Learning-Based Profiling of Buried Elliptical-Cylindrical Objects

Abstract

This letter presents a novel AI-based approach for subsurface profiling of buried dielectric objects. Using elliptical modeling, the method frames the inverse problem as a regression task, characterizing the target’s geometry with seven parameters: center coordinates, radii, tilt angle, sector angle, and permittivity. The sector angle enables the reconstruction of diverse shapes, enhancing flexibility and accuracy. The method exclusively utilizes amplitude-only scattered field data as a direct input to the convolutional neural network (CNN), eliminating the need for complex-valued data and qualitative preprocessing, thus bypassing their inherent limitations. Numerical results demonstrate the algorithm’s efficacy in reconstructing diverse profile shapes across a wide range of permittivity values, with a relative error of 16% in predicting the output of the network. The impact of factors, such as noise levels, measurement points, multi-frequency measurements, and out-of-range parameters, is also analyzed. Furthermore, a comparative analysis with the state-of-the-art global optimization techniques underscores the superior performance of the proposed method, highlighting its potential for significant advancements in the field. This algorithm also presents itself as an appealing candidate for use as an initializer for pixel-wise methods, replacing traditional qualitative approaches.