A Comparison of Detection Strategies for Solids and Organic

8
10/10/11 1 A Comparison of Detection Strategies for Solids and Organic Liquids on Surfaces Using Long Wave Infrared Hyperspectral Imaging. Neal B. Gallagher 1 , Thomas A. Blake 2 , James F. Kelly 2 1 Eigenvector Research, Inc., [email protected] 2 Pacific Northwest National Laboratory, Richland, WA USA A Comparison of Detec1on Strategies for Solids and Organic Liquids on Surfaces Using Long Wave Infrared Hyperspectral Imaging. Neal B. Gallagher 1 , Thomas A. Blake 2 , James F. Kelly 2 1 Eigenvector Research, Inc., 160 Gobblers Knob Lane, Manson, WA 98831, USA [email protected] 2 Pacific Northwest NaJonal Laboratory, P.O. Box 999, Mail Stop K888, Richland, WA 99352, USA Passive standoff detecJon of condensed chemical analytes on surfaces is of interest for rapid screening in security, environmental and precision agriculture applicaJons. This paper compares and contrasts the detecJon of solid explosive residues and thin films of liquid organophosphorus compounds on a steel plate using hyperspectral imaging in the long wave infrared at standoff distances of tens of meters. The objecJve is to idenJfy the appropriate modeling approach for detecJon and potenJal classificaJon of both types of compounds in field applicaJons. Of parJcular interest is the use of the matched filter (generalized least squares) esJmator under the assumpJon of “parJal closure” on the esJmated contribuJons. IniJal results indicate that parJal closure is appropriate for detecJon of solids on surfaces but may be less useful for detecJon of thin liquid films.

Transcript of A Comparison of Detection Strategies for Solids and Organic

Page 1: A Comparison of Detection Strategies for Solids and Organic

10/10/11  

1  

A Comparison of Detection Strategies for Solids and Organic Liquids on

Surfaces Using Long Wave Infrared Hyperspectral Imaging.

Neal B. Gallagher1, Thomas A. Blake2, James F. Kelly2

1 Eigenvector Research, Inc., [email protected] 2 Pacific Northwest National Laboratory, Richland, WA

USA

A  Comparison  of  Detec1on  Strategies  for  Solids  and  Organic  Liquids  on  Surfaces  Using  Long  Wave  Infrared  Hyperspectral  Imaging.      Neal  B.  Gallagher1,  Thomas  A.  Blake2,  James  F.  Kelly2      1  Eigenvector  Research,  Inc.,  160  Gobblers  Knob  Lane,  Manson,  WA  98831,  USA  [email protected]  2  Pacific  Northwest  NaJonal  Laboratory,  P.O.  Box  999,  Mail  Stop  K8-­‐88,  Richland,  WA  99352,  USA      Passive  standoff  detecJon  of  condensed  chemical  analytes  on  surfaces  is  of  interest  for  rapid  screening  in  security,  environmental  and  precision  agriculture  applicaJons.  This  paper  compares  and  contrasts  the  detecJon  of  solid  explosive  residues  and  thin  films  of  liquid  organophosphorus  compounds  on  a  steel  plate  using  hyperspectral  imaging  in  the  long  wave  infrared  at  standoff  distances  of  tens  of  meters.  The  objecJve  is  to  idenJfy  the  appropriate  modeling  approach  for  detecJon  and  potenJal  classificaJon  of  both  types  of  compounds  in  field  applicaJons.  Of  parJcular  interest  is  the  use  of  the  matched  filter  (generalized  least  squares)  esJmator  under  the  assumpJon  of  “parJal  closure”  on  the  esJmated  contribuJons.  IniJal  results  indicate  that  parJal  closure  is  appropriate  for  detecJon  of  solids  on  surfaces  but  may  be  less  useful  for  detecJon  of  thin  liquid  films.  

