Abstract

We treat infrared patterns of absorption or emission by nerve and blister agent compounds (and simulants of this chemical group) as features for the training of neural networks to detect the compounds’ liquid layers on the ground or their vapor plumes during evaporation by external heating. Training of a four-layer network architecture is composed of a backward-error-propagation algorithm and a gradient-descent paradigm. We conduct testing by feed-forwarding preprocessed spectra through the network in a scaled format consistent with the structure of the training-data-set representation. The best-performance weight matrix (spectral filter) evolved from final network training and testing with software simulation trials is electronically transferred to a set of eight artificial intelligence integrated circuits (ICs’) in specific modular form (splitting of weight matrices). This form makes full use of all input–output IC nodes. This neural network computer serves an important real-time detection function when it is integrated into pre- and postprocessing data-handling units of a tactical prototype thermoluminescence sensor now under development at the Edgewood Research, Development, and Engineering Center.

© 1995 Optical Society of America

Full Article  |  PDF Article

References

  • View by:
  • |
  • |
  • |

  1. A. H. Carrieri, “Infrared detection of liquids on terrestrial surfaces by CO2 laser heating,” Appl. Opt. 29, 4907–4913 (1990).
    [CrossRef] [PubMed]
  2. M. E. Munk, M. S. Madison, E. W. Robb, “Neural network models for infrared spectrum interpretation,” Mikrochim. Acta 2, 505–514 (1991).
    [CrossRef]
  3. A. Bruchmann, H. J. Götze, P. Zinn, “Application of Hamming networks for IR spectral search,” Chemometrics Intell. Lab. Syst. 18, 59–69 (1993).
    [CrossRef]
  4. J. R. M. Smits, P. Schoenmakers, A. Stehmann, F. Sijstermans, G. Kateman, “Interpretation of infrared spectra with modular neural-network systems,” Chemometrics Intell. Lab. Syst. 18, 27–39 (1993).
    [CrossRef]
  5. M. Meyer, T. Weigelt, “Interpretation of infrared spectra by artificial neural networks,” Anal. Chim. Acta 265, 183–190 (1992).
    [CrossRef]
  6. R. J. Fessenden, L. Györgyi, “Identifying functional groups in IR spectra using an artificial neural network,” J. Chem. Soc. Perkin Trans. 2, 1755–1762 (1991).
  7. E. W. Robb, M. E. Munk, “A neural network approach to infrared spectrum interpretation,” Mikrochim. Acta 1, 131–155 (1990).
    [CrossRef]
  8. H. F. Hameka, J. O. Jensen, “Theoretical prediction of the infrared and Raman spectra of O-ethyl S-2diisopropyl amino ethyl methyl phosphonothiolate,” Int. J. Quantum Chem. 50, 161–172 (1994).
    [CrossRef]
  9. M. Rrisch, J. Foresman, A. Frisch, Gaussian 92 User’s Guide (Gaussian, Inc., Pittsburgh, Pa., 1992).
  10. R. Piffath, U.S. Army Chemical and Biological Defense Command, Edgewood Research, Development, and Engineering Center, Aberdeen Proving Ground, Md. 21010-5423 (personal communication, July1993).
  11. H. F. Hameka, A. H. Carrieri, J. O. Jensen, “Calculations of the structure and the vibrational infrared frequencies of some methylphosphonates,” Phosphorus Sulfur Silicon 66, 1–11 (1992).
    [CrossRef]
  12. S. Zarrabian, “The use of first-derivative spectra in obtaining quantitative and qualitative information,” Spectroscopy 8, 43–45 (1993).
  13. D. L. Reilly, L. N. Cooper, C. Elbaum, “A neural model for category learning,” Biol. Cybern. 45, 35–41 (1982).
    [CrossRef] [PubMed]
  14. Original suggestions were made and a method of solution was developed by J. O. Jensen, U.S. Army Edgewood Research, Development, and Engineering Center, Aberdeen Proving Ground, Md. 21010-5423 (personal communication, 1994).
  15. B. A. Hess, L. J. Schaad, “Ab initio calculations of vibrational spectra and their use in the identification of unusual molecules,” Chem. Rev. 86, 709–730 (1986).
    [CrossRef]

1994

H. F. Hameka, J. O. Jensen, “Theoretical prediction of the infrared and Raman spectra of O-ethyl S-2diisopropyl amino ethyl methyl phosphonothiolate,” Int. J. Quantum Chem. 50, 161–172 (1994).
[CrossRef]

1993

S. Zarrabian, “The use of first-derivative spectra in obtaining quantitative and qualitative information,” Spectroscopy 8, 43–45 (1993).

