Buried Object Detection based on Acousto-seismic Method using

Accelerometer and Neural Network

Setyabudi

, Muhammad Rivai

and Ronny Mardiyanto

Department of Electrical Engineering, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia

Keywords:

Accelerometer, Acousto-Seismic Method, Buried Objects, Neural Network.

Abstract:

A system for detecting buried objects is often needed for inspection, exploration and security purposes. This

research has developed a system to detect buried objects based on the acousto-seismic principle. A sinusoidal

signal is ampliﬁed by an audio ampliﬁer to drive a subwoofer speaker to produce mechanical vibrations. The

seismic vibrations propagating in the ground are measured by an accelerometer. The Fast Fourier Transform

method converts vibrations in the time domain to the frequency domain. Neural Network algorithm is applied

to distinguish these wave spectrums to determine buried objects. After testing in experiments, this system can

distinguish between buried metal and non-metal objects. This system could also recognize the shallow buried

objects with an accuracy rate of 86.6%. This method can be potentially developed to detect land mines both

metal and non-metal materials.

1 INTRODUCTION

Buried object detections are often needed as an in-

spection and security machine. Metal detectors can

detect objects within a certain distance both on the

surface and inside the ground. Buried landmines with

metal casings are often found in ﬁelds which can be

dangerous to humans or animals around them. Metal

detectors are generally only able to detect metal ob-

jects. In fact, many dangerous objects are covered by

non-metal materials.

Several methods are applied to inspect buried ob-

jects. Radar-based noncontact displacement sensors

can be used to detect buried landmines using seis-

mic wave excitation (Martin et al., 2001). This sys-

tem explores the elasticity characteristics of objects

towards their environment. Detection of landmines

can be carried out by capturing gamma rays emit-

ted by hydrogen and nitrogen contained in the ex-

plosives (Yoshiyuki et al., 2007). The position of

the buried object can be localized by the pendu-

lum method to measure its acceleration (Maqsud and

Daku, 2005). An acousto-seismic method has also

been implemented in which buried objects have spe-

ciﬁc seismic (Rayleigh) wave responses in the time

and frequency domains (Bulletti et al., 2010). Detec-

tion of land mines based on Time Reverse Acoustics

can signiﬁcantly improve the accuracy of the results

(Sutin et al., 2005). The inspections of buried objects

typically use low frequency mechanical vibrations of

5-100 Hz (Song et al., 2017).

In this research, a system has been designed

and developed to detect buried objects based on the

acousto-seismic method. Seismic vibration is gen-

erated by an acoustic signal produced by a sub-

woofer. The vibration that propagate on the ground

is measured using micro-electro-mechanical systems

(MEMS) accelerometers. Fast Fourier Transform

(FFT) method converts vibrations in the time domain

to the frequency domain. Neural Network algorithm

will recognize the frequency spectrum pattern of the

vibration in order to determine the buried object.

252

Setyabudi, ., Rivai, M. and Mardiyanto, R.

Buried Object Detection based on Acousto-seismic Method using Accelerometer and Neural Network.

DOI: 10.5220/0009882202520257

In Proceedings of the 2nd International Conference on Applied Science, Engineering and Social Sciences (ICASESS 2019), pages 252-257

ISBN: 978-989-758-452-7

2 MATERIALS AND METHOD

The overall system of detection for buried objects is

illustrated by Figure 1. This prototype is a wooden

box with a size of 60cmx60cmx10cm containing 30

kg of sand. The 60 hertz sinusoidal signal is ampliﬁed

by an audio ampliﬁer to drive a 12-inch subwoofer

speaker to produce mechanical vibrations of 125 dB.

The subwoofer is located in a wooden box with a size

of 60cmx45cmx100cm mounted on sand with a dis-

tance of 30 cm, as shown in Figure 2. The objects

used in this study consisted of metal and non-metal

objects buried 10 cm deep in sand including iron, alu-

minum, zinc, stainless steel, plastic, polyvinyl chlo-

ride (PVC), and acrylic, as shown in Figure 3.

Figure 1: The overall system of detection for buried objects.

Figure 2: The layout of the buried object detection.

Figure 3: The buried objects used for the experiments.

space

Figure 4: The architecture of switching power ampliﬁer

(Kova

cevi

c et al., 2018)

Figure 5: The MDL-049 TPA3116D2 ampliﬁer module.

Class-D or switching power ampliﬁer has a high

power efﬁciency of more than 90% (Kova

cevi

c et al.,

2018). This type of ampliﬁer uses the Pulse Width

Modulation (PWM) method as depicted in Figure 4.

