an output GPR data simulator.

(2) Tested and training objects are in the di ff erent depth.

(i) Scenario number 4. The sand and the testedobject have the same parameters as in thetraining data (the sand: r = 3, r = 1, andthe object: perfect conductor), and depth of the

tested object is 40 cm.(ii) Scenario number 5. The sand and the tested

object have the same parameters as in thetraining data (the sand: r = 3, r = 1, andthe object: perfect conductor), and depth of thetested object is 55 cm.

(iii) Scenario number 6. The sand has the sameparameter as in training data and the object tobe tested has diff erent material (the sand: r =3, r = 1, and the object: fresh water, r = 81),and the depth of the tested object is 40 cm.

(iv) Scenario number 7. The sand has the sameparameter as in training data and the tested

object has diff erent material (the sand: r = 3,r =1, and the object: fresh water, r =8), anddepth of the tested object is 55 cm.

2.3. The Test Result Based on Algorithm Number 1. Anexample of interpretation of the test result for scenarionumber 6 is shown in Figure 7. This scenario in whichtested and training objects are in the di ff erent environmentalcondition, generates the extreme result, that is, the most of interpretation is false. On the other hand, scenario number0, in which tested andtraining objects are in the sameenvironmental condition, generates the best result, that, is allinterpretation is true.

The complete test results are summarized in Table 1.From 96 tested data samples, the system can truly wellinterpret only for 50 samples (52%). Algorithm 1, which rep-resent direct comparison between test and training data withtemplate matching method, has inaccurate interpretation

8/9/2019 919741.pdf

6/14


Device under test Results of the interpretation

D:\ALGORTM#1>T.exe -f training# training#.ah

C4.5 [release 8] decision tree interpreter

Tx5Time278: inserting 1.151200

Tx5Time282: inserting 5.543930Decision:



Tx5Time278: inserting 2.353900Decision:



Tx5Time278: inserting 1.680600Tx5Time282: inserting 1.186690Decision:

TRI#1 21 40.tst

CIRCULAR CF = 1.00 [0.871.00]

1.00 [0.871.00]

1.00 [0.871.00]

CIR#1 8 40.tst

RECTANGULAR CF =

REC#1 9 40.tst

TRIANGULAR CF =

Figure 7: An example of the partial test result for scenario number6.

for condition where the tested and the training object havediff erent environmental condition.

3. The Feature ExtractionAn eff ort to enhance the performance of the system is doneby adding a feature extraction of the object to algorithm 1.This improved the algorithm called as algorithm 2.

Investigation of reected signal E(z ) received by receiverfor 30 samples of three types of (circular, triangular, andrectangular) object of training data carries out an importantinformation, that is, each type of object has a unique patternsignal. In the same type of object, diff erent size will generatediff erent amplitude signal with the same pattern signal. Thisphenomenon can be seen in Figure 8.

Afterward, the unique pattern signal of object is extractedas the objectfeature, that is, a representation of the object.To simplify the feature and to reduce the number attributeof complex extracted signal pattern, the time variable isexcluded by sorting values of the E(z ) from the smallestto the largest, and then a linear extrapolation is applied.Changing the testing condition ( r , r , r , and depth) willshift the position in time axis and change the amplitudesignal without changing the pattern of object.

By this process, each reected signal pattern of anobject, which in algorithm 1 has 10180 attributes, will betransformed to a straight line that only has3 attributes, calledslope of the line (a), intersection point of the line with theE(z ) axis (b), and standard deviation ( s). Furthermore, theseattributesare used as the object representations areprocessed

Table 4: Size of prediction of tested object.