A. Bruchmann, H. J. Götze, P. Zinn, “Application of Hamming networks for IR spectral search,” Chemometrics Intell. Lab. Syst. 18, 59–69 (1993).
[CrossRef]

J. R. M. Smits, P. Schoenmakers, A. Stehmann, F. Sijstermans, G. Kateman, “Interpretation of infrared spectra with modular neural-network systems,” Chemometrics Intell. Lab. Syst. 18, 27–39 (1993).
[CrossRef]

1992

M. Meyer, T. Weigelt, “Interpretation of infrared spectra by artificial neural networks,” Anal. Chim. Acta 265, 183–190 (1992).
[CrossRef]

H. F. Hameka, A. H. Carrieri, J. O. Jensen, “Calculations of the structure and the vibrational infrared frequencies of some methylphosphonates,” Phosphorus Sulfur Silicon 66, 1–11 (1992).
[CrossRef]

1991

R. J. Fessenden, L. Györgyi, “Identifying functional groups in IR spectra using an artificial neural network,” J. Chem. Soc. Perkin Trans. 2, 1755–1762 (1991).

M. E. Munk, M. S. Madison, E. W. Robb, “Neural network models for infrared spectrum interpretation,” Mikrochim. Acta 2, 505–514 (1991).
[CrossRef]

1990

A. H. Carrieri, “Infrared detection of liquids on terrestrial surfaces by CO2 laser heating,” Appl. Opt. 29, 4907–4913 (1990).
[CrossRef] [PubMed]

E. W. Robb, M. E. Munk, “A neural network approach to infrared spectrum interpretation,” Mikrochim. Acta 1, 131–155 (1990).
[CrossRef]

1986

B. A. Hess, L. J. Schaad, “Ab initio calculations of vibrational spectra and their use in the identification of unusual molecules,” Chem. Rev. 86, 709–730 (1986).
[CrossRef]

1982

D. L. Reilly, L. N. Cooper, C. Elbaum, “A neural model for category learning,” Biol. Cybern. 45, 35–41 (1982).
[CrossRef] [PubMed]

Bruchmann, A.

A. Bruchmann, H. J. Götze, P. Zinn, “Application of Hamming networks for IR spectral search,” Chemometrics Intell. Lab. Syst. 18, 59–69 (1993).
[CrossRef]

Carrieri, A. H.

H. F. Hameka, A. H. Carrieri, J. O. Jensen, “Calculations of the structure and the vibrational infrared frequencies of some methylphosphonates,” Phosphorus Sulfur Silicon 66, 1–11 (1992).
[CrossRef]

A. H. Carrieri, “Infrared detection of liquids on terrestrial surfaces by CO2 laser heating,” Appl. Opt. 29, 4907–4913 (1990).
[CrossRef] [PubMed]

Cooper, L. N.

D. L. Reilly, L. N. Cooper, C. Elbaum, “A neural model for category learning,” Biol. Cybern. 45, 35–41 (1982).
[CrossRef] [PubMed]

Elbaum, C.

D. L. Reilly, L. N. Cooper, C. Elbaum, “A neural model for category learning,” Biol. Cybern. 45, 35–41 (1982).
[CrossRef] [PubMed]

Fessenden, R. J.

R. J. Fessenden, L. Györgyi, “Identifying functional groups in IR spectra using an artificial neural network,” J. Chem. Soc. Perkin Trans. 2, 1755–1762 (1991).

Foresman, J.

M. Rrisch, J. Foresman, A. Frisch, Gaussian 92 User’s Guide (Gaussian, Inc., Pittsburgh, Pa., 1992).

Frisch, A.

M. Rrisch, J. Foresman, A. Frisch, Gaussian 92 User’s Guide (Gaussian, Inc., Pittsburgh, Pa., 1992).

Götze, H. J.

A. Bruchmann, H. J. Götze, P. Zinn, “Application of Hamming networks for IR spectral search,” Chemometrics Intell. Lab. Syst. 18, 59–69 (1993).
[CrossRef]

Györgyi, L.

R. J. Fessenden, L. Györgyi, “Identifying functional groups in IR spectra using an artificial neural network,” J. Chem. Soc. Perkin Trans. 2, 1755–1762 (1991).

Hameka, H. F.

H. F. Hameka, J. O. Jensen, “Theoretical prediction of the infrared and Raman spectra of O-ethyl S-2diisopropyl amino ethyl methyl phosphonothiolate,” Int. J. Quantum Chem. 50, 161–172 (1994).
[CrossRef]

H. F. Hameka, A. H. Carrieri, J. O. Jensen, “Calculations of the structure and the vibrational infrared frequencies of some methylphosphonates,” Phosphorus Sulfur Silicon 66, 1–11 (1992).
[CrossRef]

Hess, B. A.

B. A. Hess, L. J. Schaad, “Ab initio calculations of vibrational spectra and their use in the identification of unusual molecules,” Chem. Rev. 86, 709–730 (1986).
[CrossRef]

Jensen, J. O.

H. F. Hameka, J. O. Jensen, “Theoretical prediction of the infrared and Raman spectra of O-ethyl S-2diisopropyl amino ethyl methyl phosphonothiolate,” Int. J. Quantum Chem. 50, 161–172 (1994).
[CrossRef]

H. F. Hameka, A. H. Carrieri, J. O. Jensen, “Calculations of the structure and the vibrational infrared frequencies of some methylphosphonates,” Phosphorus Sulfur Silicon 66, 1–11 (1992).
[CrossRef]

Original suggestions were made and a method of solution was developed by J. O. Jensen, U.S. Army Edgewood Research, Development, and Engineering Center, Aberdeen Proving Ground, Md. 21010-5423 (personal communication, 1994).

Kateman, G.

J. R. M. Smits, P. Schoenmakers, A. Stehmann, F. Sijstermans, G. Kateman, “Interpretation of infrared spectra with modular neural-network systems,” Chemometrics Intell. Lab. Syst. 18, 27–39 (1993).
[CrossRef]

Madison, M. S.

M. E. Munk, M. S. Madison, E. W. Robb, “Neural network models for infrared spectrum interpretation,” Mikrochim. Acta 2, 505–514 (1991).
[CrossRef]

Meyer, M.

M. Meyer, T. Weigelt, “Interpretation of infrared spectra by artificial neural networks,” Anal. Chim. Acta 265, 183–190 (1992).
[CrossRef]

Munk, M. E.

M. E. Munk, M. S. Madison, E. W. Robb, “Neural network models for infrared spectrum interpretation,” Mikrochim. Acta 2, 505–514 (1991).
[CrossRef]

E. W. Robb, M. E. Munk, “A neural network approach to infrared spectrum interpretation,” Mikrochim. Acta 1, 131–155 (1990).
[CrossRef]

Piffath, R.

R. Piffath, U.S. Army Chemical and Biological Defense Command, Edgewood Research, Development, and Engineering Center, Aberdeen Proving Ground, Md. 21010-5423 (personal communication, July1993).

Reilly, D. L.

D. L. Reilly, L. N. Cooper, C. Elbaum, “A neural model for category learning,” Biol. Cybern. 45, 35–41 (1982).
[CrossRef] [PubMed]

Robb, E. W.

M. E. Munk, M. S. Madison, E. W. Robb, “Neural network models for infrared spectrum interpretation,” Mikrochim. Acta 2, 505–514 (1991).
[CrossRef]

E. W. Robb, M. E. Munk, “A neural network approach to infrared spectrum interpretation,” Mikrochim. Acta 1, 131–155 (1990).
[CrossRef]

Rrisch, M.

M. Rrisch, J. Foresman, A. Frisch, Gaussian 92 User’s Guide (Gaussian, Inc., Pittsburgh, Pa., 1992).

Schaad, L. J.

B. A. Hess, L. J. Schaad, “Ab initio calculations of vibrational spectra and their use in the identification of unusual molecules,” Chem. Rev. 86, 709–730 (1986).
[CrossRef]

Schoenmakers, P.

J. R. M. Smits, P. Schoenmakers, A. Stehmann, F. Sijstermans, G. Kateman, “Interpretation of infrared spectra with modular neural-network systems,” Chemometrics Intell. Lab. Syst. 18, 27–39 (1993).
[CrossRef]

Sijstermans, F.

