This PDF file contains the front matter associated with SPIE Proceedings Volume 10628, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.

The purpose of this paper is to present the results obtained from unmanned aerial vehicle (UAV) using multispectral with thermal imaging sensors and field spectroscopy campaigns for detecting underground structures. Airborne thermal prospecting is based on the principle that there is a fundamental difference between the thermal characteristics of underground structures and the environment in which they are structure. This study aims to combine the flexibility and low cost of using an airborne drone with the accuracy of the registration of a thermal digital camera. This combination allows the use of thermal prospection for underground structures detection at low altitude with high-resolution information. In addition vegetation indices such as the Normalized Difference Vegetation Index (NDVI) and Simple Ratio (SR), were utilized for the development of a vegetation index-based procedure aiming at the detection of underground military structures by using existing vegetation indices or other in-band algorithms. The measurements were taken at the following test areas such as: (a) vegetation area covered with the vegetation (barley), in the presence of an underground military structure (b) vegetation area covered with the vegetation (barley), in the absence of an underground military structure. It is important to highlight that this research is undertaken at the ERATOSTHENES Research Centre which received funding to be transformed to an EXcellence Research Centre for Earth SurveiLlance and Space-Based MonItoring Of the EnviRonment (Excelsior) from the HORIZON 2020 Widespread-04-2017: Teaming Phase 1(Grant agreement no: 763643).

Ground penetrating radar (GPR) is an ultra-wideband electromagnetic sensor used not only for subsurface sensing but also for the detection of objects which may be hidden behind a wall or inserted within the wall. Such applications of the GPR technology are used in both military and civilian operations such as mine or IED (improvised explosive device) detection, rescue missions after earthquakes and investigation of archeological sites. Detection of concealed objects with low metal content is known to be a challenging problem in general. Use of A-scan, B-scan and C-scan GPR data in combination provides valuable information for target recognition in such applications. In this paper, we study the problem of target detection for potentially explosive objects embedded inside a wall. GPR data is numerically simulated by using an FDTD-based numerical computation tool when dielectric targets and targets with low metal content are inserted into different types of walls. A small size plastic bottle filled with trinitrotoluene (TNT) is used as the target with and without a metal fuse in it. The targets are buried into two different types of wall; a homogeneous brick wall and an inhomogeneous wall constructed by bricks having periodically located air holes in it. Effects of using an inhomogeneous wall structure with internal boundaries are investigated as a challenging scenario, paying special attention to preprocessing.

Laser Doppler vibrometers (LDVs) have been successfully used for ground vibration imaging in acoustic detection of buried objects. LDVs operating from a stationary platform or from a moving platform with a beam looking down can provide high sensitivity vibration measurement of the ground. However, operation from a moving vehicle with laser beams looking forward induces Doppler shift in the LDV beam. This shift can be much greater than the modulation bandwidth of the LDV. The demodulation must allow for the shift either by increasing the processing bandwidth, or by tracking the Doppler shift. The former increases the LDV noise while the latter can result in complex LDV design and signal processing. We developed a novel Laser Multi Beam Differential Interferometric Sensor (LAMBDIS) which provides measurement of vibration fields of objects with high sensitivity, while having low sensitivity to the whole body motion of the object, or sensor itself. The principle of operation of the LAMBDIS is based on the interference of light reflected from different points on the object surface illuminated with a linear array of laser beams. The Doppler shift induced by the sensor motion is approximately the same for all beams and is automatically subtracted from the measurements. Scanning the linear array of laser beams in the transverse direction provides a vibration image of the surface. Performance of the sensor for vibration imaging of a buried object was experimentally investigated. The experimental results and description of the sensor are presented in the paper.

Analysis of homemade and nonstandard explosives and their post-blast residues may be a bit of a challenge for forensic analysis. Analysis of these materials becomes more and more important concerning both the current global situation and the considerably easy access to precursors that are often commonly obtainable chemicals. The aim of two year running project is prepare some of these substances and carry out experimental explosions and tests, and map analyses possibilities using a wide range of available analytical techniques in forensic labs. Samples of primary substances, prepared explosives and post-blast residues are analysed in a complex way in terms of organic and inorganic components. All data obtained, including visual documentation, are stored in a specialized database for security forces and their expert workplaces. The first version of the database is to be introduced.

Phoenix has demonstrated direct detection of buried explosive material by interrogation with an intense neutron source in a laboratory environment. The technique analyzes neutron-induced emission of characteristic gamma rays by each element, so it senses the explosive material itself. The high yield of the Phoenix neutron generator (up to 3x10^11 neutrons/second) represents a leap in detection times and standoff distances for neutron-based IED detection technology. Detection experiments ranged from standoff distances of up to 7 meters, which was the limit of the laboratory space. Simulants for nitrogen-based explosives were buried in sand of different moisture levels at depths of up to 28 cm (distance to top of explosive). The fast, high-resolution gamma-ray detector array is placed 50 cm above the suspect location. The method follows a concept of operations which assumes a selection of high-risk locations have been identified using ground-penetrating radar or satellite. The location is then scanned to determine the presence of explosives. The gamma radiation emitted due to the activation is analyzed to determine a presence of 10.8 MeV nitrogen gammas higher than background. The technique can also be used as a primary detection method. The measurements validate Monte Carlo modeling of neutron-based activation techniques. Example detection times: 1.4 seconds for 30-liter jug of TNT buried 8 cm in wet soil at a standoff of 5 m, or 37 seconds for 10-liter of TNT buried 30 cm in dry sand at a standoff of 10 m.

