This PDF file contains the front matter associated with SPIE Proceedings Volume 8357, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.

Pedemis (PortablE Decoupled Electromagnetic Induction Sensor) is a time-domain handheld electromagnetic induction (EMI) instrument with the intended purpose of improving the detection and classification of UneXploded Ordnance (UXO). Pedemis sports nine coplanar transmitters (the Tx assembly) and nine triaxial receivers held in a fixed geometry with respect to each other (the Rx assembly) but with that Rx assembly physically decoupled from the Tx assembly allowing flexible data acquisition modes and deployment options. The data acquisition (DAQ) electronics consists of the National Instruments (NI) cRIO platform which is much lighter and more energy efficient that prior DAQ platforms. Pedemis has successfully acquired initial data, and inversion of the data acquired during these initial tests has yielded satisfactory polarizabilities of a spherical target. In addition, precise positioning of the Rx assembly has been achieved via position inversion algorithms based solely on the data acquired from the receivers during the "on-time" of the primary field. Pedemis has been designed to be a flexible yet user friendly EMI instrument that can survey, detect and classify targets in a one pass solution. In this paper, the Pedemis instrument is introduced along with its operation protocols, initial data results, and current status.

Current electromagnetic induction (EMI) sensors of the kind used to discriminate buried unexploded orndance (UXO) can detect targets down to a depth limited by the geometric size of the transmitter (Tx) coils, the amplitudes of the transmitting currents, and the noise floor of the receivers (Rx). The last two factors are not independent: for example, one cannot detect a deeply buried target simply by increasing the amplitude of the Tx current, since this also increases the noise and thus does not improve the SNR. The problem could in principle be overcome by increasing the size of the Tx coils and thus their moment. Current multi-transmitter instruments such as the TEMTADS sensor array can be electronically tweaked to provide a big Tx moment: they can be modified to transmit signals from two, three or more Tx coils simultaneously. We investigate the possibility of enhancing the deep-target detection capability of TEMTADS by exploring different combinations of Tx coils. We model different multi-Tx combinations within TEMTADS using a full-3D EMI solver based on the method of auxiliary sources (MAS).We determine the feasibility of honing these combinations for enhanced detection and discrimination of deep targets. We investigate how to improve the spatial resolution and focusing properties of the primary magnetic field by electronically adjusting the currents of the transmitters. We apply our findings to data taken at different UXO live sites.

This paper illustrates the discrimination performance of a set of advanced models at an actual UXO live site. The suite of methods, which combines the orthonormalized volume magnetic source (ONVMS) model, a data-preprocessing technique based on joint diagonalization (JD), and differential evolution (DE) minimization, among others, was tested at the former Camp Beale in California. The data for the study were collected independently by two UXO production teams from Parsons and CH2M HILL using the MetalMapper (MM) sensor in cued mode; each set of data was also processed independently. Initially all data were inverted using a multi-target version of the combined ONVMS-DE algorithm, which provided intrinsic parameters (the total ONVMS amplitudes) that were then used to perform classification after having been inspected by an expert. Classification of the Parsons data was conducted by a Sky Research production team using a fingerprinting approach; analysis of the CH2M HILL data was performed by a Sky/Dartmouth R&D team using unsupervised clustering. During the classification stage the analysts requested the ground truth for selected anomalies typical of the different clusters; this was then used to classify them using a probability function. This paper reviews the data inversion, processing, and discrimination schemes involving the advanced EMI methods and presents the classification results obtained for both the CH2M HILL and the Parsons data. Independent scoring by the Institute for Defense Analyses reveals superb all-around classification performance.

