Convolutional Models for Landmine Identification with Ground Penetrating Radar

Abstract

This thesis presents new developments in the area of target identification with ground penetrating radar (GPR) intended for the identification of plastic and metal cased antipersonnel (AP) landmines from a single measured GPR return signal, called A-scan. The target identification is formulated as a deconvolution problem. As such, the measured target response is represented through a convolutional model, which describes the sequence of radiation, propagation, target scattering and receiving, and we apply deconvolution to estimate an impulse response from which target characteristics (its outer dimensions or material properties) and target depth may be inferred. These characteristics in turn provide information on the likely identity of the detected target, i.e. âmineâ or ânot mineâ. The three main contributions of this thesis are: 1) The systematic derivation of a convolutional GPR model including closed-form expressions for the target transfer function/impulse response of an AP landmine in terms of its size, shape, electromagnetic contrast and internal structure. 2) The development of a deconvolution based target characterization procedure for circular disk-shaped minelike targets, which are representative for a large class of AP landmines. 3) The design of preprocessing algorithms (weighted moving average background subtraction and target frame transformation), which extract a target response suitable for target characterization from the measured GPR data. The target characterization procedure and the underlying convolutional models have been validated with success based on 3D finite-difference time-domain (FDTD) simulations and experimental data acquired with a video impulse GPR. In particular, the possibility to estimate the outer dimensions and depth of a minelike target with millimeter accuracy is demonstrated under laboratory conditions.