The design and subsequent testing of suitable antennas and of complete ground-penetrating radar (GPR) systems can be both time consuming and expensive, with the need to understand the performance of a system in realistic environments of great importance to the end user. Through the use of suitably validated simulations, these costs could be significantly reduced, allowing an economical capability to be built which can accurately predict the performance of novel GPR antennas and existing commercial-off-the-shelf (COTS) systems in a user defined environment. This paper focuses on a preliminary validation of the open source software gprMax¹ which features the ability to custom define antennas, targets, clutter objects and realistic heterogeneous soils. As an initial step in the assessment of the software, a comparison of the modelled response of targets buried in sand to experimental data has been undertaken, with the variation in response with antenna stand-off height investigated. This was conducted for both a simple bespoke bow-tie antenna design as well as for a Geophysical Survey Systems, Inc. (GSSI) commercial system,² building upon previous work³ which explored the fidelity of gprMax in reproducing the S11 of simple antenna designs.

This study presents a novel landmine identification method that estimates intrinsic parameters of buried objects from their primary and secondary GPR reflections to reduce false alarm rates of GPR-based landmine detection algorithms. To achieve this, two different features are extracted from A-scan GPR data of buried objects. The first feature identifies significant GPR signal length. The second feature estimates intrinsic impedance of the object. These two features are classified with support vector machine (SVM) classifier. The experimental results show that the proposed features have very high discrimination power which reduces false alarm rates to a great extent.

This work performs scene analysis in order to represent and understand the elements contained in a defined area under the ground. Elements of interest are the ground layer, sub-surface layers, explosive hazards, and non-explosive (clutter) objects. The scene is composed of data collected by hand-held and vehicular-mounted ground penetrating radar (GPR) devices. In previous work, we segmented scenes into super-voxels and used a Markov Random Field (MRF) to combine super-voxels into layer regions. Here, we provide users with a training tool to annotate exemplar regions in sample data. Annotations associate must-link and cannot-link regions. Semi-supervised clustering is used to implement the Probability-Based Training Realignment (PBTR) algorithm. PBTR influences region labeling and increases the accuracy of scene representation.

Down-looking ground penetrating radar (DLGPR) has been used extensively for buried target detection. Performance of a DLGPR is typically measured by calculating the probability of detection (PD) and the false alarm rate (FAR) against a target set in a particular soil type. Variability in target sets, including target construction, size, layout, and burial depth, make comparing performance of a DLGPR across test sites and soil compositions a challenge. This paper describes a recent effort to collect data against a standardized set of target types, layouts, and depths. The goal of this effort is to have data sets collected in a uniform manner at various test sites in Australia and Canada for more meaningful comparisons of DLGPR performance in a range of soil types. The data is to be used to improve algorithms for the automatic detection of targets. This paper will describe test planning and execution, and discuss high-level DLGPR results and ongoing analyses from the Australian data collection.

A great deal of research has been focused on the development of computer algorithms for buried threat detection (BTD) in ground penetrating radar (GPR) data. Most recently proposed BTD algorithms are supervised, and therefore they employ machine learning models that infer their parameters using training data. Cross-validation (CV) is a popular method for evaluating the performance of such algorithms, in which the available data is systematically split into ܰ disjoint subsets, and an algorithm is repeatedly trained on ܰ−1 subsets and tested on the excluded subset. There are several common types of CV in BTD, which vary principally upon the spatial criterion used to partition the data: site-based, lane-based, region-based, etc. The performance metrics obtained via CV are often used to suggest the superiority of one model over others, however, most studies utilize just one type of CV, and the impact of this choice is unclear. Here we employ several types of CV to evaluate algorithms from a recent large-scale BTD study. The results indicate that the rank-order of the performance of the algorithms varies substantially depending upon which type of CV is used. For example, the rank-1 algorithm for region-based CV is the lowest ranked algorithm for site-based CV. This suggests that any algorithm results should be interpreted carefully with respect to the type of CV employed. We discuss some potential interpretations of performance, given a particular type of CV.

A large number of algorithms have been proposed for automatic buried threat detection (BTD) in ground penetrating radar (GPR) data. Convolutional neural networks (CNNs) have recently achieved groundbreaking results on many recognition tasks. This success is due, in part, to their ability to automatically infer effective data representations (i.e., features) using training data. This capability however results in a high capacity model (i.e., many free parameters) that is difficult to train, and more prone to overfitting, than models employing hand-crafted feature designs. This drawback is pronounced when training data is relatively scarce, as is the case with GPR BTD. In this work we propose to combine the relative advantages of hand-crafted features, and CNNs, by constructing CNN architectures that closely emulate successful hand-crafted feature designs for GPR BTD. This makes it possible to apply supervised training to traditional hand-crafted features, allowing them to adapt to the unique characteristics of the GPR BTD problem. Simultaneously, this approach yields a much lower capacity CNN model that incorporates substantial prior research knowledge, making the model much easier to train. We demonstrate the feasibility and effectiveness of this approach by designing a “neural” implementation of the popular histogram of oriented gradient (HOG) feature. The resulting neural HOG (NHOG) implementation is much smaller and easier to train than standard CNN architectures, and achieves superior detection performance compared to the un-trained HOG feature. In theory, neural implementations can be developed for many existing successful GPR BTD algorithms, potentially yielding similar benefits.

In this work, we consider the development of algorithms for automated buried threat detection (BTD) using Ground Penetrating Radar (GPR) data. When viewed in GPR imagery, buried threats often exhibit hyperbolic shapes, and this characteristic shape can be leveraged for buried threat detection. Consequentially, many modern detectors initiate processing the received data by extracting visual descriptors of the GPR data (i.e., features). Ideally, these descriptors succinctly encode all decision-relevant information, such as shape, while suppressing spurious data content (e.g., random noise). Some notable examples of successful descriptors include the histogram of oriented gradient (HOG), and the edge histogram descriptor (EHD). A key difference between many descriptors is the precision with which shape information is encoded. For example, HOG encodes shape variations over both space and time (high precision); while EHD primarily encodes shape variations only over space (lower precision). In this work, we conduct experiments on a large GPR dataset that suggest EHD-like descriptors outperform HOG-like descriptors, as well as exhibiting several other practical advantages. These results suggest that higher resolution shape information (particularly shape variations over time) is not beneficial for buried threat detection. Subsequent analysis also indicates that the performance advantage of EHD is most pronounced among difficult buried threats, which also exhibit more irregular shape patterns.

