Algorithm Overview Based On Image Processing with Electromagnetic

(EM) Techniques in X Band and GA Approach for Depth Estimation of

Shallow Burried Dummy Mines
KC Tiwari*, D. Singhb@ and M. Arora*

*Department of Civil Engineering ( Geomatics), IIT Roorkee, Roorkee - 247667, India

@Department of Electronics and Computer Engineering, IIT Roorkee, Roorkee - 247667, India
Email: chtphd

mail.com,

armfcajitr. rnetinandmanocenaiitrernetin

Radar remote sensing is a very encouraging field

of
research
with many advantages over various other
Detetio ofurid lndmies nd stimtio of techniques. The
conditions existing in the western borders
depth by modeling layered media is a complex and of India match that of sandy7 deserts where surface
computationally intensive task. Microwave remote
..
conditions
areraymjrln
smooth and theetrs
area is hs
also
thecapbiltyopnetatesubroughness
withto penetrate subsurface and itS
brf fayvgtto
sensing with the sensing
capability
conditiofn make them eremey sable foraplcTion
ability to resolve landmines as well as non-lethal targets
cofdar rmoesensingrfordetrectio of minefields.
can therefore be used for subsurface landmine detection.
Various signal deimg ocesin lgrh
A model based on electromagnetic scattering and image
analysis techniques at X-band frequency for sub-surface are under development world over to accurately solve the
detection dmm
of
dandmines bed uroaosad layer mine detection problem. Literature reports neural network
[3], SVM [4] and Markov random models [5] based
paper. An extensive set of lab experiments have bbeen techniques
the detection
applied for solving
problem. Ina a
s
ives cmaaieeprmn
utatv
uz ehiu
carried out using dummy landmines (without explosives)
utatv uz ehiu
a
cmaaieeprmn
and
backscatter observed at different depths. The ra
raw data
found
to
better
when
the
andbackscatterobserved
a
perform
training
and
testing
data
deriesoffer
gendepthces.iThe
esaesprt 4.Sm inlpoesn
ehd
iS processed through a at
series
of image
processing steps,
the detection is carried out using Otsu's
and
. . the landmine
s
for the hyperbola for charecterising
. thresholding .
~~~~~search
the depth estimated by optimizing .through
a GA (genetic
to predict landmine detection, its geometry and
algorithm) based electromagnetic model developed. The response
depth
[5].
Most
works have however
ignored
otwrshv
oee
[]
dph
goe
method does not have any
requirement of separate electromagnetic interactions at the air-ground-landmine
a
ny
mthodnn andotesnt
hatavset
oftimszeparat
training
and test data
set to trequire
train the optimizer
and
inefc. Dail eta aeeautd hoeia oe
nterface.
Daniels et
al have
evaluated
in P-a theoretical
microwave
remote
band (441 model
validte
reults Theresuls
lboraory
MHz,
using
sensing
the teresults.
The results uner
under laboratory
Abstract

paprier Ant extsivesetmoflabdminesexperim tent

validate

conditionusindicatethatsholdetion
fchnidummy lanmins
possible
using thresholding techniques
with dataia
generated in X band and the proposed model is capable of
estimating depth of the buried landmine to a significant

68 cm) en
for buried
H and the results
hv
wl reflectors
upre at depth
yteeprmna
have been well supported by the experimental
investigations conducted in Negev Desert [6]. Models
analyzing electromagnetic interactions at layered rough
surfaces are suitable for estimation of depth of landmine.
Ulaby et al have also proposed a relationship between
radar observation depth as a function of observation
frequency and soil moisture content by considering the
power of an electromagnetic wave incident upon a soil
surface [7]. Despite extensive research however, either
the models developed are very complex and
computationally intensive or there is a requirement of
some a priori information.
Lower frequencies penetrate higher, while higher
frequencies resolve better. Microwave X-band at 10 GHZ,
3 cm with significan subsurface penetration capability
provides an optimal choice as it also has sufficient
bandwidth to permit resolution of mine targets as well as
non-lethal targets such as rocks etc [7]. Besides, at this
wavelength volume scattering from inhomogenities of the
layered media ( small rough surfaces such as in sand
layer) can be ignored because of the dimensions of
inhomogeneous particles and the distance between them

accuracy.

Microwave X band, landmines,

Keywords
electromagnetic scattering, detection, feature extraction,
entropy, thresholding, image analysis. genetic algorithm.

I. INTRODUCTION
Thousands of innocent civilians are killed/
maimed annually due to buried mines scattered all over
the world in which the conflict has long ceased [1,2].
Detection and removal of landmines however is
complicated by variety in types of mines, soil types,
scattering from layered media, vegetation etc. Landmines
are often laid flush with the ground or at shallow depths,
hence their responses to any emitted pulse overlaps with
soil clutter. Landmine detection poses two main
challenges in detecting small shallow buried landmines
which contain little or no metal i.e. reduction of soil effect
and mine feature extraction.

1-4244-0284-0/07/$20.OO 2007 IEEE

331

The experimental setup consists of a monostatic

scatterometer with one pyramidal horn antenna connected
through a circulator to microwave transmitter on one side
and power meter on the other side. A wooden box
measuring 120 cm by 120 cm has been created and filled
with sand for mines to be buried. The system has
provision to move the antennae in both X and Y planes.
The horizontal bars on the two sides (called X-Y