ESTCP live-site UXO classification results are presented for cued data collected with two advanced EMI instruments, the cart-based 2 × 2 3D TEMTADS array and the Man Portable Vector (MPV) handheld sensor, at the former Camp Beale in California. There were two sets of targets of interest (TOI): the main set consisted of 105-mm, 81-mm, 60-mm, 37-mm and ISO projectiles, and the other (optional) set comprised site-specific fuzes and fuze fragments of varous sizes. The advanced models used for inversion and classification combine: 1) a joint-diagonalization (JD) algorithm that estimates the number of potential targets generating an anomaly directly from the measured data without need for inversion; 2) the ortho-normalized volume magnetic source (ONVMS) model, which locates targets, represents their EMI responses, and extracts their intrinsic feature vectors; and 3) a Gaussian mixture algorithm that uses extracted discrimination features to classify the corresponding buried objects as TOI or clutter. Initially the data are inverted using a combination of ONVMS and the differential evolution direct-search algorithm; this allows the determination of relevant intrinsic parameters, which in turn are classified by a mixture of clustering and library-matching techniques. This paper describes in more detail the main steps of the classification process and demonstrates the results obtained for the 2 × 2 3D TEMTADS and MPV data taken at Camp Beale, as scored independently by the Institute for Defense Analyses. The advanced models are seen to produce superb classification in both cases.

ESTCP live-site UXO classification results are presented for cued data collected with two advanced EMI instruments, the cart-based 2 × 2 3D TEMTADS array and the Man Portable Vector (MPV) handheld sensor, at the former Camp Beale in California. There were two sets of targets of interest (TOI): the main set consisted of 105-mm, 81-mm, 60-mm, 37-mm and ISO projectiles, and the other (optional) set comprised site-specific fuzes and fuze fragments of varous sizes. The advanced models used for inversion and classification combine: 1) a joint-diagonalization (JD) algorithm that estimates the number of potential targets generating an anomaly directly from the measured data without need for inversion; 2) the ortho-normalized volume magnetic source (ONVMS) model, which locates targets, represents their EMI responses, and extracts their intrinsic feature vectors; and 3) a Gaussian mixture algorithm that uses extracted discrimination features to classify the corresponding buried objects as TOI or clutter. Initially the data are inverted using a combination of ONVMS and the differential evolution direct-search algorithm; this allows the determination of relevant intrinsic parameters, which in turn are classified by a mixture of clustering and library-matching techniques. This paper describes in more detail the main steps of the classification process and demonstrates the results obtained for the 2 × 2 3D TEMTADS and MPV data taken at Camp Beale, as scored independently by the Institute for Defense Analyses. The advanced models are seen to produce superb classification in both cases.

This paper presents an active source Electromagnetic Induction (EMI) sensor that offers extended detection ranges (> 2m) with minimal sensitivity to magnetic geology. The Ultra Deep Search (ULTRA) EMI system employs a large (20 - 40m), stationary, surface-laid transmitter loop that produces a relatively uniform magnetic field within the search region. This primary field decays slowly with depth due to the non-dipolar nature of the field within the search volume. An array of 3-axis receiver cubes measures the time derivative of secondary field decays produced by subsurface metallic objects. The large-loop transmitter combined with the vector sensing induction coil receivers produces a deep search capability that remains robust in environments containing highly magnetic soils. In this paper, we assess the general detection capabilities of the ULTRA system and present data collected over a set of standardized UXO targets. Additionally, we evaluate the potential for target feature extraction through dipole fit analysis of several data sets.

Frequency-domain electromagnetic induction (EMI) sensors have been shown to provide target signatures which enable discrimination of landmines from harmless clutter. In particular, frequency-domain EMI sensors are well-suited for target characterization by inverting a physics-based signal model. In many model-based signal processing paradigms, the target signatures can be decomposed into a weighted sum of parameterized basis functions, where the basis functions are intrinsic to the target under consideration and the associated weights are a function of the target sensor orientation. When spatial data is available, the diversity of the measured signals may provide more information for estimating the basis function parameters. After model inversion, the basis function parameters can be used as features for classifying the target as landmine or clutter. In this work, feature extraction from spatial frequency-domain EMI sensor data is investigated. Results for data measured with a prototype frequency-domain EMI sensor at a standardized test site are presented. Preliminary results indicate that Structured relevance vector machine (sRVM) regression model inversion using spatial data provides stable, and sparse, sets of target features.

