Elimination of Clutter through Signal Processing for Landmine / Ordnance Detection (ECSPLOD)

Gabriel Ford

Aniket Hirebet

Vasileios T. Nasis

Norman Butler

Advisor:

• Dr. Athina Petropulu

May 29, 2002

Nic Dunlop

What are landmines?

Buried explosive devices.

Triggered by pressure or tripwire.

Types

• Anti-personnel

• Anti-tank

Cost 3 dollars to purchase.

Cost 1000 dollars to remove. Nic Dunlop

The Human Toll

Over 110 million mines are scattered across the world.

500 people are killed or injured every week by triggering landmines.

80% of the victims are civilians.

Nic Dunlop

Why Metal Detectors Fail

Only effective when locating mines with substantial metal content.

Modern anti-personnel mines are made of plastics and contain dielectric explosives.

National Defense Magazine

Ground Penetrating Radar (GPR)

Ground Penetrating Radar (GPR) can detect all types of landmines, including modern plastic mines.

Our project focuses on preprocessing to improve landmine detection.

Preprocessing Detection ClassificationPreprocessing

DeTec - EPFLDeTec - EPFL

Experimental Setup

Ground Penetrating Radar (GPR) data was collected by DeTeC with a commercial SPRscan system.

Mines were buried in the center of a scanning pattern.

Each line in the scanning pattern represents a separate B-scan.

DeTec - EPFL

GPR Data: A-scans and B-scans

A B-scan is a collection of A-scans recorded along a scanning line.

The vertical (time) axis corresponds to depth.

A B-scan represents a vertical slice of the ground.

B-scan – Composed of A-scans

The received signal at any measurement point is called an A-scan.

A-scan

The Problem: Signal Clutter

GPR images are marred by a high degree of clutter.

For homogeneous media, the clutter varies little over a B-scan.

As a result, the mine signal is still evident.

PFM-1 Mine in SAND

PFM-1 Mine in SOIL

For non-homogeneous media, the clutter is more random.

In this case, the mine signal is highly obscured. ?

Signal Clutter Defined

A-scan Model:

w = c + b + s + e

The received signal w:c: antenna cross-talk

b: ground bounce

s: target signal

e: measurement noise

Clutter is due primarily to ground bounce and antenna cross-talk.

e

sb

c

Tx Rx

w

c

b

Clutter Reduction

Common clutter reduction methods include:• subtracting a signal measured in the absence of a

target• subtracting a moving average estimate of the

background • ensemble mean subtraction

These procedures are ineffective for the general case of mines buried in non-homogeneous media.

We have developed ECSPLOD to specifically overcome this problem

Method of Solution

ECSPLOD extends mean subtraction with adaptive filtering for interference cancellation.

ECSPLOD updates the adaptive filter input signal to compensate for variation in clutter properties.

x = n’ d = s+nx = n’ d = s+nx = n’ d = s+nx = n’ d = s+nx = n’ d = s+nx = n’ d = s+nx = n’ d = s+n

d = s+n

x = n’ Adaptive Filter SUM

y

-

+e

Mine Response

Raw Data

Mean Subtraction

Plastic Mine in Sand – Mean Subtraction

Raw Data

Mean Subtraction

Plastic Mine in Soil – Mean Subtraction

Plastic Mine in Soil: Moving Average Subtraction vs. ECSPLOD

MA

ECSPLOD-KALMAN

ECSPLOD-NLMS

Summary – ECSPLOD

Raw Data

Mean Subtraction

ECSPLOD - NLMS

Plastic Mine in Soil – Detection

Raw Data Moving Average Subtraction

ECSPLOD NLMS ECSPLOD Kalman

Detection Evaluation – ROC Curves

PFM - 1 in Soil, NLMS Type 72 in Soil, NLMS

Economic Summary

• Out of pocket costs were under $500 (reference material and printing costs)

• Cost in private sector summarized below:

Salaries $ 80,400.00

Fringe Benefits $ 11,656.80

Consulting Services $ 11,440.00

Hardware / Software / Equipment $ 7,608.00

Subtotal: $111,104.80

Overhead @ 100 % $111,104.80

TOTAL $222,209.60

Summary and Conclusions

The Problem:

• There is a need for a reliable demining technology

• Metal detectors do not work on all types of mines

• Ground Penetrating Data is suitable for all types of mines

• But GPR is plagued with the problem of excessive signal clutter

Summary and Conclusions

The Solution:

• Mean Subtraction only works in homogeneous media

• Moving Average Subtraction shows some improvement

• ECSPLOD utilizes Mean Subtraction followed by Adaptive Filtering

• ROCs and visual comparisons show that ECSPLOD works

Questions and Comments

Before

After

ECSPLOD

Special thanks to:

Athina PetropuluWayne HillMaja BystromMoshe KamBrian Kravitz