NEURAL-NETWORK PATTERN-RECOGNITION OF THERMAL-SIGNATURE SPECTRA FOR CHEMICAL DEFENSE

We treat infrared patterns of absorption or emission by nerve and blis ter agent compounds (and simulants of this chemical group) as features for the training of neural networks to detect the compounds' liquid l ayers on the ground or their vapor plumes during evaporation by extern al heating. Training of a four-layer network architecture is composed of a backward-error-propagation algorithm and a gradient-descent parad igm. We conduct testing by feed-forwarding preprocessed spectra throug h 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 e ight artificial intelligence integrated circuits (ICs') in specific mo dular form (splitting of weight matrices). This form makes full use of all input-output IC nodes. This neural network computer serves an imp ortant real-time detection function when it is integrated into pre- an d postprocessing data-handling units of a tactical prototype thermolum inescence sensor now under development at the Edgewood Research, Devel opment, and Engineering Center.