This paper investigates the use of the Single Transmit Multiple Receive (STMR) metal detector (MD) array to estimate the depth of metal targets, such as 155mm shells. The depth estimation problem using MD has been investigated by a number of researchers and the processing was performed along the down-track. The proposed method takes a different approach by exploring the MD responses in cross-track to achieve the depth estimation. It is found that the normalized energy spread of the MD output is narrower for shallow targets and wider for deeper targets. Based on this observation, a method is derived to estimate the depth of a target. Experimental results from the data collected at an U.S. Army test site validate the performance of the proposed depth estimator.

The EMI response of a target can be accurately modeled by a sum of real exponentials. However, it is difficult to obtain the model parameters from measurements when the number of exponentials is unknown. We have previously proposed estimation methods for the model parameters from a single measurement. In this paper, we propose to increase the estimation accuracy by utilizing the multiple measurements often available for a given target. The property of measurements sharing the same relaxation frequencies is exploited to increase the estimation performance. The proposed method is shown to deliver robust estimation using synthetic, laboratory data, and field data.

Joint Orthogonal Matching Pursuits (JOMP) is used here in the context of landmine detection using data obtained from an electromagnetic induction (EMI) sensor. The response from an object containing metal can be decomposed into a discrete spectrum of relaxation frequencies (DSRF) from which we construct a dictionary. A greedy iterative algorithm is proposed for computing successive residuals of a signal by subtracting away the highest matching dictionary element at each step. The nal condence of a particular signal is a combination of the reciprocal of this residual and the mean of the complex component. A two-tap approach comparing signals on opposite sides of the geometric location of the sensor is examined and found to produce better classication. It is found that using only a single pursuit does a comparable job, reducing complexity and allowing for real-time implementation in automated target recognition systems. JOMP is particularly highlighted in comparison with a previous EMI detection algorithm known as String Match.

This work presents recent progress on a system for measuring and characterizing wide-band electromagnetic induction (EMI) target responses. The problems involved in the system design are discussed and a number of measurements of discrete targets and samples of distributed targets are provided. Measurements are mapped to a wide-band model that represents a large amount of data about the EMI response of the target in a very compact form. The issues and techniques involved in this mapping are discussed. When theoretical predictions are possible, measurements show good agreement with theory.

Dictionary matching techniques are an eective way to invert EMI measurements to detect properties of buried targets. A problem arises when trying to create a comprehensive dictionary that can account for location and orientation information for all types of useful targets because populating the dictionary would require the enumeration of parameters in a 7-dimensional space. This paper shows that the discrete spectrum of relaxation frequencies (DSRF) can be used to eliminate two of the dimensions entirely, and the singular value decomposition (SVD) will compress the dictionary by another two orders of magnitude. These dimensionality reducing techniques lead to an ecient dictionary matching algorithm for location and orientation estimates.

The smuggling of bulk cash across borders is a serious issue that has increased in recent years. In an effort to curb the illegal transport of large numbers of paper bills, a detection scheme has been developed, based on the magnetic characteristics of bank notes. The results show that volumes of paper currency can be detected through common concealing materials such as plastics, cardboard, and fabrics making it a possible potential addition to border security methods. The detection scheme holds the potential of also reducing or eliminating false positives caused by metallic materials found in the vicinity, by observing the stark difference in received signals caused by metal and currency. The detection scheme holds the potential to detect for both the presence and number of concealed bulk notes, while maintaining the ability to reduce false positives caused by metal objects.

Metal detectors are used not only to detect but also to locate targets. The location performance has been evaluated previously only in laboratory. The performance probably differs that in the field. In this paper, the evaluation of the location performance based on the analysis of pinpointing error is discussed. The data for the evaluation were collected in a blind test in the field. Therefore, the analyzed performance can be seen as the performance under field conditions. Further, the performance is discussed in relation to the search head and footprint dimensions.

