Handheld, vehicle mounted and air-borne Ground Penetrating Radar (GPR) systems have been identified as potential technology solutions for detection of current and evolving buried threat objects. However, the success rate of the GPR systems are limited by operational conditions and the robustness of automatic target recognition (ATR) algorithms embedded with the systems. With the ever-increasing complexity of target configuration and their deployment scenarios it is becoming a challenge to develop ATR algorithms robust enough to detect and identify GPR signatures of a wide variety of threat objects. The aim of this research is to design a potential solution for detection of threat objects using GPR data and reducing the number of false alarms. In this paper, a Machine Learning (ML) based ATR algorithm applicable to GPR data is developed to detect complex patterns and trends relevant to a multitude of threat objects. The proposed ATR algorithm has been validated using a data set acquired by a vehicle mounted GPR array. The data set utilized in this investigation involved GPR data of threat objects (both conventional and improvised) commonly found in realistic operational scenarios. Lane based summaries of the algorithm performance are presented in terms of the probability of detection threat objects and false alarm rate. Preliminary results of the proposed ML techniques have shown promise of achieving a high detection rate and a low false alarm rate in multiple GPR data sets collected in challenging geographical locations.

In underwater synthetic aperture sonar (SAS) imagery, there is a need for accurate target recognition algorithms. Automated detection of underwater objects has many applications, not the least of which being the safe extraction of dangerous explosives. In this paper, we discuss experiments on a deep learning approach to binary classification of target and non-target SAS image tiles. Using a fused anomaly detector, the pixels in each SAS image have been narrowed down into regions of interest (ROIs), from which small target-sized tiles are extracted. This tile data set is created prior to the work done in this paper. Our objective is to carry out extensive tests on the classification accuracy of deep convolutional neural networks (CNNs) using location-based cross validation. Here we discuss the results of varying network architectures, hyperparameters, loss, and activation functions; in conjunction with an analysis of training and testing set configuration. It is also in our interest to analyze these unique network setups extensively, rather than comparing merely classification accuracy. The approach is tested on a collection of SAS imagery.

The ground penetrating radar (GPR) is a remote sensing technology that has been successfully used for detecting buried explosive threats. A large body of published research has focused on developing algorithms that automatically detect buried threats using data from GPR sensors. One promising class of algorithms for this purpose is convolutional neural networks (CNNs), however CNNs suffer from overfitting due to the limited and variable nature of GPR data. One solution to this problem is to use a validation dataset during training, however this excludes valuable labeled data from training. In this work we show that two modern techniques for training CNNs – Batch Normalization and the Adam Optimizer - substantially improve CNN performance and reduce overfitting when applied jointly. We also investigate and identify useful settings for several important CNN hyperparameters: l₂ regularization, Dropout, and the learning rate schedule. We find that the improved CNN (a baseline CNN, plus all of our improvements) substantially outperforms two competing conventional detection algorithms.

A novel feature extraction and buried object identification method for ground penetrating radar data are presented. Discriminative features are obtained by modelling the most dynamic peaks of GPR A-scan signals, utilizing principal component analysis (PCA). Landmine/clutter discrimination is then achieved using fuzzy k-nearest neighbor algorithm. The identification results are presented on a real data set of 700 surrogate landmines and clutter objects, which were collected from three different terrains with various soil types and buried object depths. We show that the proposed method gives outstanding results over this extensive data set.

In this article we discuss the differences and similarities in useful information, extracted from images, between several features that operate on pixel imagery and their analogues which act on voxel imagery. First we train voxel and, separately, pixel feature-based classifiers to distinguish targets from clutter using a probabilistic classification algorithm. The relative usefulness of information of these features is then measured by comparing the performance of the classifiers. Our experiments utilize voxel imagery constructed from ultra-wideband synthetic aperture radar; the pixel images analyzed are two-dimensional (2D) subimages of the voxel images. The work primarily uses features commonly employed, such as image intensity statistics, elements of an images fourier decomposition and various energy measures. The cost of different features is also discussed.

New ground penetrating radars for the vehicle mounted system can penetrate further into the ground, providing the opportunity of detecting objects that are deeply buried. A deep target has less obvious edge behaviors, especially if it has low-metal or non-metal content, making the edge based algorithm less effective for the detection. We shall explore the use of multiresolution time-frequency analysis technique by the log-Gabor filters to improve the detection of deep targets. They act on the 2-D image at an alarm location, generate multiresolution outputs and produce classification features for detection. The multiresolution analysis is able to preserve the edge behavior while at the same time forms the extra dimension of resolution (frequency) to better characterize a target signature for distinction with false alarms. Results on the detection performance at two government test sites validate the encouraging performance of the proposed detector.

