Ground Penetrating Radar Data Analysis with Nonlinear Regression on Artificial Neural Network

Abstract

Herein, a Ground Penetrating Radar (GPR) problem is defined and modelled with CST 3-D full-wave electromagnetic (EM) simulation environment. The target which has various radius is placed at different depth of soil, then reflected normalized power is obtained by using C-Band conventional horn antenna for determined points on the aperture with the help of time domain solver. Also, without target simulations are applied for the same points and background subtraction algorithm is used to eliminate soil reflection measures and other effects like noise, ground anomalies. After that, nonlinear regression function is used to obtain hyperbola for all 1-D time signals in other words A-scan data, so that one normalized power amplitude of value as an output is received. With these outputs, different Artificial Neural Networks (ANN) are worked to predict approximate backscattering normalized power amplitudes from the buried objects. Finally, the presented nonlinear regression algorithm constructs 1-D signals which are reduced from B-scan GPR images. The constructed networks can be able to correlate with the target specifications and power of the reflected signals and results are discussed.