Size of testedobject r (cm)

Prediction result (cm)Rectangular Circular object Scenario

Size r Size r 8 7.5 8.0

Sc no. 09 8.4 8.921 21.0 21.822 21.2 21.8

8 8.1 10.6

Sc no. 49 9.2 11.721 21.0 26.822 21.7 27.1

8 7.5 9.6

Sc no. 59 8.3 10.521 20.6 22.722 21.0 22.7

8 7.9 9.6

Sc no. 29 8.9 11.221 22.2 23.522 22.6 23.9

8 7.5 5.1

Sc no. 19 8.5 5.621 21.2 19.622 21.4 22.8

8 7.6 6.7

Sc no. 79 8.4 7.721 20.1 20.622 20.7 23.1

8 8.7 6.6

Sc no. 69 10.0 8.821 23.1 21.922 24.1 25.2

8 8.4 5.5

Sc no. 39 9.5 7.121 21.1 20.622 21.6 23.0

using template matching method by comparing attribute of tested data to training data, the same as in algorithm 1.

Figure 9 shows algorithm 2 illustrated in each step,starting from extracting the reected signal, extracting thefeature, and matching the tested and training attribute datausing decision tree to generate the interpretation result.

3.1. The Test Result Based on Algorithm Number 2. Thesystem based on algorithm 2 is tested by the test data andscenarios the same as in the system based on algorithmnumber 1. An example of interpretation of the partial testresult for scenario number 5 is shown in Figure 10 and allthe test result is summarized in Table 2. For 96 test samplesin diff erent scenarios, the result showed that the systemcorrectly interprets all the samples. Algorithm number 2

8/9/2019 919741.pdf

7/14


0 0.2 0.4 0.6 0.8 1 1.2Time (s)

7 8 9 10Time (s)

10 8

1000

500

0

500

1000

1500Rectangular object, r = 13cm

Rectangular object, r = 13cm

10 9

100

50

0

50

E z

( V / m

)

E z

( V / m

)

(a) Rectangular

0

0

0.2 0.4 0.6 0.8 1 1.2Time (s)

8 9 10

0

50

Time (s)

10 8

1000

500

500

1000

1500

E z

( V / m

)

E z

( V / m

)

10 9

50

Circular object, r = 13cm

Circular object, r = 13cm

(b) Circular

8 9 10

0

20

Time (s)

Time (s)0 0.2 0.4 0.6 0.8 1 1.2

10 8

1000

500

0

500

1000

1500

E z

( V / m

)

E z

( V / m

)

10 9

40

20

Triangular object, r = 13cm

Triangular object, r = 13cm

(c) Triangular

Figure 8: Received signal pattern for three types of the basic object.

has accurate interpretations for all environmental testingconditions.

4. Calculation of the Object SizeBased on reected signal E(z ) for 30 samples training data of three types (circular, triangular, and rectangular), the maxi-mum and minimum signals amplitudecan be determined. Inthe same type of object, diff erent size will generate diff erentamplitude signal with the same pattern signal. By linear orlogarithmic extrapolation of signals amplitude, a formula tocalculate another size of objects in the same object type withsimilar environmental condition ( d , r , r , and r ) can begenerated as in Figure 11. Unfortunately for the triangularobject, the generated formula is constant. Constant relation-ship between amplitude and size of triangular shape can bedue to the backscattered rays which are not directed towardsthe antenna.

The formula for a tested object has di ff erent condition(d , r , r and r ) with the training object, which is need tobe corrected by a correction factor as can be derived fromFigure 12.

Intrinsic impedance, transmission, and reection factorsare as follows:

