![Page 1: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/1.jpg)
Prac%cal experiences from muni%ons response demonstra%ons
Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran, Nicolas Lhomme
1
![Page 2: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/2.jpg)
2008 Camp Sibert, AL 2009 San Luis Obispo, CA 2010 Camp Butner, NC
2011 Camp Beale, CA
• Identify single large target among smaller clutter and debris
• Increased TOI target classes • Significant topographic relief
2011 Pole Mountain, WY
• Significant amount of clutter similar in size & shape to 37mm
• Data support an aggressive strategy • Data requires conservative approach
Evolu%on of processing capabili%es driven by live site challenges
2
![Page 3: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/3.jpg)
Standard processing flow for UXO detec%on and classifica%on
1. Data Acquisi%on 2. Feature es%ma%on 3. Classifica%on
Feature vector
Tx, Rx Tx, Ry Tx, Rz
Ty, Rx Ty, Ry Ty, Rz
Tz, Rx Tz, Ry Tz, Rz
3
![Page 4: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/4.jpg)
UXO are generally dis2nguished by:
• large amplitude, slow-‐decaying primary (L1) polarizability
• equal secondary polarizabili2es (L2=L3).
L1 L2=L3
UXO
L1
L2≠L3
Non-‐UXO
Time Time
2. Feature Extrac%on : Target Polarizabili%es
4
![Page 5: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/5.jpg)
Size/decay Total polarizability All polarizabilities
• Size • Wall-thickness • Shape
• Size • Wall-thickness • Simplest representation
• Choice of features depends on the classification problem and data quality
Time
3. Classifica%on: Feature Selec%on
Time
5
![Page 6: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/6.jpg)
UXO Classifica%on at Pole Mountain, WY
• ESTCP Classification Study
• Small ISO • 37 mm projectiles • 57 mm projectiles • 60 mm mortar • 75 mm projectile • Horseshoes
MetalMapper data were acquired and processed for distribution in July-August 2011
6
![Page 7: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/7.jpg)
Feature Extraction
1. Estimate target parameters • Single source and two source inversions
• Loca2on, orienta2on, polarizabili2es
2. Data /Inversion QC • Look for poor fits to the data
• Determine if any anomalies are “Can’t Analyze”
3. Model Selection • Determine which of the models should be used in the
classifier
7
![Page 8: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/8.jpg)
Data/Inversion QC and Model Selec%on
• QC soXware tools essen2al for efficient iden2fica2on of data and inversion issues
8
![Page 9: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/9.jpg)
Classification
4. Request training data • Semi-‐supervised approach
5. Create Ranked Anomaly List • Library Matching Method
• Support Vector Machine Classifier
6. Determine a Stop Digging Point • Manual: Visual inspec2on of list
9
![Page 10: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/10.jpg)
Classification 4. Request training data
• Establish clusters of UXO • Determine extent of UXO clusters and boundaries with clu`er classes.
SMALL ISO 10
![Page 11: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/11.jpg)
Classification
6 cm Fuse
4. Request training data • Establish clusters of UXO • Determine extent of UXO clusters and boundaries with clu`er classes.
11
![Page 12: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/12.jpg)
Classifica2on Example:
MetalMapper data processing at Pole Mountain “Aggressive” dig lists: included polarizability quality and misfits to L1, L2 and L3
Final ROC curves for Year 1 and Year 2. Both lists found all TOI before the stop dig point.
All TOI iden2fied 88% reduc2on in digs
All TOI iden2fied 84% reduc2on in digs
12
![Page 13: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/13.jpg)
Beale Open Area: MetalMapper Processing
• MetalMapper data collected by Parsons and CH2M HILL
• Two classifica2on methods to each dataset:
1. Library based 2. Two-‐stage conserva2ve approach
• A different analyst for each method and instrument
13
![Page 14: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/14.jpg)
Understanding limita%ons of the data • Varia2on in quality of recovered polarizabili2es for ISO
ISO: MetalMapper (P) Camp Beale 2011
ISO: MetalMapper (C) Camp Beale 2011
ISO: MetalMapper Pole Mountain 2011
ISO: MetalMapper (URS) Spencer 2012
ISO: MetalMapper (NAEVA) Spencer 2012
14
![Page 15: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/15.jpg)
1105 Scrap Not Dug
334 Total Digs
205 129
MetalMapper Processing 1: Library Matching
• Aggressive method
• Two ISO Missed
Two ISO Missed 2 ISO Not Dug MetalMapper
15
![Page 16: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/16.jpg)
• Two ISO missed
BE-‐2532 ISO stats seed
BE-‐1965 ISO stats seed
• All 3 polarizabilities sometimes not well constrained – should also use primary polarizability
Classifica2on Example:
MetalMapper data processing at Camp Beale
16
![Page 17: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/17.jpg)
Classifica2on Example:
MetalMapper data processing at Camp Beale
869 Scrap Not Dug
572 Total Digs
441 Scrap Dug
131 UXO Dug
• More conservative approach
• Use all 3 polarizabilities then switch to total polarizability
NO UXO Missed
17
![Page 18: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/18.jpg)
Technology Transfer: Industry Partner Results Pole Mountain Spencer Range
NO UXO Missed NO UXO Missed
18
![Page 19: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/19.jpg)
Reduc%on in Excava%ons and Cost Reduc%on
Classifica%on
Classifica%on
MetalMapper Digging
Detec2on Survey
19
![Page 20: Praccalexperiencesfrommunions response’demonstra%ons · Praccalexperiencesfrommunions response’demonstra%ons Kevin Kingdon, Len Pasion, Stephen Billings, Barry Zelt, Laurens Beran,](https://reader033.vdocuments.site/reader033/viewer/2022042011/5e724880fd466733672cc10f/html5/thumbnails/20.jpg)
Conclusions • There are a number key components to successful classification
1. A reliable data Quality Control (QC) process 2. Understanding limitations of the data 3. Selecting an appropriate classification strategy based on the
data quality
• Effective classification could be achieved at Pole Mountain and Camp Beale demonstration sites using MetalMapper data
Acknowledgements The work in this presentation was funded by the Strategic Environmental Research and Development Program (SERDP) and the Environmental Security Technology Certification Program (ESTCP)
The University of British Columbia Geophysical Inversion Facility
20