In this study, we use the MDL-049 TPA3116D2 sub-

woofer ampliﬁer module as shown in Figure 5. This

ampliﬁer is a stereo digital ampliﬁer that can drive

speakers by 50 W per channel. This module is con-

ﬁgured as a mono ampliﬁer that can produce a power

of 100 W to a 4 Ω speaker with the Bridge-Tied-Load

method.

Figure 6: The working principle of MEMS accelerometer

(John, 2011)

Figure 7: The neuron model.

Accelerometer is a sensor that can measure

changes in speed. MEMS accelerometer is often used

in many applications because of their compact size

for three-dimensional space. This sensor can be used

Buried Object Detection based on Acousto-seismic Method using Accelerometer and Neural Network

253

to measure vibrations that propagate on the ground or

objects (Ardiansyah et al., 2018). The working princi-

ple of this sensor is based on changes in capacitance

shown in Figure 6. The output voltage of the proof

mass can be expressed as:

(1)

where x is the distance between the two plates

of the internal capacitor, d is the change in distance

between the plates, V

is the amplitude of the exci-

tation voltage. By involving spring force, the out-

put voltage will be proportional to the acceleration.

In this study, the seismic vibrations are measured by

ADXL335 analog accelerometer. This sensor oper-

ates at the voltage of 1.8–3.6V and the current con-

sumption of 350µ. This device has a bandwidth of

around 0.5-1600 hertz. The sensor signal is then dig-

itized using an Arduino Nano microcontroller board.

The FFT method converts the signal in the time

domain into a frequency domain. This frequency

spectrum feature is often used for the identiﬁcation of

sounds produced by vibrating objects (Winjaya et al.,

2017). Compared with Discrete Fourier Transform

(DFT), the FFT has fewer total number of complex

multiplies of (N/2) 2log (N) with N is the number of

signal samples. In general, Fourier transform can be

expressed as:

x[k] =

n−1

∑

n=0

x[n]W

(2)

= e

− j2πkn/N

(3)

where x [n] is a discrete signal in time domain,

and x [k] is the frequency spectrum. In this study,

the range of frequency spectrum is between 0-99 hertz

with a step of 1 hertz. This spectrum is used as input

for the Neural Network to recognize the vibrations.

Figure 8: The Neural Network architecture used in the ex-

periment.

Neural Network is a computational model inspired

by human brain. The human brain has many neurons

that are connected to each other. Figure 7 shows the

model of a neuron cell. Using the logistic sigmoid

function, a neuron model can be expressed by:

= Σ

j,i

+ b

) (4)

1 + e

−az j

(5)

Neural Networks are widely used as artiﬁcial in-

telligence to identify patterns (Rivai and Tasripan,

2015), (Rivai et al., 2016), (Rivai et al., 2014). The

Neural Network architecture used in this study is

shown in Figure 8. The input layer consists of 100

nodes that correspond to the frequency spectrum of

the acousto-seismic vibrations. The hidden layer con-

sists of three layers, each of which consists of 500,

100, and 50 neurons. Meanwhile, the 2 neurons in the

output layer represent the number of classes that will

be recognized. The learning and running phases are

accomplished by a computer using Python Program-

ming Language.

3 RESULT AND DISCUSSIONS

The prototype of the system for detecting buried ob-

jects is shown in Figure 9. The experimental results

of the frequency response without buried object is

shown in Figure 10. Whereas, the frequency spec-

trum for buried metal and non-metal objects can be

seen in Figure 11, and Figure 12, respectively. Band-

width spectrum of buried objects is wider than that

of without objects. There are also signiﬁcant differ-

ences between the spectrum patterns of buried metal

and non-metal objects.

The next experiment is the detection of buried

metal and nonmetal objects using the Neural Net-

work. In the training phase, the network uses pairs

between 120 spectrum patterns of all metal objects for

different positions and targets (Y1 = 1 and Y2 = 0), as

well as pairs between 90 spectrum patterns of all non-

metal objects for different positions and targets (Y1 =

0 and Y2 = 1).

ICASESS 2019 - International Conference on Applied Science, Engineering and Social Science

254

space

Figure 9: The prototype of the buried object detection sys-

tem.

Figure 10: The spectrum of vibration without buried object.

Figure 11: The spectrum of vibration with buried metal ob-

jects.

This training phase loss curve is shown in Figure

13. In the running phase, 35 spectrum patterns are

tested online as shown in Table 1. The experimen-

tal results indicate that Neural Network could classify

buried metal and non-metal objects with an identiﬁ-

cation level of 77%.

space

Figure 12: The spectrum of vibration with buried non-metal

objects.

