Approved for public release, distribution is unlimited / Landmines continue to be a threat to both military and civilian communities throughout the world. Current methods of detection, while better than nothing, could certainly be improved. Seismic SONAR is a promising new technology that may help save countless lives. The goal of this thesis was to advance Seismic SONAR development by introducing a mobile source which could be easily used in practical applications. A small tracked vehicle with dual inertial mass shakers mounted on top was used for a source. The source accurately transmitted the shaker signal into the ground, and its mobility made it a practical choice for field operations. It excited Rayleigh waves, as desired, but also generated undesirable P-waves and was not found to be directional. It proved incapable of finding a target. Improvements, such as a deploying an array of mobile sources and a stronger source, should vastly enhance the performance of such tracked vehicles in seismic SONAR mine detection and should be pursued. / Ensign, United States Navy

Ground penetrating radar (GPR) is a promising technique for humanitarian demining applications as it permits providing useful information about the subsurface based on wave reflections produced by electromagnetic (EM) contrasts. Yet, landmine detection using GPR can suffer from: (1) clutter, i.e, undesirable effects from antenna coupling, system ringing and soil surface and subsurface reflections; (2) false alarms, e.g., reflections from buried mine-like objects such as stones or metallic debris; (3) effects of soil properties on the GPR performance, such as attenuation. This thesis addresses these topics in an integrated approach aiming at reducing clutter, identifying landmines from false alarms and analysing GPR performance. For subtracting undesirable reflections, a new physically-based filtering algorithm is developed, which takes into account major antenna effects and soil surface reflection. It is applied in conjunction with a change detection algorithm for enhancing landmine detection. Landmine identification is performed using discriminant characteristics extracted from the pre-filtered data by a novel feature extraction approach in the time-frequency domain. For analysing the effects of soil properties, in particular soil dielectric permittivity, an EM model is coupled to pedotransfer functions for estimating the GPR performance on a given soil. The developed algorithms are validated using data acquired by two different hand-held GPR systems. Promising results are obtained under laboratory and outdoor conditions, where different types of soil (including real mine-affected soils) and landmines (including improvised explosive devices) are considered.

Ground penetrating radar (GPR) is a promising technique for humanitarian demining applications as it permits providing useful information about the subsurface based on wave reflections produced by electromagnetic (EM) contrasts. Yet, landmine detection using GPR can suffer from: (1) clutter, i.e, undesirable effects from antenna coupling, system ringing and soil surface and subsurface reflections; (2) false alarms, e.g., reflections from buried mine-like objects such as stones or metallic debris; (3) effects of soil properties on the GPR performance, such as attenuation. This thesis addresses these topics in an integrated approach aiming at reducing clutter, identifying landmines from false alarms and analysing GPR performance. For subtracting undesirable reflections, a new physically-based filtering algorithm is developed, which takes into account major antenna effects and soil surface reflection. It is applied in conjunction with a change detection algorithm for enhancing landmine detection. Landmine identification is performed using discriminant characteristics extracted from the pre-filtered data by a novel feature extraction approach in the time-frequency domain. For analysing the effects of soil properties, in particular soil dielectric permittivity, an EM model is coupled to pedotransfer functions for estimating the GPR performance on a given soil. The developed algorithms are validated using data acquired by two different hand-held GPR systems. Promising results are obtained under laboratory and outdoor conditions, where different types of soil (including real mine-affected soils) and landmines (including improvised explosive devices) are considered.