This work focuses on the development of automatic buried threat detection (BTD) algorithms using ground penetrating radar (GPR) data. Buried threats tend to exhibit unique characteristics in GPR imagery, such as high energy hyperbolic shapes, which can be leveraged for detection. Many recent BTD algorithms are supervised, and therefore they require training with exemplars of GPR data collected over non-threat locations and threat locations, respectively. Frequently, data from non-threat GPR examples will exhibit high energy hyperbolic patterns, similar to those observed from a buried threat. Is it still useful therefore, to include such examples during algorithm training, and encourage an algorithm to label such data as a non-threat? Similarly, some true buried threat examples exhibit very little distinctive threat-like patterns. We investigate whether it is beneficial to treat such GPR data examples as mislabeled, and either (i) relabel them, or (ii) remove them from training. We study this problem using two algorithms to automatically identify mislabeled examples, if they are present, and examine the impact of removing or relabeling them for training. We conduct these experiments on a large collection of GPR data with several state-of-the-art GPR-based BTD algorithms.

For the past 2 decades, detection of buried explosive hazard has been studied extensively and several machine learning algorithms have been developed and adapted to this application. First, a pre-screener is used to identify areas of interest or alarms. Each alarm consists of a 3D data cube that corresponds to spatial down-track, crosstrack, and time respectively. Ground truth information is then used to label each alarm and generate labeled data to train a classifier to discriminate between targets and clutter objects. One of the main challenges in this approach is localizing the true depth of the alarm. On one hand, the buried object signature is not expected to cover all the depth values and extracting one global feature from all depth bins may not discriminate between object and clutter signatures effectively. On the other hand, depth ground truth is not available as this depends on the target type and size, soil properties, and other environmental conditions. Moreover, visually inspecting each alarm to select the optimal depth location is tedious, ambiguous, and not practical for very large training data. Two different approaches have been considered to train learning algorithms. The first one uses simple rules and machine learning algorithms to automate the selection of the optimal depth(s) for each alarm. The second approach avoids the labeling at the depth level and uses multiple instance learning (MIL) algorithms. In this context, each alarm is represented by a bag of multiple instances. Each instance corresponds to a feature extracted at a different depth. Since labels are needed at the bag level and not at the instance level, MIL does not require true depth information. In this paper, we propose a large-scale evaluation of the two approaches. For the first approach, we consider three methods to identify optimal depth locations and analyze their effect on the KNN and SVM classifiers. For the second approach, we consider four MIL algorithms that do not require depth information. For our analysis, we use large data collections accumulated across multiple dates and multiple test sites by a vehicle mounted downward looking ground penetrating radar (GPR) sensor. The data include a diverse set of buried explosive objects of varying shapes, metal content, and underground burial depths. Performance of all algorithms is analyzed using receiver operating characteristics (ROC).

Backprojection of cross-correlated array data, using algorithms such as coherent interferometric imaging (Borcea, et al., 2006), has been advanced as a method to improve the statistical stability of images of targets in an inhomogeneous medium. Recently, the Windowed Beamforming Energy (WBE) function algorithm has been introduced as a functionally equivalent approach, which is significantly less computationally burdensome (Borcea, et al., 2011). WBE produces similar results through the use of a quadratic function summing signals after beamforming in transmission and reception, and windowing in the time domain. We investigate the application of WBE to improve the detection of buried targets with forward looking ground penetrating MIMO radar (FLGPR) data. The formulation of WBE as well the software implementation of WBE for the FLGPR data collection will be discussed. WBE imaging results are compared to standard backprojection and Coherence Factor imaging. Additionally, the effectiveness of WBE on field-collected data is demonstrated qualitatively through images and quantitatively through the use of a CFAR statistic on buried targets of a variety of contrast levels.

A buried-object detection system composed of a LWIR, a MWIR and a SWIR camera, along with a set of ground and ambient temperature sensors was constructed and tested. The objects were buried in a 1.2x1x0.3 m³ sandbox and surface temperature (using LWIR and MWIR cameras) and reflection (using SWIR camera) were recoded throughout the day. Two objects (aluminum and Teflon) with volume of about 2.5x10^-4 m³ , were placed at varying depths during the measurements. Ground temperature sensors buried at three different depths measured the vertical temperature profile within the sandbox, while the weather station recorded the ambient temperature and solar radiation intensity. Images from the three cameras were simultaneously acquired in five-minute intervals throughout many days. An algorithm to postprocess and combine the images was developed in order to maximize the probability of detection by identifying thermal anomalies (temperature contrast) resulting from the presence of the buried object in an otherwise homogeneous medium. A simplified detection metric based on contrast differences was established to allow the evaluation of the image processing method. Finite element simulations were performed, reproducing the experiment conditions and, when possible, incorporated with data coming from actual measurements. Comparisons between experiment and simulation results were performed and the simulation parameters were adjusted until images generated from both methods are matched, aiming at obtaining insights of the buried material properties. Preliminary results show a great potential for detection of shallowburied objects such as land mines and IEDs and possible identification using finite element generated maps fitting measured surface maps.

