Detection and classification of landmines using UWB antenna system and ANN analysis

• Published in: Вісник Харківського національногоуніверситету імені В.Н. Каразіна. Серія: Радіофізика та електроніка no. 33
• Main Authors: O. M. Dumin, O. A. Pryshchenko, V. A. Plakhtii, G. P. Pochanin
• Format: Journal Article
• Language: English
• Published: V. N. Karazin Kharkiv National University 01.11.2020
• ISSN: 2311-0872
• DOI: 10.26565/2311-0872-2020-33-01

Background: The problem of detecting underground objects is found in many areas of human activity in the modern world, for example, a quick survey of the territory for the presence of underground utilities for earthworks, finding the location of grounding structures, cable breakage or short circuit, remote sensing for detecting and mapping of archaeological objects. The issue of humanitarian demining in Donetsk and Lugansk regions is also important in Ukraine. The latest ground surveying devices, such as ultrawideband subsurface radar, have already come to the aid for military sappers in developed countries to make the demining process safer. Objectives: The goal of this work is to improve the recognition of subsurface objects by using an artificial neural network (ANN) for signal processing, to test the influence of interference in signals coming from ultrawideband antenna system on the reliability of determining the object in the observation area, its type and distance to subsurface radar. Materials and methods: In this work, the ANN method is used to recognize the hidden objects by ultrawideband subsurface radar. The process of electromagnetic field propagation is simulated by finite time difference method (FDTD). Neural network testing is performed by adding Gaussian noise of different levels in the input signal. Simulation of the problem is performed 1000 times to exclude the randomness of recognition for different realizations of a noise. Results: Histograms of objects recognition for two types of mines and six types of cans were obtained. A large set of false objects for neural network training gave good results in the detection of antipersonnel mines, which was reflected in the excellent stability of determining the position and type of object, even in the presence of interference with a high signal-to-noise ratio. Conclusions: The problem of subsurface survey can be solved by using a fully connected neural network with five hidden layers of neurons. It has been determined that the use of artificial intelligence gives good results in the recognition of underground objects, if a high-quality learning data set for ANN will be prepared. Satisfactory stability of noisy signal operation is shown, which gives prospects for further testing of the developed method in application to a subsurface radar in the conditions of a real experiment.