Page 2: A Comparison of Detection Strategies for Solids and Organic

10/10/11  

2  

Standoff Detection •  Passive standoff detection of condensed chemicals on surfaces

•  environmental cleanup, rapid screening for security and precision ag •  Traditional lab calibration procedures are not applicable

•  measured image and analyte spectrum available •  clutter (interferences) highly variable image to image

Experiment and Analysis

•  1 m2 bare metal plates with •  solid explosive residues RDX and Tetryl •  thin layer of organophosphorous DBP and DBBP • 14, 31 and 50 m at known areal dosage

• Telops, Inc. longwave infrared imager • End-member extraction (exploratory analysis) and • Two generalized least squares (GLS) target detection

approaches • a.k.a., matched filter

Page 3: A Comparison of Detection Strategies for Solids and Organic

10/10/11  

3  

Detection Models

x = xc + cs

The simple additive model assumes that the background is static with Gaussian noise and adding target increases the signal.

The linear mixture model assumes that the background signal is not static and can change as target is added.

x = ccxc + cs x

xc

cs s

x1

x2

x ccxc

cs s

x1

x2

x

ccxc

cs s

x1

x2

x ! xc

If target ‘blocks’ the background, the simple additive model can ‘subtract too much’ of the background signal and there is ‘loss of target information.’ (the thin magenta signal has shorter projection on s).

x

ccxc

cs s

x1

x2

x ! xc

… but how does the signal truly manifest?

Page 4: A Comparison of Detection Strategies for Solids and Organic

10/10/11  

4  

08-Jul-2008 3:51 pm S6_TETRYL_8cm-1_14m_1 _000011476 110 µg/cm2

08-Jul-2008 4:43 pm S6_TETRYL_8cm-1_31m_PASSIVE_1 _000011950 110 µg/cm2

Steel Plate S6

Tetryl

Steel Plate S6

9001000110012000

0.05

0.1

0.15

0.2

0.25

Wavenumber (1/cm)

on steel at 14m  

on steel at 31m  

Example: Tetryl on Steel. At 14 m the signal looks like reflected sky shine. At 31 m the signal looks like reflected black body.

90095010001050110011501200125013000

0.05

0.1

0.15

0.2

0.25

Wavenumber (1/cm)

Nor

mal

ized

Spe

ctru

m

TETTET MCR

Comparison of the TET reflectance spectrum and the end-member estimate of the target’s spectrum

0.2

0.4

0.6

0.8

Stee

l

row 53row 58

0 20 40 60 80 100 1200

0.2

0.4

0.6

x pixel

TET

row 53row 58

S6

20 40 60 80 100 120

10

20

30

40

50

60 0

2

4

6

8

x 10 3

S6

20 40 60 80 100 120

10

20

30

40

50

60 0

0.5

1

S6

20 40 60 80 100 120

10

20

30

40

50

60 0

0.2

0.4

0.6

Factor 1: Target

Factor 2: Steel

RMSE

The contributions shows that signal due to the steel plate decreases when target is present.

08-Jul-2008 3:51 pm S6_TETRYL_8cm-1_14m_1 _000011476 110 µg/cm2

Page 5: A Comparison of Detection Strategies for Solids and Organic

10/10/11  

5  

90095010001050110011501200125013000

0.05

0.1

0.15

0.2

0.25

0.3

Wavenumber (1/cm)

Nor

mal

ized

Spe

ctru

m

RDXRDX MCR

PLATE_S9_RDX_4cm-­‐1_14m_1_000011692  90  μg/cm2  Comparison of the RDX reflectance spectrum and the end-member estimate of the target’s spectrum

0

0.1

0.2

Load

ings RDX

Steel

2

90010001100120013000

1

2

RM

SE (x

1000

)

Wavenumber (1/cm)

0.3

0.32

0.34

Stee

l

row 51row 64

0 50 100 1500

0.1

0.2

x pixel

RD

X

row 51row 64

S9

50 100 150

20

40

60

80

100

1200

0.05

0.1

0.15

S9

50 100 150

20

40

60

80

100

1200

0.01

0.02

0.03

0.04

0.05

S9

50 100 150

20

40

60

80

100

120

0.05

0.1

0.15

0.2

0.25

0.3S9

50 100 150

20

40

60

80

100

1200

0.5

1

1.5

2

2.5

x 10 3

Factor 1: Target

Factor 2 (interference?)