Laos is one of the most seriously contaminated countries by UXO. Most of the UXO are cluster bombs, used in 1960- 70’s. There is historical political background about the UXO problems in Laos. Currently, The clearance of UXO is significantly important for future economic development for Laos. It affects not only local inhabitants, but it will affects national economy. For example, UXO clearance in mountainous area can promote construction of roads, and Laos has a good possibility to be the center of transportation in Indo-Chia region. We have started a research project for evaluation of impact of UXO clearance in Laos, “The economic impact and the technological development for the elimination of unexploded ordnance (UXO) elimination in Laos”. Lao National Unexploded Ordnance Programme (UXO Lao) is working in the nine most impacted provinces nationwide and it clears land for agriculture and community purposes. We recently visited Laos and observed the UXO clearance procedure conducted by UXOLAO. The technical survey for UXO clearance site to determine for operation, UXO detection by using metal detectors are organized by UXO Lao. In this paper, we will report the technical aspects of the UXO clearance in Laos.

Thin wires are a critical component of many types of improvised explosive devices. Short wires with lengths on the order of 30 cm to a few meters are difficult to detect using electromagnetic induction metal detectors due to the induction currents having only a small cross-section of the wire to circulate on. Longer wires on the order of tens of meters up to a kilometer are often buried at depths which preclude induction detection. We demonstrate short wire detection and identification through acquiring the electromagnetic induction response at frequencies above the traditional EMI range. In addition, long wire detection and identification is shown through electric field coupling between excitation coils and the long wire itself. We present the relevant physics of detecting both types of wires and experimental and modeling results demonstrating the utility of this high-frequency EMI regime. We present a high-frequency electromagnetic induction instrument utilizing frequencies up to 15 MHz which can detect and classify both short and long wires.

Processing electromagnetic induction signals from subsurface targets, for purposes of discrimination, requires accurate physical models. To date, successful approaches for on-land cases have entailed advanced modeling of responses by the targets themselves, with quite adequate treatment of instruments as well. Responses from the environment were typically slight and/or were treated very simply. When objects are immersed in saline solutions, however, more sophisticated modeling of the diffusive EMI physics in the environment is required. One needs to account for the response of the environment itself as well as the environment’s frequency and time-dependent effects on both primary and secondary fields, from sensors and targets, respectively. Here we explicate the requisite physics and identify its effects quantitatively via analytical, numerical, and experimental investigations. Results provide a path for addressing the quandaries posed by previous underwater measurements and indicate how the environmental physics may be included in more successful processing.

Recent work with Wideband Electromagnetic Induction (WEMI) sensors has shown that a low-rank model can be used to exploit the measurements. The low-rank model has led to a new filterless processing framework for frequency-domain WEMI sensors, where projection operators can be used in both the frequency and spatial dimensions of the data. Previous work has used a single subspace from the projected measurements to perform target detection, classification, and localization. This work investigates the eight remaining measurement subspaces created by the projection operators and how they can be exploited to extract more information for WEMI processing.

In processing electromagnetic induction (EMI) survey data for UXO discrimination, the nonlinear search for the location of subsurface objects (“targets”) is one of the most onerous and error-producing components. Here we pursue a technique that exploits fundamental analytical electromagnetic relations to extract estimates of a target’s subsurface position. The formulation requires that one be able to estimate secondary field gradients from the data, e.g. by differencing values from multiple, nearby vector receivers. Further, the field in the vicinity of the Rx units is idealized as one structured like that from an infinitesimal dipole responder; this need not restrict the model complexity applied in other aspects of the processing. Once the system is formulated from the data, direct solution of a 3×3 matrix is the only computational burden. Essentially instantaneous results in simulation tests and in processing of some field data locate a hypothetical source point that varies from nominal target position only by a distance on the order of UXO sizes.

This paper studies high (100 kHz up to 15 MHz) frequency electromagnetic responses (HFEMI) for DU metallic pieces and DU contaminated soils and derives a simple empirical expression from the measured HFEMI data for estimating DU contamination levels in soil. Depleted uranium (DU) is the byproduct of uranium enrichment and contains 33% less radioactive isotopes than natural uranium. There are at least thirty facilities at fourteen separate locations in the US, where munitions containing DU have been evaluated or used for training. At these sites, which vary in size, evaluation studies have been conducted with and without catch boxes. In addition, the DoD used DU at open firing ranges as large as thousands of acres (hundreds of hectares), for both artillery and aircraft training. These activities have left a legacy of DU contamination. Currently at military sites where DU munitions have been or are being used, cleanup activities mainly are done by excavating and shipping large volumes of site soil and berm materials to a hazardous material radiation disposal site. This approach is very time consuming, costly, and associated with the potential for exposure of personnel performing excavation and transportation. It also limits range use during the operation. So, there is an urgent need for technologies for rapid surveying of large areas to detect, locate, and removal of DU contaminants at test sites. Additionally, the technologies are needed to detect material at a depth of at least 30 cm as well as discriminate between DU metals and oxides from natural uranium and from other conductive metals such as natural and man-made range clutter. One of the potential technologies for estimating DU radiation levels in soils is HFEMI sensing. In this paper, HFEMI signals are collected for DU metal pieces, sodium diunarate (Na₂U₂ O₃) and tri-uranium octoxide (U₃O₈). The EMI signal’s sensitivity with respect to DU material composition and conditions are illustrated and analyzed. A new scheme for extracting near-surface soil’s EM parameters is formulated.

Broadband electromagnetic induction sensors are effective at detecting and classifying buried metal. Electromagnetic induction sensors operate by exciting eddy currents in conducting targets using a primary magnetic field and measuring the scattered magnetic response. Broadband sensors gather increased information about the flow of eddy currents relative to narrowband sensors. A dipole model can be fitted to the scattered response, which allows the scattering mechanism to be represented using a frequency-dependent magnetic polarizability tensor. This tensor can be decomposed into a pole expansion of frequency-independent tensors, which represent the scattering of the natural modes of the eddy current problem, and a corresponding relaxation frequency that characterizes the mode’s exponential decay in time. The pole-expansion coefficients are valuable for target classification and for discriminating targets from clutter. In this paper, a volume integral method is used to compute the pole-expansion coefficient of a few canonical three-dimensional targets. These coefficients can be used to compute the eddy-current response of a larger subset of targets.