<p>This dissertation addresses the problem of detecting buried explosive threats (i.e., landmines and improvised explosive devices) with ground-penetrating radar (GPR) and hyperspectral imaging (HSI) across widely-varying environmental conditions. Automated detection of buried objects with GPR and HSI is particularly difficult due to the sensitivity of sensor phenomenology to variations in local environmental conditions. Past approahces have attempted to mitigate the effects of ambient factors by designing statistical detection and classification algorithms to be invariant to such conditions. These methods have generally taken the approach of extracting features that exploit the physics of a particular sensor to provide a low-dimensional representation of the raw data for characterizing targets from non-targets. A statistical classification rule is then usually applied to the features. However, it may be difficult for feature extraction techniques to adapt to the highly nonlinear effects of near-surface environmental conditions on sensor phenomenology, as well as to re-train the classifier for use under new conditions. Furthermore, the search for an invariant set of features ignores that possibility that one approach may yield best performance under one set of terrain conditions (e.g., dry), and another might be better for another set of conditions (e.g., wet).</p><p>An alternative approach to improving detection performance is to consider exploiting differences in sensor behavior across environments rather than mitigating them, and treat changes in the background data as a possible source of supplemental information for the task of classifying targets and non-targets. This approach is referred to as context-dependent learning. </p><p>Although past researchers have proposed context-based approaches to detection and decision fusion, the definition of context used in this work differs from those used in the past. In this work, context is motivated by the physical state of the world from which an observation is made, and not from properties of the observation itself. The proposed context-dependent learning technique therefore utilized additional features that characterize soil properties from the sensor background, and a variety of nonparametric models were proposed for clustering these features into individual contexts. The number of contexts was assumed to be unknown a priori, and was learned via Bayesian inference using Dirichlet process priors.</p><p>The learned contextual information was then exploited by an ensemble on classifiers trained for classifying targets in each of the learned contexts. For GPR applications, the classifiers were trained for performing algorithm fusion For HSI applications, the classifiers were trained for performing band selection. The detection performance of all proposed methods were evaluated on data from U.S. government test sites. Performance was compared to several algorithms from the recent literature, several which have been deployed in fielded systems. Experimental results illustrate the potential for context-dependent learning to improve detection performance of GPR and HSI across varying environments.</p> / Dissertation

<p>The ground penetrating radar (GPR) is one of the most popular and successful sensing modalities that have been investigated for buried target detection (BTD). GPR offers excellent detection performance, however, it is limited by a low rate of advance (ROA) due to its short sensing standoff distance. Standoff distance refers to the distance between the sensing platform and the location in front of the platform where the GPR senses the ground. Large standoff (high ROA) sensing modalities have been investigated as alternatives to the GPR but they do not (yet) achieve comparable detection performance. Another strategy to improve the ROA of the GPR is to combine it with a large standoff sensor within the same BTD system, and to leverage the benefits of the respective modalities. This work investigates both of the aforementioned approaches to improve the ROA of GPR systems using statistical modeling techniques. The first part of the work investigates two large-standoff modalities for BTD systems. New detection algorithms are proposed in both cases with the goal of improving their detection performance so that it is more comparable with the GPR. The second part of the work investigates two methods of combining the GPR with a large standoff modality in order to yield a system with greater ROA, but similar target detection performance. All proposed statistical modeling approaches in this work are tested for efficacy using real field-collected data from BTD systems. The experimental results show that each of the proposed methods contribute towards the goal of improving the ROA of BTD systems.</p> / Dissertation