With the increasing popularity of using autonomous underwater vehicles (AUVs) to gather large quantities of Synthetic Aperture Sonar (SAS) seafloor imagery, the burden on human operators to identify targets in these seafloor images has increased significantly. Existing methods of automated target detection can have perfect or near-perfect accuracy, but often produce a high ratio of false positives. Thus, it is desired to find features that discriminate between targets and high-confidence false alarms. In this paper, we examine the potential of several classical methods of feature extraction in how well their generated features can separate the two classes of image tiles: those containing targets from those containing no targets. To quantify the ability of a set of features to separate these classes, we measure the region-based cross validation accuracy of a linear SVM trained on the features in question, extracted from SAS imagery provided to us by the U.S. Navy. We show that these general feature extraction methods show potential in the ATR problem, suggesting further research is warranted.

Detection of buried explosive objects has been studied extensively and several sensors have been developed. In particular, ground penetrating radar (GPR) has proved to be one of the most successful modalities and many machine learning algorithms have been developed for buried threat detection using this sensor. Large scale experiments that involved multiple detection algorithms and very large data collections have indicated that the relative performance of different algorithms can vary significantly depending on the explosive objects, geographical site, soil and weather conditions, and burial depth. In fact, it is possible for an algorithm that performs well on training data to have low probability of target detection (PD), or high false alarm rate (FAR), on new data collected in a different environment. In this paper, we investigate the possibility of developing an algorithm that can predict the performance of a discrimination algorithm on GPR data collected in different environments. This can be used to select the optimal sensor/algorithm for a given location. It can also be used to select the optimal parameters of a given discriminator for a given site. Our approach combines predictive analysis with adequate feature selection methods to boost PD modeling and improve its prediction accuracy. Starting from raw GPR data, we extract and investigate a large set of potential descriptors that can quantify noise, surface roughness, and (implicit) soil properties. Our objectives are to: (i) Identify the optimal subset of features that can affect the target PDs of a given discriminator; and (ii) Learn a regression model for PD prediction. To validate our approach, we use data collected by a GPR sensor mounted on a vehicle. We extract over 50 different features from background regions and investigate feature selection and regression algorithms to learn a model that can predict the targets PD of a given discrimination algorithm for a given lane segment. We validate our results using different cross-validation methods.

The Edge Histogram Detector (EHD) is an algorithm for buried threat detection (BTD) using sensor data generated by ground penetrating radar (GPR). It has been tested extensively and has demonstrated excellent performance on large real-world data sets. It has been implemented in real-time versions in vehicle mounted GPR. The EHD captures the spatial distribution of the edges within a 3D GPR volume. To keep the computation simple, 2D edge operators are used, and two types of edge histograms are computed. The first one, called EHD^DT, is obtained by fixing the cross-track dimension and extracting edges in the (time, down-track) B-scans. The second edge histogram, called EHD^CT, is obtained by fixing the down-track dimension and extracting edges in the (time, cross-track) B-Scans. For confidence assignment, it uses either a possibilistic K-Nearest Neighbors (p-KNN) or a Support Vector Machine (SVM) classifier. In this paper, we first propose an improvement to the EHD by adding a new feature component extracted from (down-track, cross-track) C-Scans. We show that this feature can improve the probability of detection while reducing the false alarm rate. Second, we design a large-scale experiment to compare the performance of the p-KNN and SVM classifiers and investigate their risk of over-fitting the training data. We use large datasets accumulated across multiple dates and multiple test sites by a vehicle mounted mine detector (VMMD) using a GPR sensor. The data includes a diverse set of buried explosive objects consisting of varying shapes, metal content, and underground burial depths. Performance of the EHD with the different features and classification methods are analyzed using receiver operating characteristics (ROC). To study the potential over-fitting problem, we use two different cross validation methods.

In this work we consider the problem of developing algorithms for the automatic detection of buried threats in handheld Ground Penetrating Radar (HH-GPR) data. The development of algorithms for HH-GPR is relatively nascent compared to larger downward-looking GPR (DL-GPR) systems. A large number of buried threat detection (BTD) algorithms have been developed for DL-GPR systems. Given the similarities between DL-GPR data and HHGPR data, effective BTD algorithm designs may be similar for both modalities. In this work we explore the application of successful class of DL-GPR-based algorithms to HH-GPR data. In particular, we consider the class of algorithms that are based upon gradient-based features, such as histogram-of-oriented gradients (HOG) and edge histogram descriptors. We apply a generic gradient-based feature with a support vector machine to a large dataset of HH-GPR data with known buried threat locations. We measure the detection performance of the algorithm as we vary several important design parameters of the feature, and identify those designs that yield the best performance. The results suggest that the design of the gradient histogram (GH) feature has a substantial impact on its performance. We find that a tuned GH algorithm yields substantially-better performance, but ultimately performs similarly to the energy-based detector. This suggests that GH-based features may not be beneficial for HH-GPR data, or that further innovation will be needed to achieve benefits.