The next experiment is the detection of with and

without buried objects. In the training phase, the net-

work uses pairs between 100 spectrum patterns of

without objects and targets (Y1 = 1 and Y2 = 0), as

well as pairs between 210 spectrum patterns of all

both metal and non-metal objects for different posi-

tions and targets (Y1 = 0 and Y2 = 1). This training

phase loss curve is shown in Figure 14. In the run-

ning phase, 15 spectrum patterns are tested online as

shown in Table 2. The experimental results indicate

that Neural Network could determine the buried ob-

jects with a success rate of 86.6%. The similar results

are obtained when this system is applied to identify

buried objects for true ﬁeld experiments, as shown in

Figure 15.

Figure 13: Loss curve of the Neural Network training phase

to detect buried metal and non-metal objects.

Buried Object Detection based on Acousto-seismic Method using Accelerometer and Neural Network

255

spacer

Figure 14: Neural Network identiﬁcation for buried metal

and non-metal objects.

Figure 15: Loss curve of the Neural Network training phase

to detect buried objects.

space

Figure 16: Neural Network identiﬁcation for buried objects.

Figure 17: True ﬁeld experiments for buried object identiﬁ-

cation

4 CONCLUSION

We have designed and realized a device to detect

buried objects based on the acousto-seismic method.

The 60 hertz sinusoidal signal is ampliﬁed by an audio

ampliﬁer to drive a 12-inch subwoofer speaker to pro-

duce mechanical vibrations of 125 dB. The seismic

vibrations propagating in the ground are measured

by the ADXL335 MEMS accelerometer. The FFT

method converts vibrations in the time domain to the

frequency domain. The Neural Network algorithm is

applied to distinguish these wave spectrums to deter-

mine buried objects. The experimental results show

that this system can distinguish between buried metal

and non-metal objects with an identiﬁcation rate of

77%. This system could also recognize the shallow

buried objects with a success rate of 86.6%

ICASESS 2019 - International Conference on Applied Science, Engineering and Social Science

256

REFERENCES

Ardiansyah, H., Rivai, M., and Nurabdi, L. (2018). Train

arrival warning system at railroad crossing using ac-

celerometer sensor and neural network. In AIP Con-

ference Proceedings, page 040029–1.

Bulletti, A., Borgioli, G., Calzolai, M., Capineri, L.,

and Mazzoni, M. (2010). Acoustoseismic method

for buried-object detection by means of surface-

acceleration measurements and audio facilities. IEEE

Trans. Geosci. Remote Sens, 48(8):3134.

John (2011). Mems accelerometer. http:

//www.instrumentationtoday.com/

mems-accelerometer/2011/08. [Online; Ac-

cessed 23-June-2019].

Kova

cevi

c, S., Pe

c-Brdjanin, T., and Gali

c, J. (2018).

Class d audio ampliﬁer with reduced distortion. Int.

Symp. Ind. Electron. INDEL.

Maqsud, A. and Daku, B. (2005). Characterization of a

mems accelerometer for an underground mine posi-

tioning system. In Canadian Conference on Electrical

and Computer Engineering, page 2268.

Martin, J., Scott, W., and Larson, G. (2001). Experimen-

tal model for a seismic landmine detection system.

IEEE Transactions on Geoscience and Remote Sens-

ing, 39(6):1155.

Rivai, M., Ariﬁn, A., and Agustin, E. (2016). Mixed vapour

identiﬁcation using partition column-qcms and artiﬁ-

cial neural network. In International Conference on

Information, Communication Technology and System,

page 172.

Rivai, M., Talakua, E., and L. (2014). The implementation

of preconcentrator in electronic nose system to iden-

tify low concentration of vapors using neural network

method. International Conference on Information,

Communication Technology and System, page 31.

Rivai, M. and Tasripan (2015). Fuel qualiﬁcation using

quartz sensors. ARPN Journal of Engineering and Ap-

plied Sciences, 10(16):6737.

Song, K., Tong, S., Ding, Z., and Dong, L. (2017). An

electromagnetic feedback method to improve low-

frequency response performance of geophone. In

Proc. IEEE Sensors, page 5.

Sutin, A., Johnson, P., Tencate, J., and Sarvazyan, A.

(2005). Land mine detection by time reversal acousto-

seismic method time reversal acousto-seismic method

for land mine detection. In Proc. of Society of Photo-

Optical Instrumentation Engineers, page 706.

Winjaya, F., Rivai, M., and Purwanto, D. (2017). Identiﬁ-

cation of cracking sound during coffee roasting using

neural network. International Seminar on Intelligent

Technology and Its Application, page 271.

Yoshiyuki, T., Tsuyushi, M., Cheol, H., Seiji, S., Kiyosi,

Y., Kai, M., and Teruhisa, T. (2007). Development

of landmine detection system using scintillators by

measuring radiations from landmine. IEEE Nucl. Sci.

Symp. Conf. Rec, page 273.

Buried Object Detection based on Acousto-seismic Method using Accelerometer and Neural Network

257