direction) were marked serially from 1 to 24 to make a
grid. The circulator and the horn were moved laterally (Ydirection) from 1-24 at each of the horizontal (Xdirection) positions from 1 to 24 thus making a total of
24x 24 positions. Backscatter readings were recorded at
all the 24x 24 positions.

is of the order of 2.5x10 -3 , both of which are much

smaller than the wavelength [6].
The problem of landmine detection and
estimation of depth using layered media analysis presents
a non-linear problem and hence optimizers such as
gradient descent or direct search methods are not likely to
yield good results because standard optimizers require
objective function to be continuous, differentiable, nonstochastic and linear. A variety of nonparametric
optimization techniques such as neural networks, support
vector machines and fuzzy methods etc have been put to
use [3, 4, 8] but these non-parametric methods often
require different training and testing data sets to train the
optimizer and validate the results. Thus the results are
often representative of the test data and not necessarily
the ground truth.
A new approach to the problem of detection of
landmines and estimation of the depth of landmine by
developing a theoretical model by combining image
processing and signal processing approaches has been
discussed in this paper. A theoretical model based on
electromagnetic interactions at the air-sand-landmine
interface has been used for depth analysis and optimized
using a genetic algorithm based fitness function. The
main advantage of this method is that it does not require
separate training and test data set and therefore the results
represent the ground truth more accurately.

II.

Backscatter reading has been taken for each

mine at depths of 0.5 cm, 1.0 cm, 1.5cm, 2.0 cm, 2.5cm,
3.0 cm, 4.0 cm, 5.0cm, 7.0cm and 10 cm. All the
experiments were conducted in microwave X band at 10
GHz.
The dielectric constant of dummy landmines
without explosives was assumed in the range of 4-10 and
that of smooth dry sand in the range of 3-5. The sand
surface was assumed smooth because at this wavelength (
X band, wavelength - 3 cm), the volume inhomogenieties
are of the order of 2.5x1 0 -3 cm which are much smaller
than the wavelength [6].

III. THEORETICAL MODELING AND

EXPERIMENTAL SETUP

A set of experiments have been conducted with

dummy mines (without explosives) in dry sand. In each
category dummy, anti tank, fragmentation, and influence
mines have been used. The schematic layout of the
experimental setup is shown in Fig-I and the dimensions
of the landmines used are given in Fig-4.

Tranrtikr7
l

LS

1/>

Fig- 1: Schematic Diagram of Monostatic

Scatterometer System

11

4|

IMPLEMENTATION
Clutter removal and mine feature extraction is
the basic process for detection of landmines. However,
any radar image is literally swamped with clutter. Clutter
detected by radar includes initial ground reflection and
background resulting from several scatterers within the
soil. All forms of undesired signals require estimation and
subsequent removal in order to enhance the target signal.
A common method for clutter reduction is to simply
compute the mean vector and subtract this value from
individual pixel reading. This method however fails in
case the contour of the ground surface is not smooth. The
| |aim of mine feature extraction is to classify a signal into
mine or non-mine features and to make a decision
between the two during post processing and extract the
features containing the landmine. A series of steps have
been formulated for extraction of mine features and
estimation of depth which are given in Figure -2.
Raw data was generated in a grid of 24x24 array
which was calibrated using an aluminium sheet which
having conductivity of 3.5 x107 Seimens per meter and
reflectivity coefficient -1 (perfect reflector). The
calibrated data was normalized to put all the data in one
range. The illumination area of the antenna system is not
limited to the pixel size of the image, so there is a

332

specular reflection and diffuse scattering by the air-sand

interface, reflection from the mine buried at depth H,
diffraction and scattering by surface irregularities of the
reflector and volume scattering from the sand layer [6].
The total returned field ER ( Figure-3) at the radar
receiver due to the mine buried at depth H is given by -

significant contribution from the neighboring pixels.

Therefore, a 5x5 convolution filter [11 11 1; 1 1 2 1 1; 1
2 3 2 1; 1 1 2 1 1; 1 1 1 1 1 1] was applied and the
convoluted data was then assumed to be pure reading per
pixel.
To obtain the suspected area where an object
may be buried, it is very important to select a proper
threshold value, say 't'. Otsu's thresholding was used for
detection of landmine [9].
Raiw image

Calibrationt

Convolution Filter

Normalization

ER:= ES + EC1 + EC2 + EC3 .

Ecn

ES ECi EC2 EC3

Ei

Air - Medium 1
z

Thresholding

(Otsu's I Maximum Entropy)

Cdentification of

Sand -Medium 2
T

y,

Suspected area and Detection Figure Test

Mlalugkiof

erge backsctterI

Mine - Medium 3

Suspected Rgion inormalized intesgity

o|f maskekd aresa

Fig 3: Radar wave propagation in an airsoil-mine interface

Electromagnetic Mode ing for Dep0th

Optimisation Using GA

Daniels et al [6] have derived the following

model for estimation of backscatter in a layered media
which can be used for estimation of depth.

*Fig. 2: Flow Chart of Algorithm

where

R23exp(-2y2H)

k -2 n/k= wave No,

01
R

1)

(- 2 ;1HA
H)(
~~~~~~~~~~~~~R
1-2 + R23 expP (-2
+