Producing vibration images of buried landmines using a multi-beam laser Doppler vibrometer (MB-LDV) operating from a stationary platform have been accomplished in the past. Detection from a continuously moving platform can reduce the time of detection compared to stop-and-stare measurement. However, there is a speed limitation, imposed by the required spatial and frequency resolution. NCPA proposed a concept of time division multiplexing (TDM) of laser beams of a MB-LDV to overcome that speed limitation. The system, based on 16-beam MB-LDV, has been built and experimentally tested at an Army test facility. Vibration velocity profiles of buried mines have been obtained at different system speeds. Algorithms for speckle noise reduction in continuously moving MB-LDV signals have been developed and explored. The results of the current data collection, recent past data collection as well as the results of the effectiveness of speckle noise reduction techniques are presented.

This work presents a set of measurements collected with a research prototype synthetic aperture acoustic (SAA) imaging system. SAA imaging is an emerging technique that can serve as an inexpensive alternative or logical complement to synthetic aperture radar (SAR). The SAA imaging system uses an acoustic transceiver (speaker and microphone) to project acoustic radiation and record backscatter from a scene. The backscattered acoustic energy is used to generate information about the location, morphology, and mechanical properties of various objects. SAA detection has a potential advantage when compared to SAR in that non-metallic objects are not readily detectable with SAR. To demonstrate basic capability of the approach with non-metallic objects, targets are placed in a simple, featureless scene. Nylon cords of five diameters, ranging from 2 to 15 mm, and a joined pair of 3 mm fiber optic cables are placed in various configurations on flat asphalt that is free of clutter. The measurements were made using a chirp with a bandwidth of 2-15 kHz. The recorded signal is reconstructed to form a two-dimensional image of the distribution of acoustic scatterers within the scene. The goal of this study was to identify basic detectability characteristics for a range of sizes and configurations of non-metallic cord. It is shown that for sufficiently small angles relative to the transceiver path, the SAA approach creates adequate backscatter for detectability.

The U.S. Army Engineer Research and Development Center (ERDC) developed a near-surface computational testbed (CTB) for modeling geo-environments. This modeling capability is used to predict and improve the performance of current and future-force sensor systems for surface and near-surface threat detection for a wide range of geoenvironments. The CTB is a suite of integrated models and tools used to approximately replicate geo-physical processes such as radiometry, meteorology, moisture transport, and thermal transport that influence the resultant signatures of both natural and man-made materials, as perceived by the sensors. The CTB is designed within a High Performance Computing (HPC) framework to accommodate the size and complexity of the virtual environments required for analyzing and quantifying sensor performance. Specifically, as a rule-of-thumb, the size of the scene should encompass an area that is at a minimum, the size of the spatial coverage of the sensor. This HPC capability allows the CTB to replicate geophysical processes and subsurface heterogeneity with high levels of realism and to provide new insight into identifying the geophysical processes and environmental factors that significantly affect the signatures sensed by multispectral imaging, near-infrared, mid-wave infrared, long-wave infrared, and ground penetrating radar sensors. Additionally, this effort is helping to quantify the performance and optimal time-of-use for sensors to detect threats within highly heterogeneous geo-environments by reducing false alarms from automated target recognition algorithms.