In this work we consider the problem of developing algorithms for the automatic detection of buried threats using handheld Ground Penetrating Radar (HH-GPR) data. The development of algorithms for HH-GPR is relatively nascent compared to algorithm development efforts for larger downward-looking GPR (DL-GPR) systems. One of the biggest bottlenecks for the development of algorithms is the relative scarcity of labeled HH-GPR data that can be used for development. Given the similarities between DL-GPR data and HH-GPR data however, we hypothesized that it may be possible to utilize DL-GPR data to support the development of algorithms for HH-GPR. In this work we assess the detection performance of a HH-GPR-based BTD algorithm as we vary the amounts and characteristics of the DL-GPR data included in the development of HH-GPR detection algorithms. The results indicate that supplementing HH-GPR data with DL-GPR does improve performance, especially when including data collected over buried threat locations.

We formulate the problem of the search of land mines as the inverse scattering problem for the Helmholtz Partial Differential Equation. In this problem one is looking to find location, shape and dielectric constant of a mine or mine-like target or IED. The commonly known main challenge of numerical solution of such a problem is due to non-convexity of resulting least squares cost functional. The non-convexity causes the phenomenon of multiple local minima and ravines of this functional. We overcome this challenge, we construct a weighted globally strictly convex cost functional. Its weight is the so-called Carleman Weight Function, i.e. the function involved in the Carleman estimate for the corresponding Partial Differential Operator. We will present highly accurate results of testing of our method on experimental data for microwaves.

Relative to free-space or non-responsive media, salt water (SW) alters both primary and secondary electromagnetic induction (EMI) fields that pass through it. Effects appear distinctly in the frequency domain (FD), depending on both frequency and sensor-target distance, and may be distinctly quantified. Target signatures are distorted particularly over roughly the upper half log-space of the frequency range where distinctive response patterns may appear. This paper pursues the implications of those distortions when they are translated mathematically to the time domain (TD). Effects of sensor-target standoff, target composition and orientation are pursued. Within configurations of potential interest, SW effects appear strongly in very early time (< 0.1 ms); we investigate the extent to which these effects may spill over further into the early time (ET) range (~ 0.1 ms – 1 ms), which has been important for signal interpretation in sensing on land. Lastly, computations compare the relative magnitude of these SW effects in the secondary field (SF) with those from the SW background itself.

Recent work in context-dependent processing for buried threat detection has revealed the potential of exploiting correlated environmental parameters in ground-penetrating radar (GPR) imaging, detection and discrimination, as well as data fusion approaches. In order to fully understand the physical phenomenology and develop performance predictions, we need to correlate measurable environmental parameters to GPR performance. We focus on hydro-meteorological and hydrogeologic properties such as temperature, humidity, soil water matric potential, soil water content, and sediment density and texture to understand the hydrogeophysical relationships and methods to incorporate them into optimal experimental design and data processing. Varying levels of data and information quality are assessed to quantify the sensitivity of GPR acquisition and processing methods to environmental contextual information. Both numerical modeling and experimental data collection are used to evaluate soil water controls on three-dimensional electromagnetic wave propagation. We present the analysis of experimental data collected in a new instrumented test bed facility constructed for assessing various configurations of air- or ground-coupled GPR systems. We assess factors such volume integration scale, measurement scaling and accuracy of the various hydro-electromagnetic sensing methods in terms of understanding multi-static radar angular illumination and imaging performance.

Ground-based systems for imaging objects at standoff distances usually interrogate the ground region of interest at shallow grazing angles. Thus, only a small amount of the transmitted radar energy propagates into the ground, and an even smaller portion of the target backscattered energy is returned to the sensor, thus limiting detection performance. Mounting a small size, weight and power radar on a small unmanned aerial system (sUAS) platform provides a method of interrogating the ground at more favorable angles of incidence with the radar operator being at a safe standoff distance. Reliable detection and/or imaging of a buried object from a sUAS is influenced by several issues related to sensor power, measurement of sensor position and control of the sUAS in the proximity to the ground. We performed a series of measurements of buried objects using a small radar on a tethered sUAS to explore these issues. The experiments and issues encountered will be discussed, and three-dimensional radar imagery of buried objects generated from the measurements will be presented.

Sensors which use electromagnetic induction (EMI) to excite a response in conducting bodies have long been investigated for subsurface explosive hazard detection. In particular, EMI sensors have been used to discriminate between different types of objects, and to detect objects with low metal content. One successful, previously investigated approach is the Multiple Instance Adaptive Cosine Estimator (MI-ACE). In this paper, a number of new initialization techniques for MI-ACE are proposed and evaluated using their respective performance and speed. The cross validated learned signatures, as well as learned background statistics, are used with Adaptive Cosine Estimator (ACE) to generate confidence maps, which are clustered into alarms. Alarms are scored against a ground truth and the initialization approaches are compared.