=

=

, T = 211 + 0

, = 2 11 + 2

,

T up = 2 1 /r 1 1 /r 0 +

1 /r 1 =

21 +

r 1 /r 0 =

21 + r 1 ,

T pu = 2 1 /r 0 1 /r 0 + 1 /r 1 =

2 r 1 /r 01 + r 1 /r 0 = 2 r 11 + r 1 ,

T pb = 2

1 /rb

1 /rb + 1 /r 1 = 2

r 1 /rb

1 + r 1 /rb = 2

1 + rb /rp ,

T bp = 2 1 /r 1 1 /r 1 + 1 /rb =

2 rb /r 11 + rb /r 1 = 2

1 + r p /rb,

pb =b pb + p =

1 /rb 1 /r p

1 /rb + 1 /rp = rb r p rb + rp ,

bp = p b p + b =

1 /r p 1 /rb1 /r p + 1 /rb =

r p rb rp + rb ,

Ez = Ez 0 e2 pd T up pb T pu + Ez 0e2 pd e2bd b

T up T pb bp T bp T pu,

p = rpr p

2 1 + r pr p2

11 / 2

N pm

,

b = rb rb 2 1 +

rbrb

2

11 / 2

N pm

.

(1)

8/9/2019 919741.pdf

8/14


0 0.2 0.4 0.6 0.8 1 1.2Time (s)

7 8 9 10Time (s)

Procedure algorithm number 21. Create A-scan les from the training and the tested data2. For all A-scan les do

2.2. Extract feature of the object

2.2.2. Sort attributes from smallest to largest value2.2.3. Extrapolate linearly the sorted attributes2.2.4. Take attributes a, b and s as the object feature

2.3. Match the object feature to C4.5 input format2.4. Create data les, name les, ah les and test les2.5. Save the data le and the name le in a training folder,

Save the test le and the ah le in a test folder3. While there is a training folder do3.1. Create built-tree using the C4.5 tree generator tools

4. While there is a test folder do4.1. Interpret the tested data using C4.5 tree interpreter tools

5. Show the visual of the tested objectEnd algorithm number 2

Extractedattributes asthe objectfeature:

D: \ ALGORITHM#2 >T.exe -f training training.ah REC#0 13.tst


s: inserting 0.790000

Decision:

RECTANGULAR CF = 1.00 [0.881.00]

1000

500

0

500

1000

1500

E z

( V / m

)

E z

( V / m

)



10 8

10 9

100

50

0

50

D:\ ALGORITHM#2>C4.5.exe -f trainingC4.5 [release 8] decision tree generator

Options:File stem

Read 30 cases (3 attributes) from training1.dataDecision Tree:s 0.86 :| s 0.91 : TRIANGULAR (10.0)

Tree saved

1 201 401 601 801 1001 1201 1401 1601

A m p

l i t u

d e

( V / m

)

Sorting number

120 100

80

60 40 20

020406080

REC#0 1350

REC#0 1350

Sorting number1 201 401 601 801 1001 1201 1401 1601

A m p

l i t u

d e

( V / m

)

120 100

80

60 40 20

020406080

REC#0 1350

REC#0 13 50Linear (REC#0 1350)

y = 0.043x 37.08R2 = 0.779

30 samples of training in C4.5format:0.03, 22.13,0.86,RECTANGULAR 0.03, 27.19,0.84,RECTANGULAR 0.04, 35.72,0.79,RECTANGULAR 0.04, 37.09,0.78,RECTANGULAR 0.05, 39.73,0.77,RECTANGULAR 0.05, 42.29,0.76,RECTANGULAR 0.05, 44.80,0.77,RECTANGULAR 0.05, 46.07,0.77,RECTANGULAR 0.06, 49.86,0.78,RECTANGULAR 0.07, 60.38,0.81,RECTANGULAR 0.02, 18.53,0.89,CIRCULAR 0.02, 19.67,0.88,CIRCULAR 0.03, 24.95,0.88,CIRCULAR 0.03, 27.46,0.87,CIRCULAR 0.03, 29.29,0.88,CIRCULAR 0.03, 30.17,0.88,CIRCULAR 0.04, 31.45,0.88,CIRCULAR 0.04, 32.79,0.88,CIRCULAR 0.04, 34.25,0.87,CIRCULAR 0.06, 48.67,0.91,CIRCULAR 0.01, 10.53,0.93,TRIANGULAR 0.01, 10.42,0.93,TRIANGULAR 0.01, 10.39,0.93,TRIANGULAR 0.01, 10.36,0.93,TRIANGULAR 0.01, 10.45,0.93,TRIANGULAR 0.01, 10.38,0.93,TRIANGULAR 0.01, 10.35,0.93,TRIANGULAR 0.01, 10.91,0.92,TRIANGULAR 0.01, 10.41,0.93,TRIANGULAR 0.03, 24.30,0.95,TRIANGULAR