The U.S. Army Engineer Research and Development Center (ERDC) has developed a suite of models that replicate the signicant geo-physical processes which aect the thermal signatures sensed by infrared imaging systems. This suite of models also includes an electro-optical/infrared (EO/IR) sensor model that produces synthetic thermal imagery. The EO/IR sensor model can be adapted to replicate the performance of other infrared sensor systems as well. It is well known that eld-collected IR imagery can be in uenced by the micro-topographic features of a particular location. As a result, the performance of automated target recognition algorithms and decisions based on their results can also be aected. Other signicant contributors to false alarms and issues with probabilities-of- detection include the relative locations of vegetation and local changes in soil types or properties. For example, a change in the retention of soil moisture alone is known to contribute to false alarms due to changes in radiative and thermal properties of wet versus dry soil. Many aspects of eld data collection eorts (weather, soil uniformity, etc.) cannot be controlled nor changed after the fact. Within a computational framework, however, plant and object locations, as well as weather patterns can, all be changed. In this work, the sensitivity of simulated IR imagery will be examined as it relates to initial states and boundary forcing terms due to weather conditions. Dierent approaches to these inputs will be examined using the computational testbed developed at the ERDC.

Cloud cover affects direct and diffuse solar radiation and IR downwelling, and the values for these 3 components are calculated using measured meterological data and varying the values of cloud cover and cloud type using algorithms from the literature. The effects of these 3 transient forcing function components on surface, subsurface and target interior temperatures are studied in this work. The cloud cover effects are isolated from the varying multi-day diurnal cycles by repeating the meteorological data for 1 day. Cloud cover is a subgrid variable and hence, is often reported as 0 or 100%. This study includes a comparison of the effects of these two cloud cover values on a single geographical location for 6 days, with each day repeating the meterological conditions of day 1. This work involves using predictions from the Countermine Computational Test Bed (CTB), a 3D finite element model that accounts for coupled heat and moisture transfer in soil and targets.

Rain affects the thermal properties of soil and the temperature of soils and buried targets in the penetration depth of the water. This work involves using predictions from the Countermine Computational Test Bed, (CTB), a 3-D finite element model that accounts for coupled heat and moisture transfer in soil and targets. The meteorological data set used in this work is one day of a meteorological data set, repeated over 3 days. The repeated meteorological data set is required to isolate the effects of rain. The CTB is used to predict and compare surface and subsurface soil and target temperatures with and without rain. The meteorological data set contains 24hrs without rain, followed by 14 hrs of rain at a precipitation rate of 5 mm/hr, then 10 hours plus 1 subsequent day without rain and with no cloud cover.

Schedule optimization of air platforms for IR sensors is a priority because of 1) the time sensitive nature of the IR detection of buried targets, 2) limited air platform assets, and 3) limited bandwidth for live-feed video. Scheduling optimization for airborne IR sensors depends on transient meteorological predictions, transient soil properties, target type and depth. This work involves using predictions from the Weather Research and Forecasting (WRF) model, a regional weather model, as input to the Countermine Computational Test Bed (CTB), a 3D finite element model that accounts for coupled heat and moisture transfer in soil and targets. The result is a continuous 2-day optimized schedule for airborne IR assets. In this paper, a 2-day optimized schedule for an airborne IR sensor asset is demonstrated for a single geographical location with a buried target. Transient physical surface and subsurface soil temperatures are presented as well as the phase-shifted, transient thermal response of the target.

The detection of person-borne threat objects, such as improvised explosive devices, at a safe distance is an ongoing challenge. While much attention has been given to other parts of the electromagnetic spectrum, very little is known about what potential exists to detect clothing obscured threats over the ultraviolet through the shortwave-infrared spectral region. Hyperspectral imaging may provide a greater ability to discriminate between target and non-target by using the full spectrum. This study investigates this potential by the collection and analysis of hyperspectral images of obscured proxy threat objects. The results of this study indicate a consistent ability to detect the presence of concealed objects. The study included the use of VNIR (400 nm to 1000 nm) and SWIR (1000 nm to 1700 nm), as defined here, hyperspectral imagers. Both spectral ranges provided comparable results, however, potential advantages of the SWIR spectral region are discussed.