This paper describes realization of a prototype discrete-frequency self-contained NQR system into a small portable battery powered form-factor capable of detecting and classifying explosive ingredients such as sodium nitrite, ammonium nitrate, RDX, HMX, PETN, potassium nitrate, tetryl, urea nitrate, and glycine. Size is similar to the smallest commercially available NMR spectrometer, but includes all system components. This version uses fixed frequencies and simple pulse sequences to prove feasibility. Results include signature amplitude variation as a function of both sample mass and standoff, and detection unaffected by presence of both metallic and non-metallic clutter.

Many NATO navies are in the process of replacing their dedicated minehunting vessels with systems of heterogeneous, unmanned modules. While traditional ship-based assets prosecute sonar contacts in sequence through to neutralisation, modern systems employ unmanned vehicles equipped with side-looking sonar to detect and classify minelike contacts in a full area segment before proceeding with contact identification and mine neutralisation. This shift in technology and procedure brings important operational advantages, but also introduces a need to modify the traditional minehunting performance evaluation based on the percentage clearance metric. Previous works have demonstrated that the achieved detection and classification performance of modern minehunting systems can be estimated from the collected sonar data (through-the-sensor) and reported as detailed geographical maps. This paper extends the map-based evaluation approach to the identification and neutralisation phases, and also includes the case where some of the contacts or mines intentionally are left unprosecuted, e.g. disposal of only the specific mines required for establishing a safe sailing route. Each map cell is assumed to be sufficiently small to contain at most one sonar contact and can thus be assigned a status based on the hunting results for that cell: minelike contact, identified mine, etc. To this end we derive Bayesian formulations of a new performance metric: the probability of a remaining mine in a given cell. Furthermore, we show that this metric provides consistent multi-phase performance evaluation and estimates of the mine impact risk for a follow-on ship transiting a specified route.

High-frequency electromagnetic induction (HFEMI) extends the established EMI frequency range above 100 kHz to perhaps 20 MHz. In this higher frequency range, less-conductive targets display heretofore unseen responses in their inphase and quadrature components. Improvised explosive device constituent parts, such as carbon rods, small pressure plates, conductivity voids, low metal content mines, and short wires respond to HFEMI but not to traditional EMI. Results from recent testing over mock-ups of less conductive IEDs or their components show distinctive HFEMI responses, suggesting that this new sensing realm could augment the detection and discrimination capability of established EMI technology. The electrical conductivity of soil may contribute, in effect, to the imaginary part of the permittivity of soil and may then, in turn, generate perceptible responses in traditional EMI. In HFEMI, both the real and complete imaginary parts of soil permittivity produce notable effects. Pursuing this, lab tests with tap water and variously saturated Ottawa sand were compared with results from time domain reflectometry.

A method for classifying targets using a low-rank representation of broadband electromagnetic induction data is presented. The method does not require position data, a sensor model, or a complex inversion so it is applicable to hand-held EMI systems or a simple vehicle-based system. The low-rank representation is very straightforward to compute and does not require position significant computational resources. The method will be shown for data from a cart-based Georgia Tech EMI sensor that operates in the frequency domain and collects data at 15 logarithmically spaced frequencies from 1 kHz to 90 kHz. The data for several will be presented in the low-rank form to show that they are consistent within a target type and distinct for different targets. An example using the low-rank data to classify targets will be presented.

Synthetic aperture sonar (SAS) imagery can generate high resolution images of the seafloor. Thus, segmentation algorithms can be used to partition the images into different seafloor environments. In this paper, we compare two possibilistic segmentation approaches. Possibilistic approaches allow for the ability to detect novel or outlier environments as well as well known classes. The Possibilistic Fuzzy Local Information C-Means (PFLICM) algorithm has been previously applied to segment SAS imagery. Additionally, the Possibilistic K-Nearest Neighbors (PKNN) algorithm has been used in other domains such as landmine detection and hyperspectral imagery. In this paper, we compare the segmentation performance of a semi-supervised approach using PFLICM and a supervised method using Possibilistic K-NN. We include final segmentation results on multiple SAS images and a quantitative assessment of each algorithm.

Wide-band Electromagnetic Induction Sensors (WEMI) have been used for a number of years in subsurface detection of explosive hazards. While WEMI sensors have proven effective at localizing objects exhibiting large magnetic responses, detecting objects lacking or containing very low amounts of conductive materials can be challenging. In this paper, we compare a number of target detection algorithms in the literature in terms of detection performance. In the comparison, methods are tested on two real-world data sets: one containing relatively low amounts of ground noise pollution, and the other demonstrating highly-magnetic soil interference. Results are quantitatively evaluated through receiver-operator characteristic (ROC) curves and are used to highlight the strengths and weaknesses of the compared approaches in hand-held explosive hazard detection.