The tested data in C4.5 format:0.04, 37.09, 0.78

2.1. Extract the reected signal E(z )

2.2.1. Take attributes signal E(z ) of the object

a = 0.04

b = 37.08

s = 0.78

Figure 9: Proposed algorithm number 2 and its illustration.

The correction factor for di ff erent environmental conditionis

k

= |Ez 1||Ez 2| =

Ez 0 e2 pd T up pb T pu + Ez 0e2 pd e2bd b T up T pb bp T bp T pu 1Ez 0 e2 pd T up pb T pu + Ez 0e2 pd e2bd b T up T pb bp T bp T pu 2

. (2)

As example for scenario number 1, correction factor kbecame

|Ez 1||Ez 2| =

1 pb + e2bd b T pb bp T bp

. (3)

For r p =3, d =50 cm and rb =81, the factor correction is

T pb = 2 1 /rb 1 /rb +

1 /r 1 =

2 r 1 /rb1 + r 1 /rb

= 2

1 + rb /rp = 2

1 + 81 / 3 =0.323,

T bp = 2 1 /r 1 1 /r 1 + 1 /rb =

2 rb /r 11 + rb /r 1=

21 + r p /rb =

21 + 3 / 81 =1.5,

8/9/2019 919741.pdf

9/14


Device under test Results of the interpretation

D:\ALGORTM#2>T.exe -f training training.ah

C4.5 [release 8] decision tree interpreters: inserting 0.930000Decision:



s: inserting 0.890000Decision:

Decision:



s: inserting 0.830000

TRI#0 21 55.tst

TRIANGULAR CF 1.00 [0.881.00]

CIR#0 8 55.tst

CIRCULAR CF =

=

REC#0 9 55.tst

RECTANGULAR CF =

1.00 [0.881.00]

1.00 [0.881.00]

Figure 10: An example of the partial test result for scenario number 5.

0

20

40

60

80

100

120

0 5 10 15 20 25Size (cm)

A m p

l i t u

d e

0

50

100

150

200

250

300

0 5 10 15 20 25

A m p

l i t u

d e

Size (cm)

y = 2.3305x + 43.24R2 = 0.9729

y = 27.792ln(x ) + 5.1622R2 = 0.9672

y = 8.8814x + 46.674R2 = 0.9382

y = 109.06ln(x ) 106.48R2 = 0.9891

Circular object Rectangular object

y = 2.33x + 43.24 y = 27.79ln(x ) + 5.16

y = 8.88x + 46.67 y = 109.06ln(x ) 106.48

Triangular object

y constant = 32

Figure 11: Relationship of amplitude and size for same type of the object.

pb = rb r p rb + r p =

81 3 81 + 3 =0.677,

pb = r p rb r p + rb =

3 81 81 + 3 = 0.677,

b = 2 1 +

2

11 / 2

= 2 109 81 2 3 108

1 + (0.0022)2 11 / 2

=0.2,

|Ez 1||Ez 2| =

k = 1

(0.677 + e2(0.2)0.5 0.323 0.677 1.5)

= 1.06.(4)

8/9/2019 919741.pdf

10/14


Rx Tx Ez Ez 0r 0

r 1 T up T pu

pd

d b

pb T bpT pb

brb

ba

Figure 12: Reected signal of nonperfect conductor object.

Figure 13: GPR test range in ICTR-ITB laboratory.

Table 3 shows that the correction factor for all scenarioshas been investigated in this research.