Many different coil head configurations are used in electromagnetic induction (EMI) systems for sensing buried targets. The design of these types of coils is not well described in current literature, and it is difficult to compare the performance of different coil head designs to one another. Comparing two particular implementations is not particularly challenging, but fairly comparing general coil head designs is nontrivial. This work details normalized target sensitivity and soil sensitivity metrics that account for differences such as overall coil size, length, and wire diameter and for variations in sensitivity patterns. The metrics are used to optimize double-D, dipole/quadrupole, and concentric coil head designs, which are then analayzed and compared to one another.

In this paper, we derive the Cramer-Rao Lower Bound (CRLB) for the unknown tensor amplitudes and target location parameters that can be estimated from electromagnetic induction (EMI) measurements of a target. In deriving the bound, no restrictions are placed on the target type, target orientation, measurement geometry, or the geometry of the electromagnetic induction sensor or sensor array. The analysis is applicable to both a scanned sensor and a stationary array. In doing so, we illustrate ways to use the CRLB to analyze the relationships between parameters and parameter sets. We then apply the analysis to an experimental Georgia Tech EMI sensor for a nominal 2-D scan geometry. We also identify additional considerations in designing EMI sensors and acquisition setups.

We recently developed a new, man-portable, electromagnetic induction (EMI) sensor designed to detect and classify small, unexploded sub-munitions and discriminate them from non-hazardous debris. The ability to distinguish innocuous metal clutter from potentially hazardous unexploded ordnance (UXO) and other explosive remnants of war (ERW) before excavation can significantly accelerate land reclamation efforts by eliminating time spent removing harmless scrap metal. The EMI sensor employs a multi-axis transmitter and receiver configuration to produce data sufficient for anomaly discrimination. A real-time data inversion routine produces intrinsic and extrinsic anomaly features describing the polarizability, location, and orientation of the anomaly under test. We discuss data acquisition and post-processing software development, and results from laboratory and field tests demonstrating the discrimination capability of the system. Data acquisition and real-time processing emphasize ease-of-use, quality control (QC), and display of discrimination results. Integration of the QC and discrimination methods into the data acquisition software reduces the time required between sensor data collection and the final anomaly discrimination result. The system supports multiple concepts of operations (CONOPs) including: 1) a non-GPS cued configuration in which detected anomalies are discriminated and excavated immediately following the anomaly survey; 2) GPS integration to survey multiple anomalies to produce a prioritized dig list with global anomaly locations; and 3) a dynamic mapping configuration supporting detection followed by discrimination and excavation of targets of interest.

The identification followed by avoidance or removal of explosive hazards in past and/or present conflict zones is a serious threat for both civilian and military personnel. This is a challenging task as extreme variability exists with respect to the objects, their environment and emplacement context. A goal is the development of automatic, or human-in-the-loop, sensor technologies that leverage engineering theories like signal processing, data fusion and machine learning. Herein, we explore the detection of buried explosive hazards (BEHs) in handheld ground penetrating radar (HH-GPR) via convolutional neural networks (CNNs). In particular, we investigate the potential for generative adversarial networks (GANs) to impute new data based on limited and class imbalance labeled data. Unsupervised GANs are trained and assessed at a qualitative level and their outputs are explored in different ways to quantitatively help train a CNN classifier. Overall, we found encouraging qualitative results and a list of hurdles that need to be overcome before we anticipate quantitative improvements.

The use of handheld ground penetrating radar (GPR) for subsurface object detection often faces challenges coming from the human operator effect, antenna height variation and uneven data sample spacing. This paper investigates the artifact of uneven sample spacing on the performance of the features extracted from the handheld GPR, for the discrimination between targets and false alarms at the initial detection locations reported by a prescreener. The features we examined are the log Gabor and the Local Binary Pattern (LBP). They have been shown previously to be able to improve the detection performance in the absence of sample spacing artifact. The effect of the variation in the detection alarm location from the prescreener on the features will also be examined. The detection performance with and without sample spacing artifact and alarm location accuracy will be contrasted on a data set collected at a government test site.

Tohoku University has developed dual sensor ALIS for humanitarian demining and 2 sets of ALIS are used by CAMC (Cambodian Mine Action Centre) since 2009 and detected more than 80 mines. We have demonstrated the performance of mine clearance using ALIS together with CMAC. In 2017, we launched a new advanced ALIS as a commercial product. Based on the development and field evaluation since 2002, the advanced ALIS, which weights 3100g, is a compact device, which can visualize the GPR and Metal Detector response on a display of a PAD attached to the sensor. The system can acquire the sensor position information and metal detector signal together with GPR signal. Then, we process the GPR signal and reconstruct subsurface image to detect buried mines. We found that the reconstruction algorithm is the key technology to determine buried mines under very strong clutter condition. We conducted evaluation test of ALIS in a test site in Cambodia, and could show ALIS can visualize the buried mine very clearly. We are planning to deploy ALIS in mine affected countries including Cambodia and Colombia.

Ground penetrating radar (GPR) is used in explosive hazard clearance. Handheld sensors are utilized by operators during search and localization sweeps. These two styles of ground interrogation follow a standard operating procedure (SOP) that attempts to maximize both ground coverage and rate of advance (ROA). As a result, the data collected is non-uniformly sampled, with areas of missing data. We analyze a number of data interpolation methods in order to develop discrimination algorithms and human visualizations. We analyze the performance of interpolative methods for sweep patterns in the parallel and perpendicular direction by using Mean Squared Error (MSE), Mean Absolute Error (MAE), and Structural Similarity Index (SSIM). Overall, a nearest neighbors (2-NN) method performs best.