For reduction of false points, a quantity detection

figure (D) as under was calculated where, A(FG) is
average reading for foreground pixels, A(BG) is average
reading for background pixels and A(FG+BG) is average
reading for all pixels. It was found that if detection figure
obtained is less than 40, it could be claimed that there is
no object buried in sand.

Detection figure

cos1 exp(-sin

ER =s\/4k

Estimated Detth

c7

roughness parameters

incident angle, H depth

2
R 23=3
I1-2

~~~~2-3 -

+ _
44 + F2
( , 2 & 3 refer to first, second & third medium i.e. air,
sand and mine)

A(FG)-A(BG)x1OO
A(FG + BG)

The interaction of an electromagnetic wave with

soil is a complex phenomenon. Figure-3 gives a
schematic diagram of electrical field propagation for
subsurface penetration as in the present case.
The total returned electrical field received at the
scatterometer end ER is the result of simultaneous

2ic

72 X=

333

_2

(propagation constant in second

medium i.e. sand)

genetic algorithm. The GA tool was run several times

with different initial population and the optimized results
for estimation of the depth are given in Table 1. A graph
showing variation of calculated/predicted backscatter and
the observed backscatter with increasing depth is shown
in Figure-5. The variation is as expected and shows
degradation in performance with the increase in depth.
The oscillatory behaviour of backscatter with the depth is
also noticed. An error plot of the calculated/predicted
depth is given in Figure-6 which confirms optimal results.

Genetic algorithms (GA) are robust optimizers,

stochastic search methods modeled on the principles and
concepts of natural selection and evolution. GA is
effective in solving complex problems particularly when
the goal is to find global maxima in a high dimension,
multi model function domain in a near optimal manner.
The optimization in genetic algorithms is done through a
fitness function (also called cost function or objective
function) which is used to assign a fitness value to each of
the individuals in the GA population. The theoretical
model based on electromagnetic theory for calculation of
backscatter (eqn- 1) was used to obtain EC
(predicted/calculated backscatter) and the proposed
fitness function optimized using genetic algorithm. The
proposed fitness function is :FitnessFunction = min Eo - Ec 2
where

Eo

is observed backscatter and

backscatter
.
predicted/calculated

EC

V. CONCLUSION

is

IV. RESULTS AND DISCUSSIONS

The backscatter data collected for different

depths of the experiments was processed through various
steps as discussed and plots generated at each stage to
analyze and detect the presence of landmine features.
Plots generated for raw data, calibrated data and
convoluted data and mine feature extracted after
thresholding are given in Figure-4. It is noticed that the
raw data contains severe clutter from various sources
particularly from the corner of boxes and the plots for the
raw data is highly random in nature. Convolution with a
5x5 kernel filter after calibration results in a smoothened
image with noise around the corners severely restricted.
The convolution plot for each mine, which in these
experiments was kept at the centre of the box
significantly highlights the likely area containing the
mines. The mine features were segmented using Otsu's
thresholding method. It is noticed that although the
thresholded values accurately indicate the location of the
mine but there apparently is enough clutter/noise in the
same range. This highlights the need of preprocessing the
data so as to cause statistical variation between the clutter
and the mine like feature so that detection becomes
correctly segments the mine features and eliminates the
clutter/noise. A Detection Figure test was carried out and
the values obtained in all the cases are found to be more
than 40 thus indicating presence of an object.
After masking and extracting the mine features
in the region of interest, depth was estimated using the
