Discrete tomography method for the processing of ultrawideband subsurface radiolocation by artificial neural network

Background: Recognition of subsurface objects became of a great importance because of the number of practical approaches in construction, archeology and energy branch. A perspective direction for the development of subsurface radiolocation lays in the construction of systems of detection of explosives and objects using ultrashort electromagnetic impulses, since they are the ones that can detect objects without metal components.Objectives: The main purpose of this work is to improve the work of artificial neural network (ANN) for the determination of subsurface object’s depth by preliminary processing of input information on the basis of the discrete tomography method.Materials and methods: The solving of a problem of determination of the location depth of subsurface object by means of irradiation by a plane electromagnetic wave and analysis of time dependences of amplitude of reflected wave at equidistant points above the ground surface is proposed. Analysis of received signals is carried out by artificial neural network of improved structure with the usage of additional data obtained due to knowledge of time dependences of received signals and material parameters of dielectric structure under investigation. The problem of Gaussian pulse propagation in the subsurface medium with objects is solved numerically by Finite Difference Time Domain method. Amplitudes of the electric field strength above the ground in given spatial points and time moments form the first part of set of input data for multilayered artificial neural network. The second part of input data includes a special linear superposition of data from the first part with coefficients received on the basis of the discrete tomography approach and the ray tracing method.Results: The work of the artificial neural network is verified by the problem of impulse electromagnetic wave irradiation of the cylindrical perfectly conducting object located inside the ground at given depth. The precision of the determination of the object depth and the influence of the second part of input data are investigated for test cases.Conclusion: Application of discrete tomography method allows to decrease the volume of input data with saving the good approximative characteristics of ANN.