Scribd
Upload
152 views27 pages

Ground Penetrating Radar

The document discusses landmine detection using ground penetrating radar (GPR). It outlines different GPR detection methods including background subtraction, matched filtering, wavelet packet decomposition, trimmed average power, and Kalman filtering. The Kalman filter approach provides the best performance according to receiver operating characteristic (ROC) comparisons, though it has the highest computational load. Overall, the document evaluates techniques for detecting landmines using GPR data and signals.

Uploaded by

Nag Bhushan
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
152 views27 pages

Ground Penetrating Radar

The document discusses landmine detection using ground penetrating radar (GPR). It outlines different GPR detection methods including background subtraction, matched filtering, wavelet packet decomposition, trimmed average power, and Kalman filtering. The Kalman filter approach provides the best performance according to receiver operating characteristic (ROC) comparisons, though it has the highest computational load. Overall, the document evaluates techniques for detecting landmines using GPR data and signals.

Uploaded by

Nag Bhushan
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 27

Landmine Detection

Using Impulse Ground


Penetrating Radar



A Presentation by Michael Muma
Advanced Topics in Statistical Signal Processing

outline


introduction
ground penetrating radar
the received data
detection methods
Kalman fliter
comparison of detection methods
conclusion



introduction - motivation



landmines kill
10.000 people annually
27 per day
1 during this presentation

landmines contaminate whole regions
farming impossible there
people must leave their homes
development assistance is difficult

further problems
metal content and size of landmines decrease
it costs 3 $ to produce a landmine vs. 1000 $ to remove it



introduction demining methods

conventional demining
techniques

mechanical devices
manual demining
dogs

introduction demining methods

advanced sensor techniques

advanced metall detection
nuclear quadruple resonance
thermographic imaging
electro-optical sensors
biosensing
chemical sensors
GPR
impulse ground penetrating radar
impulse ground penetrating radar

wideband radar
discrete pulses of ns-duration
centimeter resolution
digitizes the return at GHz sample rates

detects dielectric contrast in soils
also non-metallic landmines
the received data

general data processing problems

only sampled values of a continuous time function
limited time duration

special improvements of the data

zero offset removal
noise reduction
clutter reduction
time varying gain
frequency filtering
wavelet optimisation or deconvolution techniques
target resonances
migration

the received data






A-, B- and C-scan
the received data - A-scan

A-scan
returns a certain
position along the
distance axis


the received data - B-scan
B-scan
2-D matrix with
a scans as
vectors
t, x

the received data - C-scan
C-scan
3-D matrix
t, x, z

Unfocused C-scan of a set of six buried AT mine targets at
increments of 10mm(courtesy of ERA technology)
detection methods

Background subtraction
Matched Filter deconvolution
Wavelet Packet decomposition
Trimmed average power
Kalman filtering
background subtraction


Hypothesis Test
H : = 0 K : = 1
H is rejected if exceeds threshold T
Bs
Problems
The estimate for b(t) must be good.
Improvements
moving means or medians


( ) ( ) ( ) x t s t b t q = +
matched filter deconvolution
( ) ( ) ( ) x t v t h t = -
( ) ( ) ( ) v t m t t q o = +
( )

max ( ), ( ) ( )
t
deconv x t h t t o =


Hypothesis Test
H : = 0 K : > 0
H is rejected if exceeds threshold T
Mf
Problems
The estimate for h(t) must be good.
Improvements
moving means or medians


wavelet packet decomposition

background subtraction
transformation
wavelet packet decomposition
target concentration of energy in a few wavelet coefficients



trimmed average power

Average power is estimated
target change of average signal power from one trace to
another
Kalman filter

target free model
target present model
when to switch models
detection of a landmine
Kalman filter - target free model
represents one horizontal strip of the B-Scan
a set of p Kalman Filters is used



measurement vector


time evolution of the background vector









( , ) ( , ) ( , )
b
u n k s n k w n k = +
( ) ( ) ( )
b
p p p
u k s k w k = +
( ) ( 1) ( )
b b
p p p
s k s k v k = +
Kalman filter - target present model

represents one horizontal strip of the B-Scan
a set of p Kalman Filters is used



measurement vector


time evolution of the background and target vectors


( , ) ( , ) ( , ) ( , )
t b
u n k s n k s n k w n k = + +






( ) ( ) ( ) ( )
t b
p p p p
u k s k s k w k = + +
( ) ( 1)
b b
p p
s k s k = ( ) ( 1) ( )
t t
p p p
s k s k b k = + ( ) ( 1) ( )
p p p
b k b k v k = +
Kalman filter - switching models
test (Neyman-Pearson test)





For a fixed trace k, each strip p is tested.
If H is rejected for more than p
1
strips, H is rejected for the whole
trace.
target present model is initialized


: ( ) ( ) ( )
b
P p p p
H u k s k w k = +
: ( ) ( ) ( ) ( )
t b
p p p p p
K u k s k s k w k = + +

target free model
target prestent model
Kalman filter- detection of landmine

A target is declared if for at
least k
1
executive GPR
traces H is rejected
k
comparison of detection methods

the receiver operating characteristics (ROC)
computational load
comparison of detection methods


ROC
for different
landmines
and soil












The Kalman filter technique has best overall performance by a significant margin.
It is the only one that doesnt fail under any of the scenarios.

comparison of detection methods

Computational load








Kalman filter based approach is very computationally expensive
next most demanding technique, deconvolution, is approximately 20 times faster
background subtraction is least computationally complex

conclusion

rsum and outlook

ROCs show that the Kalman filter based detector provides the
best overall performance.
After the Kalman filter the trimmed average power achieves
good detection performance for light computational load.
An analyst should plot ROC curves for all techniques for the
particular environment they are working in before deciding on the
best algorithm to use.
A fused system using multiple techniques and sensors may,
ultimately, result.
Acknowledgments

[1] A.M. Zoubir Signal Processing Techniques for
Landmine Detection using Impulse GPR
[2] D.J. Daniels GPR- 2nd Edition
[3] Wickypedia http://en.wikipedia.org/wiki
[3] UCL Department of Computer Science
http://www.cs.ucl.ac.uk/staff/W.Langdon/roc
[4] Halotrust http://www.halotrust.org
[5] Chalmers University of Technonogy
http://www.s2.chalmers.se/research/signal/dbsearch.php?SFOREXP=autho
r=Lundberg%2BM&fullname=Magnus%20Lundberg
[6] Lara and Jo van Schalkwyk
http://www.anaesthetist.com/mnm/stats/roc
[7] Trotec-Bautrocknung http://trotec.de
[8] Michigan Technological University
http://www.admin.mtu.edu/urel/breaking/1999/suits.htm

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy