^{}The authors are with the U.S. Army Chemical and Biological Defense Command, Edgewood Research, Development, and Engineering Center, Aberdeen Proving Ground, Maryland 21010-5423. USA

Arthur H. Carrieri and Pascal I. Lim, "Neural network pattern recognition of thermal-signature spectra for chemical defense," Appl. Opt. 34, 2623-2635 (1995)

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.

You do not have subscription access to this journal. Citation lists with outbound citation links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

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

Binary Data-Set Representation

Correct Identification Agent (% False)

Incorrect Identification Agent (% True)

Training Spectrum

Testing Spectrum

Noise Added (%)

${\sum}_{i=0}^{8}{{A}_{i}}^{B}$

A_{0}^{B}, A_{1}^{B}, A_{2}^{B}

15

0 (98.7), 1 (98.7), 2 (98.7)

None

A_{3}^{B}, A_{4}^{B}, A_{5}^{B}

3 (98.7), 4 (98.7), 5 (98.7)

None

A_{6}^{B}, A_{7}^{B}, A_{8}^{B}

6 (98.7), 7 (98.7), 8 (98.7)

None

A_{0}^{B}, A_{1}^{B}, A_{2}^{B}

20

0 (98.7), 1 (98.7), 2 (98.7)

None

A_{3}^{B}, A_{4}^{B}, A_{5}^{B}

3 (98.7), 4 (98.7), 5 (98.7)

None

A_{6}^{B}, A_{7}^{B}, A_{8}^{B}

6 (98.7), 7 (98.7), 8 (98.7)

8 → 5 (98.4)

${\sum}_{i=0}^{8}{{A}_{i}}^{B}$

T_{0}^{B}

0

No ID

None

T_{3}^{B}

No ID

None

T_{4}^{B}

No ID

None

T_{5}^{B}

No ID

6 (98.4)

T_{6}^{B}

6 (90.0)

2 (98.7)

${\sum}_{i=0}^{8}{{A}_{i}}^{B}$

S_{2}^{B}

0

2 (98.7)

None

S_{3}^{B}

3 (98.7)

None

S_{8}^{B}

8 (98.7)

None

Input-layer neurons 350; first-hidden-layer neurons 256 and second-hidden-layer neurons 128; output-layer neurons 9; epoch error < 10^{−15}; A^{B} binary IR absorption spectrum (Nicolet 10-DX spectrometer measurement); i = (0, 1, 2, 3, 4, 5, 6, 7, 8) ≡ (VX, Vx, DIMP, DMMP, GA, GB, GD, GF, SF96); T^{B} binary Gaussian 90 model theoretical prediction spectrum; S^{B} binary interferometer sensor spectrum, final transfer function learning rate/gain 0.10/2.00; sigmoid limits ±1; and weight filter ±2.5. The binary divider is the intensity of the fourth-strongest peak per scaled absorption spectrum.

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

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

Binary Data-Set Representation

Correct Identification Agent (% False)

Incorrect Identification Agent (% True)

Training Spectrum

Testing Spectrum

Noise Added (%)

${\sum}_{i=0}^{8}{{A}_{i}}^{B}$

A_{0}^{B}, A_{1}^{B}, A_{2}^{B}

15

0 (98.7), 1 (98.7), 2 (98.7)

None

A_{3}^{B}, A_{4}^{B}, A_{5}^{B}

3 (98.7), 4 (98.7), 5 (98.7)

None

A_{6}^{B}, A_{7}^{B}, A_{8}^{B}

6 (98.7), 7 (98.7), 8 (98.7)

None

A_{0}^{B}, A_{1}^{B}, A_{2}^{B}

20

0 (98.7), 1 (98.7), 2 (98.7)

None

A_{3}^{B}, A_{4}^{B}, A_{5}^{B}

3 (98.7), 4 (98.7), 5 (98.7)

None

A_{6}^{B}, A_{7}^{B}, A_{8}^{B}

6 (98.7), 7 (98.7), 8 (98.7)

8 → 5 (98.4)

${\sum}_{i=0}^{8}{{A}_{i}}^{B}$

T_{0}^{B}

0

No ID

None

T_{3}^{B}

No ID

None

T_{4}^{B}

No ID

None

T_{5}^{B}

No ID

6 (98.4)

T_{6}^{B}

6 (90.0)

2 (98.7)

${\sum}_{i=0}^{8}{{A}_{i}}^{B}$

S_{2}^{B}

0

2 (98.7)

None

S_{3}^{B}

3 (98.7)

None

S_{8}^{B}

8 (98.7)

None

Input-layer neurons 350; first-hidden-layer neurons 256 and second-hidden-layer neurons 128; output-layer neurons 9; epoch error < 10^{−15}; A^{B} binary IR absorption spectrum (Nicolet 10-DX spectrometer measurement); i = (0, 1, 2, 3, 4, 5, 6, 7, 8) ≡ (VX, Vx, DIMP, DMMP, GA, GB, GD, GF, SF96); T^{B} binary Gaussian 90 model theoretical prediction spectrum; S^{B} binary interferometer sensor spectrum, final transfer function learning rate/gain 0.10/2.00; sigmoid limits ±1; and weight filter ±2.5. The binary divider is the intensity of the fourth-strongest peak per scaled absorption spectrum.

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