Ground-Penetrating Radar (GPR) is a popular non-destructive geophysical technique with a wide range of diverse applications. Civil engineering, hydrogeophysics, forensic, glacier geology, human detection and borehole geology are some of the fields in which GPR has been applied with successful or promising results. One of the most mainstream applications of GPR is landmine detection. A lot of methods have been suggested over the years to assist the landmine detection issue. Metal detectors, trained rats or dogs, chemical methods and electrical resistivity tomography are –amongst others– some of the suggested techniques. The non-destructive nature of GPR makes it an attractive choice for a problem such as demining in which contact to the ground is not allowed. The main advantage of GPR is its ability to detect both metallic and non-metallic targets. Furthermore, GPR can provide an insight regarding the nature of the target (e.g. size, burial depth, type). From the above, it is evident that GPR can potentially reduce the false alarms emerging from small metallic objects (e.g. bullets, wires, etc.) usually encountered in battle-fields and industrialised areas. Combining the robustness of the metal detector with the resolution of GPR results in a reliable and efficient detection framework which has been successfully applied in Cambodia and Afghanistan. Despite the promising, and in some cases impressive results, aspects of GPR can be further improved in an effort to optimise GPR’s performance and decrease its limitations. The validation of a GPR system is usually achieved through the so called Receiver Operation Characteristics (ROC) which depicts the probability of detection with respect to the false alarm rate. ROC is a highly nonlinear function which is sensitive to the environment as well as to the antenna unit. Landmines are typically small objects, often less than 10 cm diameter, which are shallow buried, usually in less than 10 cm depth, and sometimes almost exposed. In order for the landmines to be resolved, high frequency antennas are essential. The latter are sensitive to soil’s inhomogeneities, rough surface, water puddles, vegetation and so on. Apart from that, the near field nature of the problem makes the antenna unit part of the medium which contributes to the unwanted clutter. The above, outlines the multi-parametric nature of the problem for which no straightforward approach has yet to be proposed. Numerical modelling is a practical and solid approach to understand the physical behaviour of a system. In the case of GPR for landmine detection, numerical modelling can be a practical tool for designing and optimising antennas in synthetic but nonetheless realistic conditions. Apart from that, evaluation of a processing method only to a specific environment is not a robust approach and does not provide any evidence for its wider inclusivity and limitations. However, evaluation in different conditions can become costly and unpractical. Numerical modelling can tackle this problem by providing data for a wide range of scenarios. An extensive database of simulated responses, apart from being a practical testbed, can be also employed as a training set for machine learning. A multi-variable problem like demining, in order to be addressed using machine learning, requires a large amount of data. These must equally include all possible different scenarios i.e. different landmines, in different media with stochastically varied properties and topography. Additionally, different heights of the antenna and different depths of the landmines must also be examined. Numerical modelling seems to be a practical approach to achieve an equally distributed and coherent dataset like the one briefly described above. Numerical modelling of GPR for landmine detection has been applied in the past using generic antennas in simplified and clinical scenarios. This approach can be used in an educational context just to provide a rough estimation of GPR’s performance. In the present thesis a realistic numerical scheme is suggested in which, simplifications are kept to a minimum. The numerical solver, employed in the suggested numerical scheme, is the Finite- Difference Time-Domain (FDTD) method. Both the dispersive properties and the Absorbing Boundary Condition (ABC) are implemented through novel and accurate techniques. In particular, a novel method which implements an inclusive susceptibility function is suggested and it is shown that surpasses the performance of the previous approaches while retaining their computational efficiency. Furthermore, Perfectly Matched Layer (PML) and more specifically Convolutional Perfectly Matched Layer (CPML) is implemented through a novel time-synchronised scheme which it is proven to be more accurate compared to the traditional CPML with no additional computational requirements. An accurate numerical solver, although essential, is not the only requirement for a realistic numerical framework. Accurate implementation of the geometry and the dielectric properties of the simulated model is highly important, especially when it comes to high-frequency near-field scenarios such as GPR for landmine detection. In the suggested numerical scheme, both the soil’s properties as well as the rough surface are simulated using fractal correlated noise. It is shown, that fractals can sufficiently represent Earth’s topography and give rise to semi-variograms often encountered in real soils. Regarding the dielectric properties of the soils, a semi-analytic function is employed which relates soil’s dielectric properties to its sand fraction, clay fraction, sand density, bulk density and water volumetric fraction. Subsequently, the semi-analytic function is approximated using a Debye function that can be easily implemented to FDTD. Vegetation is also implemented to the model using a novel method which simulates the geometry of vegetation through a stochastic process. The experimentally-derived dielectric properties of vegetation are approximated –similarly to soil’s dielectric properties– with a Debye expansion. The antenna units tested in the numerical scheme are two bow-tie antennas based on commercially available transducers. Regarding the targets, three landmines are chosen, namely, PMN, PMA-1 and TS-50. Dummy landmines are used in order to obtain their geometrical characteristics and comparison between measured and numerically evaluated traces are used to tune the dielectric properties of the modelled landmines. Lastly, water puddles are realistically implemented in the model in an effort to realistically simulate high-saturated scenarios. The proposed numerical scheme has been employed in order to test and evaluate widely used post-processing methods. The results clearly illustrate that post-processing methods are sensitive to the antenna unit as well as the medium. This highlights the importance of an accurate numerical scheme as a testbed for evaluating different GPR systems and post-processing approaches in wide range of scenarios. Using an equivalent 2D numerical scheme –restricted to 2D due to computational constrains– preliminary results are given regarding the effectiveness of Artificial Neural Network (ANN) subject to an adequate and equally distributed database. The results are promising, showing that ANN can be successfully employed for detection as well as classification using only a single trace as input. A basic requirement to do so is a representative training set. This can be synthetically generated using a realistic numerical framework. The above, provide solid arguments for further expanding the proposed machine learning scheme to the more computationally demanding 3D case.