The goal of this paper is to describe a novel parallel high-resolution 3D numerical method for the solution of high-frequency electromagnetic wave propagation. The sequential numerical method was developed by the first author in 2014. The discussed parallel algorithm will be used later by the authors to computationally simulate data for the solution of the inverse problem of imaging mine-like targets. Thus the solution of the forward problem presented in this paper is a necessary prelude to the future solution of a related inverse problem. In this paper, land mines are modeled as small abnormalities embedded in an otherwise uniform media with an air-ground interface. These abnormalities are characterized by the electrical permittivity and the conductivity, whose values differ from those of the host media. The main challenge in the calculation of the scattered electromagnetic signal in these settings is the requirement of solving the Helmholtz equation for high frequencies. This is excessively time-consuming using standard direct solution techniques. A high-resolution and scalable numerical procedure for the solution of this equation is described in this paper. The kernel of this algorithm is a combination of a second, fourth or sixth order compact finite-difference scheme and a preconditioned Krylov subspace approach. Both fourth and sixth order compact approximations for the Helmholtz equation are considered to reduce approximation and pollution errors, thereby softening the point-per-wavelength constraint. The coefficient matrix of the resulting system is not Hermitian and possesses positive as well as negative eigenvalues. This represents a significant challenge for constructing an efficient iterative solver. In our approach, this system is solved by a combination of Krylov subspace-type method with a direct parallel FFT-type preconditioner. The resulting numerical method allows a natural and efficient implementation on parallel computers. Numerical results for realistic ranges of parameters in soil and mine-like targets confirm the high efficiency of the proposed parallel iterative algorithm.

Detection of landmines at modest distances from a moving vehicle is desired to help protect soldiers from explosive threats. The US Department of Defense developed a sensor system to observe seismic vibration patterns that lead to detection of buried objects in unpaved roads. The system is capable of exciting soil acoustically and observing seismic responses while advancing at 1m/s with a 30m standoff. The optical design and build created 960 simultaneous independent spatial observations that collect data over a 0.5m x 1m area. A gimbal then repositions the sensor to collect a new region. In this sequential manner the system can scan 2m x 1km in less than 30min. Active gimbal stabilization kept beam positions relatively stable on the ground while the vehicle was in motion. Inertial sensing reduced gross Doppler components common to all channels. Even with these corrections, channel dropouts remained a challenge, so filtering of erroneous samples in time and space was required to improve the data quality. Automated target recognition algorithms quickly process the spatial vibration data to warn operators of a threat. We present here an overview of the system and collected data. The system proved effective at finding buried threats that produced seismic anomalies at the surface, but additional challenges lie in differentiating target responses from clutter. The seismic response of naturally occurring environmental clutter in roads produces responses that appear similar to objects of interest, reducing the effectiveness of target detection algorithms.

This paper investigates diffusion behavior of electromagnetic induction (EMI) signals in multilayer structures to enhance unexploded ordnances (UXO) detection and classification in a marine environment. To date, advanced EMI systems and models have demonstrated excellent classification performance for detecting and discriminating subsurface metallic targets on land. However, the marine environment introduces complexities in both primary and secondary EMI signals. These complexities, such as salinity, air-water-sediment boundaries, etc., could negatively affect target classification performance. The main objective of this paper is to analyze time domain EMI signal diffusion and understand the factors affecting the performance of advanced EMI systems in the marine environment. We use the method of auxiliary sources and the cylindrical plane wave expansion technique to model the performance of current state of the art EMI systems for detecting and classifying underwater UXO in the frequency domain. This model accounts for the spatial (air-sea, and sea-sediment boundaries) and temporal variability of EMI fields in UW environment. Then the corresponding time domain signals are obtained using the Fourier cosine/sine transforms and Anderson filters. While others have shown the significance of environmental EMI response in marine environments, here we focus exclusively on the effect of layer boundaries in that domain. Sensitivity is shown with respect to transmitter coil sizes, sediment conductivity and magnetic permeability, and the target placement in the conducting sediment.

In this work, plastic wires buried in dry, damp and wet soils are being detected from ground penetrating radar (GPR) images. Such detection is hard, mainly due to three facts: (1) detection of buried targets made of different materials but of the same shape is difficult from GPR images as their signatures look very similar; (2) the same object buried in different soils shows different signatures in a GPR image; and (3) obtaining GPR data in the millions range is not a viable option because of the difficulties in data collection. Therefore in this work, first, domain adaptation (DA) is used to bring the information from previously trained deep learning models on standard image processing tasks into the GPR domain. It is shown that with DA, high classification rates can be achieved even with small GPR datasets, and that these rates surpass the classification rates achieved by convolutional neural networks (CNNs). However, detecting the targets in different soils still remains a problem. Therefore, secondly, a multi-task CNN is proposed, in which, soil and target classification are stitched together. In doing so, our customized classifier detects targets according to soil type, and results in superior classification rates. To the best of our knowledge, we are the first group to use multi-task learning for buried target detection with GPR.