J. R. M. Smits, P. Schoenmakers, A. Stehmann, F. Sijstermans, G. Kateman, “Interpretation of infrared spectra with modular neural-network systems,” Chemometrics Intell. Lab. Syst. 18, 27–39 (1993).
[CrossRef]

Smits, J. R. M.

J. R. M. Smits, P. Schoenmakers, A. Stehmann, F. Sijstermans, G. Kateman, “Interpretation of infrared spectra with modular neural-network systems,” Chemometrics Intell. Lab. Syst. 18, 27–39 (1993).
[CrossRef]

Stehmann, A.

J. R. M. Smits, P. Schoenmakers, A. Stehmann, F. Sijstermans, G. Kateman, “Interpretation of infrared spectra with modular neural-network systems,” Chemometrics Intell. Lab. Syst. 18, 27–39 (1993).
[CrossRef]

Weigelt, T.

M. Meyer, T. Weigelt, “Interpretation of infrared spectra by artificial neural networks,” Anal. Chim. Acta 265, 183–190 (1992).
[CrossRef]

Zarrabian, S.

S. Zarrabian, “The use of first-derivative spectra in obtaining quantitative and qualitative information,” Spectroscopy 8, 43–45 (1993).

Zinn, P.

A. Bruchmann, H. J. Götze, P. Zinn, “Application of Hamming networks for IR spectral search,” Chemometrics Intell. Lab. Syst. 18, 59–69 (1993).
[CrossRef]

Anal. Chim. Acta

M. Meyer, T. Weigelt, “Interpretation of infrared spectra by artificial neural networks,” Anal. Chim. Acta 265, 183–190 (1992).
[CrossRef]

Appl. Opt.

Biol. Cybern.

D. L. Reilly, L. N. Cooper, C. Elbaum, “A neural model for category learning,” Biol. Cybern. 45, 35–41 (1982).
[CrossRef] [PubMed]

Chem. Rev.

B. A. Hess, L. J. Schaad, “Ab initio calculations of vibrational spectra and their use in the identification of unusual molecules,” Chem. Rev. 86, 709–730 (1986).
[CrossRef]

Chemometrics Intell. Lab. Syst.

A. Bruchmann, H. J. Götze, P. Zinn, “Application of Hamming networks for IR spectral search,” Chemometrics Intell. Lab. Syst. 18, 59–69 (1993).
[CrossRef]

J. R. M. Smits, P. Schoenmakers, A. Stehmann, F. Sijstermans, G. Kateman, “Interpretation of infrared spectra with modular neural-network systems,” Chemometrics Intell. Lab. Syst. 18, 27–39 (1993).
[CrossRef]

Int. J. Quantum Chem.

H. F. Hameka, J. O. Jensen, “Theoretical prediction of the infrared and Raman spectra of O-ethyl S-2diisopropyl amino ethyl methyl phosphonothiolate,” Int. J. Quantum Chem. 50, 161–172 (1994).
[CrossRef]

J. Chem. Soc. Perkin Trans.

R. J. Fessenden, L. Györgyi, “Identifying functional groups in IR spectra using an artificial neural network,” J. Chem. Soc. Perkin Trans. 2, 1755–1762 (1991).

Mikrochim. Acta

E. W. Robb, M. E. Munk, “A neural network approach to infrared spectrum interpretation,” Mikrochim. Acta 1, 131–155 (1990).
[CrossRef]

M. E. Munk, M. S. Madison, E. W. Robb, “Neural network models for infrared spectrum interpretation,” Mikrochim. Acta 2, 505–514 (1991).
[CrossRef]

Phosphorus Sulfur Silicon

H. F. Hameka, A. H. Carrieri, J. O. Jensen, “Calculations of the structure and the vibrational infrared frequencies of some methylphosphonates,” Phosphorus Sulfur Silicon 66, 1–11 (1992).
[CrossRef]

Spectroscopy

S. Zarrabian, “The use of first-derivative spectra in obtaining quantitative and qualitative information,” Spectroscopy 8, 43–45 (1993).

Other

M. Rrisch, J. Foresman, A. Frisch, Gaussian 92 User’s Guide (Gaussian, Inc., Pittsburgh, Pa., 1992).

R. Piffath, U.S. Army Chemical and Biological Defense Command, Edgewood Research, Development, and Engineering Center, Aberdeen Proving Ground, Md. 21010-5423 (personal communication, July1993).

