METHOD OF DETECTING AND IDENTIFYING EXPLOSIVES, CHEMICAL WARFARE AGENTS AND NARCOTIC SUBSTANCES AND DEVICE FOR ITS IMPLEMENTATION Russian patent published in 2024 - IPC G01N33/22 G01N27/12 

FIELD: measurement; sampling.

SUBSTANCE: invention is intended for solving tasks of continuous sampling in automatic mode and high-speed data processing with detection of signs of explosive, chemical warfare and narcotic substances in sampled air. To detect and identify explosives, chemical warfare agents and narcotic substances, ambient air samples are taken with subsequent measurement of responses of signals from a set of semiconductor gas sensors. Data are processed by an artificial neural network and tuned to a wide range of detectable substances. Probability of the presence of a specific explosive, chemical warfare agent and narcotic substance is determined on the basis of calculating the value of the output neuron of the trained artificial neural network. Values for each individual type of substance are formed by a pre-trained neural network on the corresponding training set of air samples in number of 100 and more at ratio of 50:50 with the presence of the determined substance in the range of concentrations of the sensitivity of the device and without it, which identifies a specific substance by determining the degree of similarity of the gas pattern of the analysed air with the gas patterns of the substances of the training set of the artificial neural network. If the value of the output neuron is from 0.55 to 1, a separate explosive, military poison or narcotic substance is detected and identified. Obtained data are processed by a pre-trained neural network which identifies the substance by determining the degree of similarity of the gas pattern signals of the analysed air sample with the signals of the gas patterns of the training set of the artificial neural network. Tuning for the additionally determined substance is carried out by training and adding a new neural network. Value of the output neuron of the trained artificial neural network is estimated by formula

,

where: w_i are synoptic weights of the trained neural network, x_i is vector of input values from semiconductor non-selective sensors, h is threshold of separation of positive and negative signal of output neuron, σ is neuron activation function. Device for detection and identification of explosive, chemical warfare and narcotic substances consists of a housing, a module for measuring signals and neural network data processing, a fan and a module for displaying results. Measurement and neural network data processing module contains a set of at least three commercially available semiconductor non-selective gas sensors, having different sensitivity and operating in thermal cycling mode, and a microcontroller for providing a neural network data processing algorithm in real time when estimating the value of the output neuron of the pretrained artificial neural network by formula

,

where: w_i are synoptic weights of the trained neural network, x_i is vector of input values from semiconductor non-selective sensors, h is threshold of separation of positive and negative signal of output neuron, σ is neuron activation function.

EFFECT: detection of trace concentrations of substances at the level of pressure of saturated vapours.

2 cl, 3 dwg, 2 ex