The purpose of this paper is to present the combined use of Remote Sensing (RS) data and Geographic Information System (GIS) techniques for detection of military underground structures in Cyprus. The combination of RS data and GIS tools for detecting these types of structures entails some benefits, such as accessibility, accuracy, easy collection of data, rapid production of maps and better decision-making. In this study, Sentinel-2A satellite images have been used in order to detect underground structures. Atmospheric correction was performed using the Sen2cor plugin in Sentinel Application Platform (SNAP). In addition, maps of vegetation indices throughout the plants’ phenological development cycle were generated in ArcGIS, for: NDVI (Normalized Difference Vegetation Index), SR (Simple Ratio), DVI (Difference Vegetation Index), GEMI (Global Environment Monitoring Index), RDVI (Renormalized Difference Vegetation Index) and EVI (Enhanced Vegetation Index). The measurements were taken at two test sites: Area (a) comprises a vegetated area covered with barley, in the presence of an underground military structure, and Area (b) a vegetated area covered with barley, in the absence of an underground military structure.

Advances in training and simulation technologies, particularly in the arena of augmented reality systems, not only enable a more immersive training experience for users, but also provide more opportunities to execute training both within and outside of operational environments. The Counter-Mine Augmented Reality Training System (CMARTS) effort was an investigation of fusion of augmented reality and embedded training capabilities into fielded hand-held mine detectors to support performance training at home stations, as well as in operational environments. The resulting system, including both metal detection and ground penetrating radar (GPR) sensors, provides: (1) real-time operator feedback in the form of augmented visualizations and (2) embedded “anywhere-anytime” mine detection training using simulated targets. The real-time feedback consists of a head-mounted display and tablet visualizations to indicate what areas of ground have already been scanned; markers for where devices have already been located; any problems with user swing; detector height and swing speed; and also device power status. The embedded training capability enables operators to practice mine detection in both indoor and outdoor environments with synthetic mines, to include actual device responses to simulated detection with real-time feedback. CMARTS will provide the basis for further investigation of augmented reality applications to support both real-time operations and training, not only for mine detection but for other sensing modalities and target types as well. This paper will start with a discussion of mine detection training challenges, followed by the design and development of CMARTS, concluding with possible future efforts.

Plastics are often used in mine and IEDs. Difficult to detect with traditional approaches, plastics are spectrally active in the shortwave and mid-infrared due to vibrational absorptions from the C-H bonds of which they are composed; bonds and vibrations that are diagnostic of and spectrally vary with composition. Hyperspectral infrared imaging has proven exceedingly capable of detecting and categorizing plastics. Here we pursue a dual-band imaging approach that leverages the ubiquitous presence of the ~1.7-micron harmonic of the ~3.4-micron fundamental absorption feature for a low SWaP (Size, Weight, and Power) instrument concept. The 1.7-micron band is also in a spectral region free of telluric and almost all geologic absorption features, making its presence in a reflectance spectrum almost a unique marker for plastics. We have developed and tested a two-camera, dual-band sensor, emphasizing imaging over spectroscopy and implementing on-camera processing to achieve near real-time, partially autonomous detection and imaging of plastic objects. The sensor has proven successful in discriminating and imaging plastics such as fiberglass, styrene, and acrylics from background materials such as grass, dirt, rocks, and brush. The sensor is challenged by certain plastics, especially thin, transparent plastics (less relevant to mines and IEDs) even if they are spectrally active near 1.7 microns. Also, photometric variations in the observing conditions can mask weak plastic signatures. We will discuss our current measurement and technical approach, the results and the challenges that remain to implementing an effective low SWaP sensor for the detection and imaging of plastic objects.

Non-standard and homemade explosives can represent a certain risk for security units as the result of their, practically, unlimited variability both in content and in the construction of explosive objects. Identification of matters that have already gone through an explosion can be problematic. The number of different combinations is really high and it is not difficult to obtain precise and functional instructions for production and the construction itself is quite easy. That is why a few-years-long project was commenced the aim of which is to obtain new complex analytical data of improvised explosives residues. A whole range of experimental substances was selected for experiments including liquid explosives that can be prepared fast and easily. Experimental explosions are carried out on a freshly ploughed soil of dumps of open pit mines (where explosives are not used for mining) to avoid possible contamination. After the explosion, the residues from the stub are analyzed in a standard way and a complex analysis of both organic and inorganic phase is carried out. For selected non-stoichiometric mixtures, explosion characteristics are further calculated unless they are already known. Analysis of post-blast residues of non-standard and homemade compounds is usually started with the SEM analysis. Concerning the character of the data obtained in the frameworks of the project, a special information system is programmed. All the data obtained are entered into this system. The information system is a database layer allowing data to be entered, classified and analysed. The system shall serve the Forensic Laboratories and Expert Services of the Police of the Czech Republic and other specialized units.