model proposed. Fitness function formulated as difference
of
normalised
observed
backscatter
and
calculated/predicted backscatter was minimized using'

334

Subsurface landmine detection and estimation of

depth has been carried out successfully in X- band (10
GHz, 3 cm) upto a depth of 10 cm. Experimental data was
generated with different dummy mines without explosives
buried in dry sand at different depths in a laboratory
environment. Detection was carried out using Otsu's
thresholding. The results were validated with known
location of the mines used in the experiments. A new
model has also been developed and implemented for
estimation of depth of landmine which was optimized
using genetic algorithm. However, it was found that
optimisation using genetic algorithm was highly
dependent on choice of initial populations and it affected
the results significantly. The model does not have any
requirement of training and test data sets. A Detection
Figure test was carried out before processing the data for
depth analysis to assess the presence of any object in
thresholded image.

ACKNOWLEDGEMENT
The authors are thankful to the Defence
Research and Development Organisation, Ministry of
Defence, India for providing financial support for the

project.

[1].

REFERENCES
Bureau of Political & Military Affairs, "Hidden

Killers", US Department of StatePublicationlO575,

http:/www.state. gov/www/global/arms/rpt/-9809\-

demine\-loc.html, September 1998,

D Potin, P Vanheeghe, "An Abrupt Change
[2].
Detection Algorithm For Buried Landmine Localization",
IEEE Transactions on Geoscience And Remote Sensing,

Vol 44, No 2. Feb 2006

Carosi S and Cevini G, "An Electromagnetic
[3].
Approach Based On Neural Networks For the GPR
Investigation of Buried Cylinders", IEEE Geoscience and
RemoteSensing Letters, Vol 2, No 1, Jan 2005.
'

Mine / Depth

Anti Tank Mine

Depth 1.5 cm

Fragmentation
Mine, Depth 1.5

Influence Mine
Depth 1.5 cm

cm

Image
Dimensions

2a_
W1 7

Ln

-2

20j

Raw Plot

Calibrated
Plot

Convoluted
Plot
Otsu's

Thresholding

Fig-4 Mine detection

Collins L et al, " A Comparision of the

[4].
Performance of Statistical and Fuzzy Algorithms for
Unexploded Ordnance Detection", IEEE Transactions
on Fuzzy Systems Vol 9 No 1, Feb 2004.
Gader et al, " Landmine detection with ground
[5].
penetrating radar using hidden Markov models", IEEE_ ~~33
Transactions Geoscience and Remote Sensing", Vol 4 1,
No 4, Jun 2001.
Daniels, et al, Microwave Remote Sensing of
[6].
Physically Buried Objects in Negev Desert
Implications for Subsurface Martian Exploration,
Journal of Geophysical research, Vol. 108 No.48033,
2003

using threshholding

[7].
Ulaby F T, Moore R K and Fung A K, "Radar
Remote Sensing and Surface Scattering Emission
Theory, Vol II & III", Addison Wesley Publishing
Company, 1982.
Bermani B et al, " An Innovative Real Time
[8].
Technique for Buried Object Detection", IEEE
Transactions on Geoscience and Remote Sensing, Vol
41 No 4, April 2003
[9].
Tian H et al, "Implementing Otsu's
Thresholding Process Using Area-Time Efficient
Logarithmic Approximation Unit", 0-7803-776 13/03/$ 17.00 C 2003 IEEE, Mar 2003.

0-9

g~ ~0

Normalised Observed E Field

Predicted E Field

,,~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~.............................
Normalised

n3..

01

'i...
~~~~~~~~~~~~~~~~~~~~~~~~~~~...

0.3
ffi~ ~ ~~
~~~~~~~~~~~~~~~~~~~........ ...
01

0.01

0.02

003

004

0.05

006

.06

0.09

0.1

Figure - 5 Depth vs Predicted & Observed Backscatter

TABLE -I
Actual and Predicted Depth

Dielctric constant of dry Sand = 5

(From Ulabys graph) and Dielectric
______
Constant of Mine = 4
GA
Actual
% Error
Depth(H) Optimised
epth

0.005
0.01
0.015
0.02
0.025
0.03
0.04
0.05

0.07
0.1

0.00605
0.0096
10.01668
0.02361
-0.02583
0.0316
0.03438

111.2000
18.0500
3.3200
5.3333
-14.0500

0.04496

-10.0800

0.0523

0.13529

150

21.0000

- 3.6000

0
Actual Depth (cmn)

-25.2857
35.2900

TABLE
-I ~

mu10

150

Figure -6 Error plot - Actual vs predicted depth

~

33