Plakhtii V. Ultrawideband Electromagnetic Fields in Problems of Subsurface Recognition Objects by Artificial Neural Networks

The dissertation is devoted to solving the actual problem, i.e. experimental and theoretical study of the processes of radiation and propagation of transient electromagnetic fields in layered media, their diffraction on subsurface objects, reception of reflected waves and recognition of these objects and their location using artificial neural networks. The scientific novelty of the dissertation is in the following results: 1. For the first time, it is shown that the actual value of the near boundary of the far zone is that observed at the moment of passage of that part of the pulse that has the greatest information load, the greatest rate of change at the receiving point. 2. For the first time, it was possible to radically improve the wave formation in the near-field of the electric impulse radiator by placing the magnetic radiator in it. The advantage of such a combined radiator, an ultra-wideband analog of the Clevin antenna, is the effective formation of a pulse wave with small post-pulse oscillations in a small physical volume without the use of additional artificial ohmic losses, which is relevant for its use in various applications, including information transmission and sensing. 3. By stitching the field components in time space, it was possible to obtain analytical expressions for the reflected field and the field transmitted into the medium in the first approximation. The possibility of forming an "electromagnetic missile" in the medium irradiated by an ultrashort duration pulsed electromagnetic wave is demonstrated. 4. The new methodology for calculating the error of determining the complex permittivity by comparing the calculated and measured databases is proposed. 5. It is shown for the first time by comparing ANNs and correlation methods for angle recognition that artificial neural networks can demonstrate better accuracy than the correlation approach. It is reliable to use ANNs up to the value of SNR = 10 dB and above and the method of mutual correlation for SNR = 20 dB and above. However, even for the SNR = 0 dB, the ANN provides correct angle recognition after statistical averaging of the classification results. In numerical modeling, the ANN demonstrates the calculation time three orders of magnitude less than we need to calculate the mutual correlation function. 6. For the first time, it was demonstrated that the use of the SoftMax layer makes the ANN responses more contrasting in subsurface sensing tasks, but leads to barely noticeable errors. The use of the Dropout method generally improved the quality of the ANN for this task. 7. For the first time, it was determined that in the presence of white noise of different levels, there are no significant advantages in the final results of object position recognition for both ANN and mutual correlation approaches. The mutual correlation method does not require time synchronization between the transmitter and the receiver, unlike the ANN, but requires significant calculation time, so it is possible to improve the quality of object location classification by combining these two approaches. Practical significance of the results: 1. The obtained expressions for the fields in the near-field of nonstationary radiators are of practical value for radar systems where the objects of study are in the near-field, and non-destructive testing systems. These results are important for the health of the personnel of radar and railgun systems. 2. Methods for determining the dielectric constant with high accuracy have been developed and radiators have been created, which is important for accurate determination of local inhomogeneities in objects in subsurface sensing tasks. 3. The radiators proposed in this work can be used in wireless local area networks, non-destructive testing, environmental monitoring, biology, medicine, etc. 4. The ultra-wideband combined Clevin antenna is practically valuable because of its relatively small size, it effectively radiates pulsed electromagnetic waves both as a single antenna and as an element of an antenna array. 5. The practical value of the proposed ANN recognition approach is based on the use of a source of fields with limited energy close to a real source, located at a low height from the ground, and applied to such a difficult object to recognize as a real anti-personnel mine. Of considerable practical value is the implementation of direct training of an artificial neural network for mine detection from experimental radar data taken in conditions close to real ones. 6. The proposed impulse-wave positioning system is practically interesting because of the absence of requirements for the time synchronization of transmitters and receivers and interference resistance in relation to traditional narrowband suppression systems.