Acoustic intensity probes are readily available commercial assets often found in the automobile industry. Pairing of vector mechanics based readings of acoustic output coupled with video allow users to quickly observe visually where a noise source exists on a target of interest. The advent of undersea acoustic vector sensor technology in the last few decades has allowed for the possibility of compact low frequency measurement tools on the order of hand-held for diver use size applications. By coupling the relative omni-directional beam patterns of an acoustic vector sensor with the relative field of view of a hand-held camera, it becomes possible to map acoustic wavefront estimations to the pixelated video output of the camera. An example using variable drive motors in a remote operated vehicle is presented.

The ability to discern the characteristics of the seafloor has many applications. Due to minimal visibility, Synthetic Aperture Sonar Imagery (SAS) uses sonar to produce a texture map of the seabed below. In this paper, we discuss an approach to detecting targets from varying seafloor contexts. The approach begins with one or more anomaly detecting prescreeners that use minimal information about targets and that can be applied under various seafloor conditions. In addition, these anomaly detectors see multiple fusion experiments and manipulation to bolster and account for unique target characteristics. Suppressed hits or peaks in the resultant confidence surface, are further processed for scoring. Through ROC curve production and areas under their curves, detection effectiveness becomes simple to distinguish. Attention is paid to determine performance with respect to seafloor type from various locations. The approach is tested on a SAS data collection conducted by the U.S. Navy.

Fractal analysis of an image is a mathematical approach to generate surface related features from an image or image tile that can be applied to image segmentation and to object recognition. In undersea target countermeasures, the targets of interest can appear as anomalies in a variety of contexts, visually different textures on the seafloor. In this paper, we evaluate the use of fractal dimension as a primary feature and related characteristics as secondary features to be extracted from synthetic aperture sonar (SAS) imagery for the purpose of target detection. We develop three separate methods for computing fractal dimension. Tiles with targets are compared to others from the same background textures without targets. The different fractal dimension feature methods are tested with respect to how well they can be used to detect targets vs. false alarms within the same contexts. These features are evaluated for utility using a set of image tiles extracted from a SAS data set generated by the U.S. Navy in conjunction with the Office of Naval Research. We find that all three methods perform well in the classification task, with a fractional Brownian motion model performing the best among the individual methods. We also find that the secondary features are just as useful, if not more so, in classifying false alarms vs. targets. The best classification accuracy overall, in our experimentation, is found when the features from all three methods are combined into a single feature vector.

Automated anomaly and target detection are commonly used as a prescreening step within a larger target detection and target classification framework to find regions of interest for further analysis. A number of anomaly and target detection algorithms have been developed in the literature for application to target detection in Synthetic Aperture Sonar (SAS) imagery. In this paper, a comparison of two anomaly and one target detection algorithm for target detection in synthetic aperture sonar is presented. In the comparison, each method is tested on a large set of real sonar imagery and results are evaluated using receiver operating characteristic curves. The results are compiled and quantitatively shown to highlight the strengths and weakness of the variety of approaches within various sea-floor environments and on particular target shapes and types.

The Possibilistic Fuzzy Local Information C-Means (PFLICM) method is presented as a technique to segment side-look synthetic aperture sonar (SAS) imagery into distinct regions of the sea-floor. In this work, we investigate and present the results of an automated feature selection approach for SAS image segmentation. The chosen features and resulting segmentation from the image will be assessed based on a select quantitative clustering validity criterion and the subset of the features that reach a desired threshold will be used for the segmentation process.

The U.S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate (NVESD) Countermine Division has developed a program to investigate multiple sensor modalities to detect side-attack explosive hazards. Sensor modalities include vehicle-mounted forward- and side-looking radar, side-looking acoustic, and forward- and sidelooking electro-optical/infrared sensors. NVESD is collaborating with the Institute for Defense Analyses and multiple universities to execute data collections and conduct data analysis and algorithm development. A variety of sensors have been tested at a U.S. Army test site, and performance has been measured by calculating the probability of detection and the false-alarm rate for a given target set. Reliable detection is challenging due to variations in target design, target emplacement, environment, and obscuration. Current analysis is focused on developing feature-extraction methods and determining sensors’ abilities to penetrate concealment. This paper will discuss several data analysis efforts to date that have resulted in consideration of a high-frequency 3D radar for detection and discrimination.

The detection of side-attack explosive hazards (SAEHs) is a challenging task especially if the SAEHs are camouflaged. Three-Dimensional Radar is one of the most prominent sensors that has shown a great capacity for detecting concealed SAEHs. This system produces high-resolution volumetric images where each voxel’s intensity represents the radar signal return at a specific point in the three-dimensional space. It has the capability to enhance the signal response from a SAEH nested in camouflage materials and suppress the interference from the surroundings. Nevertheless, processing the radar data in the spatial domain has some limitations in differentiating SAEH from clutter objects. In this paper, we propose the use of the discrete Fourier transform (DFT) to analyze the voxel data and capture the spatial frequency characteristics of the radar signal from SAEHs. Through a machine learning approach, our proposed algorithm is able to identify the frequency signatures of SAEHs and to differentiate them from anomalies caused by the background or clutter. This approach yields a confidence value indicating the likelihood of a SAEH at a particular location. The detection ability of the proposed algorithm is demonstrated by the receiver operating characteristic (ROC) curves generated using a dataset collected from a U.S. Army test site.