Original suggestions were made and a method of solution was developed by J. O. Jensen, U.S. Army Edgewood Research, Development, and Engineering Center, Aberdeen Proving Ground, Md. 21010-5423 (personal communication, 1994).

Cited By

OSA participates in CrossRef's Cited-By Linking service. Citing articles from OSA journals and other participating publishers are listed here.

Alert me when this article is cited.


Figures (5)

Fig. 1
Fig. 1

IR geometry optimum of the molecular structure of liquid chemical nerve agent Ethyl S-(2-Diisopropylaminoethyl) Methylphosphonothiolate (VX). Reproduced by permission of H. F. Hameka.

Fig. 2
Fig. 2

Scaled mid-IR absorption spectrum of the liquid chemical nerve agent Ethyl S-(2-Diisopropylaminoethyl) Methylphosphonothiolate (VX). The dashed curve is a theoretical prediction.8 The solid curve is an experimental measurement where the numbers above the peaks (see Table 1) identity four key vibrational modes in the VX molecule. The horizontal line A′ = 0.2498 corresponds to the fourth-strongest vibrational intensity; a binary divider of this spectrum is binary +1 for A′ ≥ 0.2498 and binary −1 for A′ < 0.2498. The training data are of a 25 × 14 binary matrix form shown below the plot. The array bits are separated by 2 cm−1, read left to right by row [element [1, 1] is binary A′(702) … element [25, 14] is binary A′(1400)]. The rightmost column is a vector that associates this map with VX.

Fig. 3
Fig. 3

First-derivative-scaled mid-IR absorption spectrum of the liquid chemical nerve agent Ethyl S-(2-Diisopropylaminoethyl) Methylphosphonothiolate (VX) shown in Fig. 2. These network training data are of a 25 × 14 decimal matrix form shown below the plot. The array numerics are separated by 2 cm−1, read left to right by row [element [1, 1] is decimal A′(702) … element [25, 14] is decimal A′(1400)]. The rightmost column is a vector that associates this map with VX.

Fig. 4
Fig. 4

Neural network failure analyses. In these experiments we gradually increased noise in the training-data set until the network failed to identify the analyte, until the network made incorrect identifications, or until both failure modes were present. Column A is the analyte compound training-data set plotted in fourth-peak binary (top, SF96 ≡ 8), scaled decimal-absorption (middle, GD ≡ 6), and scaled decimal-derivative-absorption (bottom, GA ≡ 4) formats (see Table 1). Column B shows the noise threshold for the first network failure: 20% in B (SF96), 27% in B (GD), and 19% in B (GA). In B (SF96), the fourth-peak-based binary filter identified two compounds: correct identification of SF96 to 98.7% certainty and incorrect identification of agent GB to 89.4% certainty. In B (GD), the absorption-based filter also identified two compounds: a correct identification of agent GD to 98.7% certainty and an incorrect identification of agent GF to 90.3% certainty. In B (GA) the derivative-absorption-based filter failed to identify any of the nine analyte compounds (see Tables 2 and 3). Noise above the tabulated values caused rapid deterioration in the ability of these network filters to pattern match 1 of 9 compounds correctly.

Fig. 5
Fig. 5

Linkage and modular transfer of these neural network models, for a four-layer architecture with trained weight matrices, to eight Intel 80170NX chips housed on a mother circuit board through DynaMind software, Intel version 3.0. Incoming TL sensor interferometric data are operated on by high-speed digital signal preprocessing array electronics, buffered, then sent to the network’s 350-PE input layer in staggered groups of 128 PE’s. The network hidden layers contain 256 and 128 PE’s, and its output layer contains 9 PE’s. This modularized network design (splitting of weight matrices) makes full and efficient usage of each 80170NX chip. Postprocessing involves several systems-related functions, including global positioning system analyses of the mobile TL sensor unit. The modular IC linkage code for this network architecture is given on the right-hand side of the figure.

Tables (3)

Tables Icon

Table 1 Chemical Agents and Simulants of Agents with Molecular Vibrational Modes Determined from Experimental Measurement

Tables Icon

Table 2 Pattern-Matching Results for a Four-Layer, BEP Neural Network Fully Trained with Binary Spectral Representations of Nine Analyte Compoundsa

Tables Icon

Table 3 Pattern-Matching Results for a Four-Layer, BEP Neural Network Fully Trained with (Left) Decimal and (Right) Derivative-Decimal Spectral Representations of Nine Analyte Compoundsa

Metrics