<p>In the modern sensing environment, large numbers of sensor tasking decisions must be made using an increasingly diverse and powerful suite of sensors in order to best fulfill mission objectives in the presence of situationally-varying resource constraints. Sensor management algorithms allow the automation of some or all of the sensor tasking process, meaning that sensor management approaches can either assist or replace a human operator as well as ensure the safety of the operator by removing that operator from a dangerous operational environment. Sensor managers also provide improved system performance over unmanaged sensing approaches through the intelligent control of the available sensors. In particular, information-theoretic sensor management approaches have shown promise for providing robust and effective sensor manager performance.</p><p>This work develops information-theoretic sensor managers for a general static target detection problem. Two types of sensor managers are developed. The first considers a set of discrete objects, such as anomalies identified by an anomaly detector or grid cells in a gridded region of interest. The second considers a continuous spatial region in which targets may be located at any point in continuous space. In both types of sensor managers, the sensor manager uses a Bayesian, probabilistic framework to model the environment and tasks the sensor suite to make new observations that maximize the expected information gain for the system. The sensor managers are compared to unmanaged sensing approaches using simulated data and using real data from landmine detection and unexploded ordnance (UXO) discrimination applications, and it is demonstrated that the sensor managers consistently outperform the unmanaged approaches, enabling targets to be detected more quickly using the sensor managers. The performance improvement represented by the rapid detection of targets is of crucial importance in many static target detection applications, resulting in higher rates of advance and reduced costs and resource consumption in both military and civilian applications.</p> / Dissertation

The problem of subsurface imaging with Ground Penetrating Radar (GPR) is a challenging one. Due to the low-pass nature of soil sensors must utilise wave-lengths that are of the same order of magnitude as the object being imaged. This makes imaging difficult as straight ray approximations commonly used in higher frequency applications cannot be used. The problem becomes even more challenging when the target is shallowly buried as in this case the ground surface reflection and the near-field parameters of the radar need to be considered. This thesis has investigated the problem of imaging shallowly buried targets with GPR. Two distinct problems exist in this field radar design and the design of inverse scattering techniques. This thesis focuses on the design of inverse scattering techniques capable of taking the electric field measurements from the receiver and providing accurate images of the scatterer in real time. The thesis commences with a brief introduction to GPR theory. It then provides an extensive review of linear inverse scattering techniques applied to raw GPR data. As a result of this review the thesis draws the conclusion that, due to its strong foundations in Maxwell's equations, diffraction tomography is the most appropriate approach for imaging shallowly buried targets with GPR. A three-dimensional diffraction tomographic technique is then developed. This algorithm forms the primary contribution of the thesis. The novel diffraction tomography technique improves on its predecessors by catering for shallowly buried targets, significant antenna heights and evanescent waves. This is also the first diffraction tomography technique to be derived for a range of antenna structures. The advantages of the novel technique are demonstrated first mathematically then on synthetic and finally practical data. The algorithm is shown to be of high practical value by producing accurate images of buried targets in real time.