Factor 3: Steel

RDX End-Member Contributions The contributions suggest that signal due to the steel plate increases slightly when target is present.

RMSE (Highest residuals are at highest target contributions.)

Page 6: A Comparison of Detection Strategies for Solids and Organic

10/10/11  

6  

Factor  1  referenced  to  99%  CL  for  T-­‐stat  GLS  2    >  GLS  1  

Factor  2  referenced  to  99%  CL  for  T-­‐stat  decrease  in  bkg  

Q  referenced  to  99%  CL  GLS  2  <  GLS  1  

GLS  2  

S6

20 40 60 80 100 120

10

20

30

40

50

60 0

5

10

15

20 S6

20 40 60 80 100 120

10

20

30

40

50

60 0

5

10

15

20

S6

20 40 60 80 100 120

10

20

30

40

50

60

20

10

0

S6

20 40 60 80 100 120

10

20

30

40

50

60

10

20

30

40

50 S6

20 40 60 80 100 120

10

20

30

40

50

60

10

20

30

40

50

GLS  1   GLS  2  

Target detection results for TET: GLS 2 performs better than GLS 1

GLS  1   GLS  2  

S9

50 100 150

20

40

60

80

100

120

10

20

30

40

50

60

S9

50 100 150

20

40

60

80

100

1200

2

4

6

8

S9

50 100 150

20

40

60

80

100

120

5

10

15

20

25

S9

50 100 150

20

40

60

80

100

1200

2

4

6

8

S9

50 100 150

20

40

60

80

100

120

10

20

30

40

50

60

Target detection results for RDX: GLS 1 performs ~better than GLS 2

Factor  1  referenced  to  99%  CL  for  T-­‐stat  GLS  2    <  GLS  1  

Factor  2  referenced  to  99%  CL  for  T-­‐stat  ~increase  in  bkg  

Q  referenced  to  99%  CL  GLS  2  ~  GLS  1  

GLS  1   GLS  2  

Page 7: A Comparison of Detection Strategies for Solids and Organic

10/10/11  

7  

MCR GLS Image effect on bkg contributions dist dosage when target target bkg resids chem (m) (μg/cm2) present 2?1 (2 only) (2?1) RDX 14 40 ~> < > ~ RDX 14 90 ~> < ~> ~ RDX 31 40 ~> < > ~ TET 14 110 < > < < TET 31 110 < > < < DBP 14 90 ~> < ~> ~ DBP 30 90 ~> > < < DBBP 14 90 - ~ ~< ~

Summary Table

Conclusions

• When GLS 2 works better (when target blocks background) it works significantly better than GLS 1. • e.g., TET

• When GLS 2 works worse (when target has little effect on background) it works slightly worse. • e.g., RDX. • RDX end-member showed shift of ~ 1000 1/cm peak.

• GLS 2 added information via the background factor that may lead to enhanced detection capability.

Page 8: A Comparison of Detection Strategies for Solids and Organic

10/10/11  

8  

Acknowledgements

• Funding by the Pacific Northwest National Laboratory • Laboratory Directed Research and Development program,

Initiative for Explosives Detection. • Thanks to David Atkinson, Greg Holter, Robert Ewing

and John Hartman at PNNL. • Thanks to Vincent Farley and Jean-Pierre Allard of

Telops, Inc.

20

Blake, T.A., Kelly, J.F., Gallagher, N.B, Gassman, P.L., Johnson, T.J., “Passive Detection of Solid Explosives in Mid-IR Hyperspectral Images” Anal. Bioanal. Chem. (2009). DOI 10.1007/s00216-009-2907-5.

Gallagher, N.B., Kelly, J.F., Blake, T.A., "Passive Infrared Hyperspectral Imaging for Standoff Detection of Tetryl Explosive Residue on a Steel Surface," proceedings WHISPERS'10, Reykjavik, Iceland June 14-16 (2010), DOI: 10.1109/WHISPERS.2010.5594839.