The detection of side-attack explosive hazards remains challenging due to the significant variation in size, shape, construction materials, and placement on or above the surface. Some of the most challenging-to-detect side-attack explosive hazards are those placed inside of naturally occurring clutter such as vegetation. High-frequency radar systems with 3D resolution have been observed to be an effective technology for detecting and discriminating surface-laid sideattack explosive hazards from both natural and manmade clutter. Automated target recognition on the 3D voxel radar data is a complex problem that is well suited for deep convolutional neural networks. The main drawback of such approaches is the requirement for a large amount of training data, which is expensive and time-consuming to collect. Ad hoc and generative models have been used to augment data for deep learning with some degree of success; however, these methods often generate examples closely resembling instances from the training data, and any deviations are potentially not physically realistic for the sensing phenomenology. More accurate and effective augmentation can be accomplished by leveraging sensor physics along with large amounts of readily available background data. Observations of target signatures under clutter-free conditions can be inserted into a cluttered scene in a way consistent with the physics governing the sensor. We show that our physics-based data augmentation technique yields realistic synthetic data that is useful for augmenting the available training data and leads to improved discrimination performance.

Ground based, mobile surface anomaly sensing and detection is a critical area of research in explosive hazard detection as well as local situational awareness and even autonomous operations. Increasingly, achieving reliable detection is coming to rely on a suite of different (often orthogonal) sensing modalities from optical to infrared to lidar and radar. Radar is of particular interest because it offers advantages when attempting to detect obscured surface anomalies and has the potential for large observation areas. Radar’s chief disadvantage in this context is that limited physical antenna aperture degrades the spatial localization of scattering returns. This is particularly troubling in highly cluttered surface environments. To address this shortcoming of spatial localization of scattering returns, this paper discusses the use of a MIMO X-band radar system configured in a high-resolution side-looking instantiation. By configuring the system this way it can be operated in a traditional stripmap synthetic aperture mode and since it is a MIMO array it has vertical aperture allowing for three-dimensional imagery to be formed. This paper details system elements, configuration and operation of a high resolution ground based, mobile MIMO X-band radar for side-looking anomaly detection. The system operates in X-band and utilizes a digitally synthesized frequency modulated continuous waveform. The system has previously been configured for forward looking mobile anomaly detection. The work presented in this paper is concerned with synthesizer and radio frequency electronics upgrades, side-looking specific configuration issues and image formation issues. Example side-looking three-dimensional imagery is shown using canonical calibration targets.

In this paper, we present a methodology for detecting side-attack explosive hazards using three-dimensional radar imaging. Our methodology is based on clustering intensities of voxel cubes extracted around points of interest generated from prescreener filters. To make our computations easier, and results visually comprehensible, we break our voxel cubes into slices within the x-, y-, and z-directions of equal dimensions. With these slices, we explore various feature extraction algorithms: K-Means, Fuzzy C-Means (FCM), and statistical moments on the radar intensity slices to create feature vectors based on a set number of cluster centers and number of slices within the extracted cubes. We evaluate the performance of the features produced using Hidden Markov Model (HMM) classifiers on a set of lane data supplied by the US Army.

This paper presents an approach for estimating the confidence level in automatic multi-target classification performed by an imaging sensor on an unmanned vehicle. An automatic target recognition algorithm comprised of a deep convolutional neural network in series with a support vector machine classifier detects and classifies targets based on the image matrix. The joint posterior probability mass function of target class, features, and classification estimates is learned from labeled data, and recursively updated as additional images become available. Based on the learned joint probability mass function, the approach presented in this paper predicts the expected confidence level of future target classifications, prior to obtaining new images. The proposed approach is tested with a set of simulated sonar image data. The numerical results show that the estimated confidence level provides a close approximation to the actual confidence level value determined a posteriori, i.e. after the new image is obtained by the on-board sensor. Therefore, the expected confidence level function presented in this paper can be used to adaptively plan the path of the unmanned vehicle so as to optimize the expected confidence levels and ensure that all targets are classified with satisfactory confidence after the path is executed.

Superpixel segmentation methods have been found to be increasingly valuable in image processing and analysis. Superpixel segmentation approaches have been used as a preprocessing step for a wide variety of image analysis tasks such as full scene segmentation, automated scene understanding, object detection and classification, and have been used to reduce computation time during these tasks. While many quantitative evaluation metrics have been developed in the literature to analyze traditional image segmentation and clustering results, these metrics have not been used or adapted to quantitatively evaluate superpixel segmentations. In this paper, multiple superpixel segmentation algorithms are applied to synthetic aperture sonar (SAS) imagery and the results are evaluated using cluster validity indices that have been adapted for superpixel segmentation. Both cluster validity metrics that rely only on internal measures as well as those that use both internal and external measures are considered. Results are shown on a synthetic aperture sonar (SAS) data set.

The attraction of synthetic aperture sonar (SAS) is the promise of achieving high resolution across the entire sonar image. However, this theoretical resolution depends on maintaining phase coherence over the full synthetic aperture length or integration angle. There may be some advantages to reducing the integration angle for SAS processing. For example, when considering the large amount of sonar data to be processed for a single mission, there may be a computational savings to be gained by truncating the synthetic aperture length. This could be critical when the system needs to meet post processing time requirements or when sonar data is being processed in situ to enable image queued autonomous behaviors. Another advantage may be to have the option to narrow the integration angle in the case of uncompensated vehicle motion or incorrect estimates of the sound speed. The narrowing may improve the image quality without significantly compromising the information content of the data and the subsequent Automated Target Recognition (ATR) performance. The ATR performance is investigated using simulated SAS data over 10 different integration angles and three backgrounds whose sound speed ratios range from 0.98 to 1.28. An Ada-Boosted Decision Tree classifier was used to calculate the probability of classification (Pc) and false alarm rate (FAR) and generate receiver operator characteristic (ROC) curves. Additionally, a measure of the Fisher information of the contact image snippets is investigated as a function of integration angle and object pose angle.

