• Toykan Ozdeger

Student thesis: Phd


Metal detection (MD) has been commonly used in humanitarian demining since MD can detect the metal components in Anti-Personnel landmines. However, land in post-conflict areas contain metal clutter such as shrapnel or bullets, resulting in a high false alarm rate (FAR). MD has been augmented with ground penetrating radar (GPR) in recent years to reduce FAR caused by metal clutter. However, GPR is also prone to FAR, caused by other clutter in the soil such as burrows, water pockets or rocks. A better discrimination for both MD and GPR could help speed up humanitarian demining by reducing FAR. The Magnetic Polarizability Tensor (MPT) is a representative electromagnetic property of a metal object, which depends on the size, material, shape, and excitation frequency of the object. If an MPT library of threat and non-threat metal objects was generated, an MD capable of measuring the MPT of the target objects could utilise this library to improve classification performance and reduce FAR. This thesis describes advances in techniques for characterising electromagnetic response of metallic targets using by exploiting their tensor description. The major contribution of this research is as follows. (i) This research has realised a new instrument and the associated measurement methodology for measuring the rank 2 tensor of typical metal objects encounter in landmine detection and security screening. The instrument utilises a novel coil geometry capable of generating a uniform magnetic field over a specific region containing the target object to accurately measure the rank 2 MPT. (ii) A novel methodology for fast and reliable measurement of the rank 2 MPT was also established. Performance of the instrument has been validated by comparing the measured rank 2 MPTs with previously published simulated and experimental data, where good agreement has been observed. (iii) The instrument was used to study the rank 2 MPT of four AP landmines and their metal components. (iv) Rank 2 MPT of 200 firearms and non-threat metal objects were measured to generate an MPT library with the motivation to improve classification on walk-through metal detectors. Clustering in the data was then studied using unsupervised machine learning (ML) techniques. This thesis has also considered how ML techniques can be used to advance the characterisation of buried non-metallic targets that would normally be inspected using GPR. Unlike MD, the GPR responses are more complex. Therefore, a large training set is required. However, a large training dataset is required to achieve high accuracy. To obtain the dataset, a real GPR dataset could be augmented with simulated data. The major contribution from this research in GPR is establishing a methodology for generating synthetic data representing real life scenarios. A methodology for generating a synthetic data representing real life scenarios is presented. A simulated GPR dataset involving AP landmines and clutter was then generated. Preliminary performance evaluation of two ML classifiers were tested using the simulated GPR data. The established methodology and the dataset can be used to augment a real GPR dataset in the future when the real data becomes available.
Date of Award31 Dec 2022
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorAnthony Peyton (Supervisor) & Frank Podd (Supervisor)


  • Metal detection
  • Machine Learning
  • Humanitarian Demining
  • Electromagnetic Induction Spectroscopy
  • Magnetic Polarizability Tensor
  • Metal Classification

Cite this