Using the formula and the correction factor, size of testedobject can be predicted. Table 4 shows the prediction resultfor rectangular and circular tested object. Unfortunately, dueto the constant relationship between signal amplitude andobject size in generated formula for triangular object, thisapproach did not success in predicting the triangular testedobject. The overestimate or underestimate size prediction forthe circular and rectangular objects occur due to the linear orlogarithmic approximation of the experimental result graphas shown in Figure 11.

5. Testing Measurement in the Laboratory

Testing data based on sampling experiment using real GPR tool hasbeen done in GPRtest range at ICTR-ITBlaboratory.The tool consists of test range, GPR scanner, spectrumanalyser, and personal computer with the software tools forcontrolling scanner and processing data. Figure 13 shows theGPR test range in ICTR-ITB laboratory.

Three types objects are tested. These objects are buriedobjects (circular, triangular, and rectangular cross-sectionwith diameter 9, 20, and 8 cm, resp.). The object made from

perfect conductor material is buried in the sand at 35, 25, and25 cm depth, respectively. Figure 14 shows the measurementresult for the circular object.

From Figure 15 which represents the zooming andadding cursor on B-scan, the depth of buried object as thetop of hyperbolic line in y axis is shown by the cursor on517, surface of the sand on 491, and position of the ooron 566. The real height of the sand is 100 cm represented by value 566 491 = 75 point and the position of the objectis represented by value 566 517 = 26 point. So, the realposition of the buried object is 26 / 75

100 cm

= 34.7 cm

depth.Algorithm number 2 has been applied to the measured

tested data. The reected signal received by GPR receiver thatis represented by 20 A-scan of 20 point trace experimentresult is extracted to get only the reected signal of buriedobject by capturing the reected signal in the axis rangewhere the hyperbolic line is located and then convertingthe signal to the binary data in ASCII le for furthermanipulation.

By sorting values of the binary data from the smallestto the largest and then applying a linear extrapolation, thefeature parameter of the buried object can be obtained, thatis, a =0.00007, b =0.0307, and c =0.8829.

8/9/2019 919741.pdf

11/14


(a) Files A-scan of 20 trace points

201 rowsof data

(b) A-scan for the rst trace

(c) The received signal (Ez ) (d) The raw B-scan (e) The processed B-scan

Figure 14: A-scan and B-scan of the tested circular object.

300

350

400

450

500

550

600

X : 3 Y : 491Index: 0.08842RGB: 0.752, 0.841, 0.841

X : 7 Y : 517Index: 0.01728RGB: 0.311, 0.311, 0.433

X : 8 Y : 566Index: 0.05365RGB: 0.686, 0.799, 0.799

Figure 15: Zooming and marking the B-scan.

Matching the feature parameter with the built-tree thathas been generated from training data using the C.45 toolas illustrated in Figure 9 will generate a decision of buriedobject interpretation. The extract feature parameter and theinterpretation result of this testing are shown in Figure 16.

The same process is applied to the measured tested datafor triangular and rectangular objects, generating feature

parameter with values of a = 0.0001, b = 0.048, and c =0.9524, and values of a =0.0001, b =0.055, and c =0.8592,respectively.

The amplitude of extracted signal Ez for circular objectis 119mV/m. Entering this value to the equation y =2.33x + 43.24 results in the size value, x of 32.5cm. Thetesting experiment condition in laboratory is di ff erent fromtraining condition, so the size value should be correctedusing correction factor derives from ( 2) as follows:

k =Ez 01 e2 p1d 1 T up 1 T pu1Ez 02

e2 p2d 2

T up 2

T pu2

,

d 1 =50, d 2 =35, 0 =8.86 1012,r 1 =3, =0.01, r 2 =5, =0.01,

k = 1e2(1.09)(0 .5) 0.732 1.268 5 e2(0.84)(0 .35) 0.618 1.382 =

0.29.