Image change detection methods allow to automatically detect changes on outdoor routes with respect to a previous time point and to improve the identification of potential threats for vehicles (e.g., improvised explosive devices - IEDs). The change detection task is challenging, mainly due to the arbitrary type of changes and due to the global and local illumination differences. We have developed an illumination-invariant change detection approach based on intrinsic images and differences of Gaussians. The intrinsic images are obtained by the logchromaticity space of color images and in conjunction with reflectance images obtained by homomorphic filtering, illumination-invariant differences images are determined. We employ differences of Gaussians to detect blobs of multiple sizes in the differences images, which describe detected changes. To reduce false positives we propose to weight the scores of the detected blobs based on normalized cross-correlation, and to enforce temporally consistent changes we employ a temporal filtering scheme. The approach has been successfully applied to image sequences acquired on outdoor routes at different time points, demonstrating the consistent detection of changes over time, and the invariance to illumination differences, including shadows.

The following work details a novel enhancement to metal detector coil designs with the intent of advancing the state of the art for large-scale, free-flowing threat detection in pedestrian traffic. The enhancement is achieved by increasing the signal-to-noise ratio through shaping the magnetic field and concentrating the magnetic flux to one side. Commercial metal detectors used for security suffer from decreased range and sensitivity when operated in the vicinity of benign metallic objects (e.g. rebar, metal studs, fencing, electrical wires, etc.) that generate unwanted signals. A Halbach array design provides passive enhancements by increasing the relative flux density in the direction of interest with limited additional supporting electronics. Halbach array coils were characterized and compared with single pulse induction coils for changes in performance. Comparisons of power consumption, magnetic flux density, signal-to-noise ratio, and detection range showed a 2X increase in performance of rejecting nearby benign metallic objects performance with a 25-40% loss in power efficiency to generate a magnetic field. Through these findings, metal detection designers can optimize their systems for the local metallic environment while simultaneously improving detection performance with minimal additional hardware and power requirements to advance large-scale free-flowing metal detection for crowds.

Many mobile, exploratory machines are instrumented with sensors sensors, including radar, metal, pressure, and temperature, among others, to capture information about an environment. Often, the requirements of this type of data collection include the mapping and positioning of each of the data points but this can be difficult due to environment, operation, or equipment constraints. Traditional sensing sub-systems - such as accelerometers, Global Navigation Satellite Systems (GNSS), or camera-based vision - are commonly used to record location information. We propose a new tracking methodology that enables the reconstruction of the machine's path when these traditional positioning sensors are not present. We examine our proposed approach by applying it within handheld hazardous object detection. In this work we examine the physical space modeling component of our approach. We show that the ground can be modeled as a set of sub-regions and use relational graphs to represent regions. Using simulated region information we show that the objects path can be solved for using isomorphism algorithms.

ALIS (Advanced Land Mine Imaging System) is a dual sensor, which combines EMI sensor and GPR. Tohoku University has started the development of ALIS in 2002, and after tests in Cambodian land mine fields, it is now ready to be deployed in mine affected coutries. An evaluation test of ALIS was conducted by CMAC (Cambodian Mine Action Centre) for a certificate to use ALIS in mine fields in October 2018. We trained 4 deminers of CMAC, and they demonstrated that ALIS can images typical landmines in Cambodia including PMN-2 and Type-72 in various types of soil very clearly. The EMI sensor can pin-point the location of the buried objects, and GPR can image the targets, which can be used by the deminer to judge the buried object. The test results was good, and the certificate was given to ALIS in January 2019. Since January 2019, two ALIS systems have been deployed in mine fields in Cambodia. In this paper, we introduce the technical performances of ALIS, and then reports the recent activities of ALIS in Cambodia.

Since radar imaging of buried objects involves propagation through media that are at best partially known, there is mismatch between the forward model used in the inversion and the propagation behavior actually observed in the measured data. The mismatch can cause degradation and/or reduce resolution in the imagery, which limit automatic target recognition features that can be extracted from the imagery. Recently, several research groups have advocated backpropagation of interferometric measurements as a more statistically stable estimator of targets in the presence of forward model errors and in the presence of clutter. Specifically, the lifting approach to inverse problems [Demanet and Jugnon, 2017]1 has been proposed as a robust approach to inversion in the presence of forward model mismatch that can produce reconstructions with fidelity comparable to direct inversion with the matched model. We apply this technique to radar imaging of buried targets to determine if it can produce enhanced imagery in the presence of limited knowledge of the surrounding ground geometry and/or material properties. In this paper we describe the algorithm implementation and present results for both simulated and measured data. The results show that the approach has significant potential for enhancing images of buried objects from scenarios with realistic forward model mismatch. However, we have observed significant sensitivity to surrounding clutter and to the choice of regularization. Mitigating these sensitivities is a topic of ongoing research.