This article discusses a method to identify targets in the raw data collected by a 2-dimensional radar array. A series of target simulations, in frequency space, are correlated against experimental data and these correlations are used to identify the target in the different data snippets. In each case, the target whose simulation correlates most strongly with the experimental data positively identifies the target in the data. We furthermore find that the two strongest cross-correlations between any particular experimental data sample and all target simulations has a 42% mean difference across all experimental data. From this exploratory study we conclude that this approach positively identifies all targets in our data set but that a more thorough study is necessary to determine its robustness.

The automated detection of changes occurring between multi-temporal images is of significant importance in a wide range of medical, environmental, safety, as well as many other settings. The usage of k-means clustering is explored as a means for detecting objects added to a scene. The silhouette score for the clustering is used to define the optimal number of clusters that should be used. For simple images having a limited number of colors, new objects can be detected by examining the change between the optimal number of clusters for the original and modified images. For more complex images, new objects may need to be identified by examining the relative areas covered by corresponding clusters in the original and modified images. Which method is preferable depends on the composition and range of colors present in the images. In addition to describing the clustering and change detection methodology of our proposed approach, we provide some simple illustrations of its application.

The identification followed by avoidance or removal of explosive hazards in past and/or present conflict zones is a serious threat for both civilian and military personnel. This is a challenging task as variability exists with respect to the objects, their environment and emplacement context, to name a few factors. A goal is the development of automatic or human-in-the-loop sensor technologies that leverage signal processing, data fusion and machine learning. Herein, we explore the detection of side attack explosive hazards (SAEHs) in three dimensional voxel space radar via different shallow and deep convolutional neural network (CNN) architectures. Dimensionality reduction is performed by using multiple projected images versus the raw three dimensional voxel data, which leads to noteworthy savings in input size and associated network hyperparameters. Last, we explore the accuracy and interpretation of solutions learned via random versus intelligent network weight initialization. Experiments are provided on a U.S. Army data set collected over different times, weather conditions, target types and concealments. Preliminary results indicate that deep learning can perform as good as, if not better, than a skilled domain expert, even in light of limited training data with a class imbalance.

There is interest in imaging ground target threats that are hidden in vegetation, straw grass, and foliage. In some instances, radar signals cannot penetrate through the “clutter.” Long wavelength sound waves might be capable of penetrating through the clutter so that the target can be acoustically detected and imaged. We study sound scattering by a canonical target “model disk” (aluminum, 4 inches diameter x 0.75 inch thick) in the presence of clutter. The clutter is modeled by a vertical hexagonal array of slender stainless steel rods that have an overall diameter D = 8 cm. The individual rods (diameter d = 0.089 cm, length L = 91 cm) are aligned and supported by two perforated thin (thickness = 1.8 mm) circular aluminum plates of diameter = 10 cm. Results for received backscattered tone burst echoes (at a frequency of 25.8 kHz) show that it is possible to detect the details of the target disk in the presence of a twodimensional circularly shaped cluster of rigid cylinders in air (representing clutter). The disk and “model” clutter targets were then taken out of the laboratory environment to an outdoor test site where the High Bandwidth Acoustic Detection System (HBADS) [developed by the US Army’s Night Vision and Electronic Sensors Directorate] performed an acoustic imaging experiment on the target-clutter and mechanical rigging apparatus. Using a linear frequency modulated LFM chirp pulse signal (2-15 kHz) driving a single speaker, echoes are detected by a 16 element microphone array while the HBADS vehicle is traveling ~ 1 m/s along a road. The strip-mapped synthetic aperture acoustic array SAA can image certain features of the apparatus.

High Bandwidth Acoustic Detection System (HBADS) is an emerging active acoustic sensor technology undergoing study by the US Army’s Night Vision and Electronic Sensors Directorate. Mounted on a commercial all-terrain type vehicle, it uses a single source pulse chirp while moving and a new array (two rows each containing eight microphones) mounted horizontally and oriented in a side scan mode. Experiments are performed with this synthetic aperture air acoustic (SAA) array to image canonical ground targets in clutter or foliage. A commercial audio speaker transmits a linear FM chirp having an effective frequency range of 2 kHz to 15 kHz. The system includes an inertial navigation system using two differential GPS antennas, an inertial measurement unit and a wheel coder. A web camera is mounted midway between the two horizontal microphone arrays and a meteorological unit acquires ambient, temperature, pressure and humidity information. A data acquisition system is central to the system’s operation, which is controlled by a laptop computer. Recent experiments include imaging canonical targets located on the ground in a grassy field and similar targets camouflaged by natural vegetation along the side of a road. A recent modification involves implementing SAA stripmap mode interferometry for computing the reflectance of targets placed along the ground. Typical strip map SAA parameters are chirp pulse = 10 or 40 ms, slant range resolution c/(2*BW) = 0.013 m, microphone diameter D = 0.022 m, azimuthal resolution (D/2) = 0.01, air sound speed c ≈ 340 m/s and maximum vehicle speed ≈ 2 m/s.

Full waveform inversion of Ground Penetrating Radar (GPR) data is a promising strategy to estimate quantitative characteristics of the subsurface such as permittivity and conductivity. In this paper, we propose a methodology that uses Full Waveform Inversion (FWI) in time domain of 2D GPR data to obtain highly resolved images of the permittivity and conductivity parameters of the subsurface. 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 velocity model that is updated at each iteration until an acceptable decrease of the cost function is reached. The use of FWI with GPR are expensive computationally because it is based on the computation of the electromagnetic full wave propagation. Also, the commercially available acquisition systems use only one transmitter and one receiver antenna at zero offset, requiring a large number of shots to scan a single line.