(5)

The size prediction of the test circular object can be derivedby multiplying x by factor k, that is, 0.29 32.5 =9.4cm.

Similar process can be applied to rectangular and trian-gular tested object. All the test result of measured test data inthe ICTR-ITB laboratory can be summarized in Figure 17.

8/9/2019 919741.pdf

12/14


a = 0.00007b = 0.0307c = 0.8829

0.06

0.04

0.02

0

0.02

0.04

0.06

0.08

0 200 400 600 800 1000

y = 7E 05x 0.0307R2 = 0.8829

2

7 05 0. 3072 . 2

0.3

0.2

0.1

0

0.1

0.2

0.3

0.40 200 400 600 800 1000

Figure 16: Feature extraction and test result of the circular object.

Object under test Test result

Buried object Depth(cm)

Size(cm)

Depth(cm)

Size(cm)Results of the interpretation

35 9

D:\C45\ALGORTM#2>T.exe -f training



34.7 9.4

25 20


C4.5 [release 8] decision tree interpreters: inserting 0.952400Decision:

25.6

25 8




25.3 7.7

training.ah TEST SILINDER.tst

CIRCULAR CF = 1.00 [0.871.00]

1.00 [0.871.00]

1.00 [0.871.00]

training.ah TEST LIMAS.tst

training.ah TEST KUBUS.tst

TRIANGULAR CF =

RECTANGULAR CF =

Figure 17: The test result of measured data in the ICTR ITB Laboratory.

8/9/2019 919741.pdf

13/14


6. Conclusions

This paper proposed a system for identifying and predictingtarget shape and size of buried basic object on surfaceground-penetrating radar system. Feature parameters of object are extracted by using simple proposed algorithmwhich is based on linear extrapolation. 96 samples of basicobjects which are laid underground and have both same anddiff erent condition with the data training are tested. Resultsshow that all basic objects can be correctly interpreted andthe proposed algorithm number 2 can improve the perfor-mance of the shape object interpretation from 52% to 100%.Three sampling experiments of buried object have been donein GPR test range ICTR-ITB laboratory and the result of thetest showed that the developed system works properly.

The proposed method is appropriate to be implementedin the industry to enhance the capability of GPR tools.The next output GPR target is not only in a hyperbolicrepresentation but also can directly interpret the shape andthe size of the detected object. The main application isfor the underground utility infrastructure detection andinterpretation such as the telecommunication and electricalcable and duct which have regular shape and size of therectangular or circular cross-section object.

However, there is a weakness in predicting the size of triangular object due to the constant relationship betweenamplitude and size of the triangular shape that is generatedfrom B-scan GPR data. For further research, instead of B-scan GPR data, using C-scan GPR data as the system inputmight solve this problem, even though the generation of C-scan GPR data needs more time.

References[1] P. D. Beukelaar, J. Dijkstra, M. Meinster, and P. Wilms, Locat-

ing a buried culvert and the detection of underground cablesand pipelines using GPR, in Proceedings of the 10th European Meeting of Environmental and Engineering Geophysics, Utrecht,The Netherlands, September 2004.

[2] L. Capineri, S. Ivashov, T. Bechtel et al., Comparison of GPR sensor types for landmine detection and classication,in Proceedings of the 12th International Conference on Ground Penetrating Radar , Birmingham, UK, June 2008.

[3] N.Aggarwaland W. C. Karl,Line detection in images throughregularized hough transform, Journal of IEEE Transactions onImage Processing , vol. 15, no. 3, pp. 582591, 2006.

[4] S. Shihab and W. Al-Nuaimy, Image processing and neuralnetwork techniques for automatic detection and interpre-tation of GPR, http://www.wseas.us/e-library/conferences/crete2002/papers/444-717.pdf .

[5] M. Nishimoto and K. Shimo, A Method for detecting shal-lowly buried landmiles using GPR signatures, in Proceedingsof the International Symposium on Antennas and Propagation(ISAP 04), Sendai, JAPAN, 2004.