Most scatterers of very low frequency electromagnetic waves can be treated as equivalent dipoles. When the incident field is a plane wave, polarization of the scattered field can be exploited. In this preliminary report we assume multiple receivers lie in a plane orthogonal to the incident electric field. Receivers are positioned around the circumference of a five-meter radius circle enclosing the region of interest (ROI). Receivers are in the near field and phase of the measurements varies slowly with position, particularly for highly conductive scatterers. From noise-free (simulated) data, coordinates of a single scatterer can be recovered from nulls of the measured data. Accuracy is limited by the density of the receiver positions around the ROI. When multiple scatterers are present, some coordinates can be approximated from amplitude measurements. Additive white noise reduces accuracy of single-scatterer localization by as much as 10% when SNR=10 dB.

Ground Penetrating Radar is a geophysical method based on the propagation of electromagnetic waves for the prospecting of subsurface layers. The characterization of the subsurface from data obtained at the surface is an inverse process. Full Waveform Inversion (FWI) is an iterative method that requires: a cost function to measure the misfit between observed and modeled data; a wave propagator to compute the modeled data; and an initial parameters model that is updated, at each iteration step, until reaches an acceptable decrement of the cost function. FWI has a high computational cost because use the electromagnetic wave equation is discretized using Finite Difference in Time Domain. Although shielded antenna can be configured in multi-channel mode to allow large wavenumbers, this increases the budgets of the acquisitions, the processing time and the human resource in the data collection in the field. Therefore, in this paper a methodology to obtain a simultaneous inversion of permittivity, permeability, and conductivity from 2D GPR data using FWI in time domain to short-offset is proposed. The proposed methodology takes advantage of Graphical Processing Units (GPUs), through the programming language CUDA-C developed by NVIDIA, to reduce the execution time. The FWI reaches the best result (measured as the lowest cycle skipping values % CS) when there is no conductivity in the numerical experiments. This paper shows that the conductivity effect has a bigger negative impact during the inversion process than the Full Waveform inversion of noisy data.

A standoff detector is required to reduce the damage caused by explosive objects such as landmine and improvised explosive devices (IEDs). A forward looking ground penetrating radar (FLGPR) has the benefit of detecting buried objects at a significant standoff distance. We fabricated a vehicle-mounted FLGPR comprising 4 × 32 multiple-input multiple-output (MIMO) radar antennas, and then conducted a field test for the detection of buried landmines and projectiles. Here we show that the FLGPR system with LS-band succeeded in detecting the buried landmines and projectiles during vehicle running. We found that the LS-band radar was able to detect the buried landmines and projectiles while the vehicle was running by using motion compensation and multi-look processing. The experimental results suggest that the LS-band can be available for standoff detection with a vehicle-mounted FLGPR system.

Soil and meteorological conditions hamper improvised explosive device (IED)/mine detection yielding inconsistent and in some cases unacceptable probability of detection (PD) and false alarm rates (FAR). To assess and identify the environmental parameters impacting standoff thermal infrared (IR) utilization over varying temporal and spatial scales a three-month study evaluated the associated degree of variance. The lessons learned include; 1) the considerable spatial variance in surface soil temperatures at varying scales of observation, 2) spatial and temporal impact of buried objects on the thermal signature of soil, 3) identification of the environmental parameters impacting the thermal spatial and temporal temperature variance for disturbed and undisturbed soil, and 4) development of a data analysis technique taking into account temperature variance (ΔT) over time (Δς) as a approach to assess buried objects.

This paper explores using Orbital Angular Momentum (OAM) controlled electromagnetic waves for enhanced ground penetrating radar (GPR) imaging and detection. A macroscopic interpretation of OAM is propagating waves with vortexshaped wave fronts. At the photon level OAM appears as a quantum degree of freedom with integer quanta of angular momentum added to each photon. This is in addition to Spin Angular Momentum (SAM). The use of OAM in GPR has at least two potential advantages. The vortex shape may enable better discernment of cylindrical versus non-cylindrical buried objects. At the quantum level entanglement of OAM with other quantum degrees of freedom may enable enhanced imaging, such as the ghost imaging of objects that produce weak signal returns. The results include experiments that demonstrate the generation and reception of EM waves with a circular pattern of antennas operating as phased arrays to produce vortex-shaped waves at frequencies and dimensions typical of conventional GPRs.