<p>The goal of a machine learning problem is to learn useful patterns from observations so that appropriate inference can be made from new observations as they become available. Based on whether labels are available for training data, a vast majority of the machine learning approaches can be broadly categorized into supervised or unsupervised learning approaches. In the context of supervised learning, when observations are available as labeled feature vectors, the learning process is a well-understood problem. However, for many applications, the standard supervised learning becomes complicated because the labels for observations are unavailable as labeled feature vectors. For example, in a ground penetrating radar (GPR) based landmine detection problem, the alarm locations are only known in 2D coordinates on the earth's surface but unknown for individual target depths. Typically, in order to apply computer vision techniques to the GPR data, it is convenient to represent the GPR data as a 2D image. Since a large portion of the image does not contain useful information pertaining to the target, the image is typically further subdivided into subimages along depth. These subimages at a particular alarm location can be considered as a set of observations, where the label is only available for the entire set but unavailable for individual observations along depth. In the absence of individual observation labels, for the purposes of training standard supervised learning approaches, observations both above and below the target are labeled as targets despite substantial differences in their characteristics. As a result, the label uncertainty with depth would complicate the parameter inference in the standard supervised learning approaches, potentially degrading their performance. In this work, we develop learning algorithms for three such specific scenarios where: (1) labels are only available for sets of independent and identically distributed (i.i.d.) observations, (2) labels are only available for sets of sequential observations, and (3) continuous correlated multiple labels are available for spatio-temporal observations. For each of these scenarios, we propose a modification in a traditional learning approach to improve its predictive accuracy. The first two algorithms are based on a set-based framework called as multiple instance learning (MIL) whereas the third algorithm is based on a structured output-associative regression (SOAR) framework. The MIL approaches are motivated by the landmine detection problem using GPR data, where the training data is typically available as labeled sets of observations or sets of sequences. The SOAR learning approach is instead motivated by the multi-dimensional human emotion label prediction problem using audio-visual data, where the training data is available in the form of multiple continuous correlated labels representing complex human emotions. In both of these applications, the unavailability of the training data as labeled featured vectors motivate developing new learning approaches that are more appropriate to model the data. </p><p>A large majority of the existing MIL approaches require computationally expensive parameter optimization, do not generalize well with time-series data, and are incapable of online learning. To overcome these limitations, for sets of observations, this work develops a nonparametric Bayesian approach to learning in MIL scenarios based on Dirichlet process mixture models. The nonparametric nature of the model and the use of non-informative priors remove the need to perform cross-validation based optimization while variational Bayesian inference allows for rapid parameter learning. The resulting approach is highly generalizable and also capable of online learning. For sets of sequences, this work integrates Hidden Markov models (HMMs) into an MIL framework and develops a new approach called the multiple instance hidden Markov model. The model parameters are inferred using variational Bayes, making the model tractable and computationally efficient. The resulting approach is highly generalizable and also capable of online learning. Similarly, most of the existing approaches developed for modeling multiple continuous correlated emotion labels do not model the spatio-temporal correlation among the emotion labels. Few approaches that do model the correlation fail to predict the multiple emotion labels simultaneously, resulting in latency during testing, and potentially compromising the effectiveness of implementing the approach in real-time scenario. This work integrates the output-associative relevance vector machine (OARVM) approach with the multivariate relevance vector machine (MVRVM) approach to simultaneously predict multiple emotion labels. The resulting approach performs competitively with the existing approaches while reducing the prediction time during testing, and the sparse Bayesian inference allows for rapid parameter learning. Experimental results on several synthetic datasets, benchmark datasets, GPR-based landmine detection datasets, and human emotion recognition datasets show that our proposed approaches perform comparably or better than the existing approaches.</p> / Dissertation

Humanitarian clearance of minefields is most often carried out by hand, conventionally using a a metal detector and a probe. Detection is a very slow process, as every piece of detected metal must treated as if it were a landmine and carefully probed and excavated, while many of them are not. The process can be safely sped up by use of Ground-Penetrating Radar (GPR) to image the subsurface, to verify metal detection results and safely ignore any objects which could not possibly be a landmine. In this thesis, we explore the possibility of using Full Wave Inversion (FWI) to improve GPR imaging for landmine detection. Posing the imaging task as FWI means solving the large-scale, non-linear and ill-posed optimisation problem of determining the physical parameters of the subsurface (such as electrical permittivity) which would best reproduce the data. This thesis begins by giving an overview of all the mathematical and implementational aspects of FWI, so as to provide an informative text for both mathematicians (perhaps already familiar with other inverse problems) wanting to contribute to the mine detection problem, as well as a wider engineering audience (perhaps already working on GPR or mine detection) interested in the mathematical study of inverse problems and FWI.We present the first numerical 3D FWI results for GPR, and consider only surface measurements from small-scale arrays as these are suitable for our application. The FWI problem requires an accurate forward model to simulate GPR data, for which we use a hybrid finite-element boundary-integral solver utilising first order curl-conforming N\'d\'{e}lec (edge) elements. We present a novel `line search' type algorithm which prioritises inversion of some target parameters in a region of interest (ROI), with the update outside of the area defined implicitly as a function of the target parameters. This is particularly applicable to the mine detection problem, in which we wish to know more about some detected metallic objects, but are not interested in the surrounding medium. We may need to resolve the surrounding area though, in order to account for the target being obscured and multiple scattering in a highly cluttered subsurface. We focus particularly on spatial sensitivity of the inverse problem, using both a singular value decomposition to analyse the Jacobian matrix, as well as an asymptotic expansion involving polarization tensors describing the perturbation of electric field due to small objects. The latter allows us to extend the current theory of sensitivity in for acoustic FWI, based on the Born approximation, to better understand how polarization plays a role in the 3D electromagnetic inverse problem. Based on this asymptotic approximation, we derive a novel approximation to the diagonals of the Hessian matrix which can be used to pre-condition the GPR FWI problem.