[6] X. Xu, E. L. Miller, C. M. Rappaport, and G. D. Sower,Statistical method to detect subsurface objects using array ground-penetrating radar data, Journal of IEEE Transactionson Geoscience and Remote Sensing , vol. 40, no. 4, pp. 963976,2002.

[7] E. Pasolli, F. Melgani, M. Donelli, R. Attoui, and M. de Vos,Automatic detection and classication of buried objects in

GPR images using genetic algorithms and support vectormachines, in Proceedings of the IEEE International Geoscienceand Remote Sensing Symposium (IGARSS 08), pp. II525II528, Boston, Mass, USA, July 2008.

[8] P. Gamba and S. Lossani, Neural detection of pipe signaturesin ground penetrating radar images, IEEE Transactions onGeoscience and Remote Sensing , vol. 38, no. 2, pp. 790797,2000, http://www.ieeexplore.ieee.org/ .

[9] H. S. Youn and C. C. Chen, Automatic GPR target detectiionand clutter reduction using neural network, in 9th Interna-tional Conference on Ground Penetrating Radar , Proceedingsof SPIE, pp. 579582, Santa Barbara, Calif, USA, April 2002.

[10] A. C. Gurbuz, Feature detection algorithms in computer images[Ph.D. thesis], Georgia Institute of Technology, Atlanta, Ga,USA, 2008.

[11] D. J. Daniels, Ed., Ground Penetrating Radar , The Institutionof Electrical Engineering, 2nd edition, 2004.

[12] A. J. Devaney, Geophysical diff raction tomography, IEEE Transactions on Geoscience and Remote Sensing , vol. 22, no. 1,pp. 313, 1984.

[13] E. M. Johansson and J. E. Mast, Three-dimensional ground-

penetrating radar imaging using multifrequency di ff ractiontomography, in Advanced Microwave and Millimeter-WaveDetectors, Proceedings of SPIE, pp. 196203, July 1994.

[14] D. Laksameethanasan, 3D Modelling and Imaging Based onTransmission Di ff raction Tomography and Algebraic Recon-struction Techniques as Applied in NDT , University of Kassel,Germany, 2004.

[15] N. Rachmana, N. Iftari, Sugihartono, and A. B. Suksmono,Comparison of surface and cross-hole tomography radar,in Proceedings of the International Conference on Electrical Engineering and Informatics, vol. 1, no. 172175, June 2007.

[16] N. Rachmana, Sugihartono, Hendrawan, and A. B. Suksmono,Interpretation target pattern of a buried basic object onsurface GPR system, International Journal on Electrical Engi-neering and Informatics, vol. 1, no. 1, pp. 5363, 2009.

[17] J. R. Quinlan, C4. 5: Program For Machine Learning , MorganKauff man, 1993.

[18] K. S. Yee, Numerical solution of initial boundary valueproblems involving Maxwells equation in isotropic media, Journal of IEEE Transactions on Antennas Propagation, vol. 14,no. 3, pp. 302307, 1996.

[19] A. Giannopoulos, GprMax2D/3D, Users Guide, 2005,http://www.gprmax.org/ .

[20] A. Giannopoulos, Modelling ground penetrating radar by GprMax, Construction and Building Materials, vol. 19, no. 10,pp. 755762, 2005.

[21] J. Irving and R. Knight, Numerical modeling of ground-penetrating radar in 2-D using MATLAB, Computers & Geosciences, vol. 32, no. 9, pp. 12471258, 2006.

[22] J.D.Irving andR. J. Knight,Numerical simulation of antennatransmission and reception for crosshole ground-penetratingradar, Journal of Geophysics, vol. 71, no. 2, pp. K37K45, 2006.

8/9/2019 919741.pdf

14/14

Submit your manuscripts athttp://www.hindawi.com

919741.pdf

Documents

Transcript of 919741.pdf