The main disadvantage of applying the IRT method is presence of plenty false indications in thermograms. A simple use of IRT equipment with better temperature resolution would not help in distinguishing the mines, since noise comes not from a camera, but from soil surface. Recognizing the role of moisture and density of sand and possibilities to express it quantitatively plays an important role. In our model of thermal properties of the soil the volumetric unit of the soil consists of mineral and organic particles, as well as water and air. All needed parameters can be calculated. Calculations of thermal signatures of the underground objects were made basing on 3D-heat equation for the sinus type heating of 3D model and cooling by convection. Measurements were made for field and laboratory stand-ups, using methodologies typical for "single-shot" measurements as well as analyses of transient processes based on sequence of thermograms. Results of simulations and measurements confirm expectation tha that high level of "radiant noises" is caused mainly by differences in the moisture and sand density levels.

We have engaged in research on buried mine/IED detection by remote sensing method using LWIR camera. A IR image of a ground, containing buried objects can be assumed as a superimposed pattern including thermal scattering which may depend on the ground surface roughness, vegetation canopy, and effect of the sun light, and radiation due to various heat interaction caused by differences in specific heat, size, and buried depth of the objects and local temperature of their surrounding environment. In this cumbersome environment, we introduce fractal geometry for analyzing from an IR image. Clutter patterns due to these complex elements have oftentimes low ordered fractal dimension of Hausdorff Dimension. On the other hand, the target patterns have its tendency of obtaining higher ordered fractal dimension in terms of Information Dimension. Random Shuffle Surrogate method or Fourier Transform Surrogate method is used to evaluate fractional statistics by applying shuffle of time sequence data or phase of spectrum. Fractal interpolation to each line scan was also applied to improve the signal processing performance in order to evade zero division and enhance information of data. Some results of target extraction by using relationship between low and high ordered fractal dimension are to be presented.

This paper presents new experimental results from a prototype Spectral LADAR, which combines active multispectral and 3D time-of-flight point cloud imaging. The physical domain unification of these imaging modalities based on a pulse modulated supercontinuum source enables substantially higher fidelity images of obscured targets compared to the data domain fusion of passive hyperspectral cameras and conventional LADAR imagers. Spectral LADAR produces 3D spectral point clouds with unambiguously associated 3D image points and spectral vectors, promoting improved object classification performance in cluttered scenes. The 3D shape and material spectral signature of objects may be acquired in daylight or darkness, behind common glass, and behind obscurants such as foliage and camouflage. These capabilities are demonstrated by data obtained from test scenes. These scenes include plastic mine-like objects obscured by foliage, distinction of hazardous explosives inside plastic containers versus innocuous decoy materials, and 3D spectral imaging behind ordinary glass windows. These scenes, at effective ranges of approximately 40 meters, are imaged with nanosecond-regime optical pulses spanning 1.08 μm to 1.62 μm divided into 25 independently ranged spectral bands. The resultant point cloud is spectrally classified according to material type. In contrast to other active spectral imaging techniques, Spectral LADAR is well suited to operate at high pixel and frame rates and at considerable stand-off distances. A combination of favorable attributes, including eye safe wavelengths, relatively small apertures, and very short (single pulse) receiver integration time, bear the potential for this technique to be used on robotic platforms for on-the-move imaging and high area coverage rates.

To assist the warfighter in visually identifying potentially dangerous roadside objects, the U.S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate (NVESD) has developed an elevated video sensor system testbed for data collection. This system provides color and mid-wave infrared (MWIR) imagery. Signal Innovations Group (SIG) has developed an automated processing capability that detects the road within the sensor field of view and identifies potentially threatening buried objects within the detected road. The road detection algorithm leverages system metadata to project the collected imagery onto a flat ground plane, allowing for more accurate detection of the road as well as the direct specification of realistic physical constraints in the shape of the detected road. Once the road has been detected in an image frame, a buried object detection algorithm is applied to search for threatening objects within the detected road space. The buried object detection algorithm leverages textural and pixel intensity-based features to detect potential anomalies and then classifies them as threatening or non-threatening objects. Both the road detection and the buried object detection algorithms have been developed to facilitate their implementation in real-time in the NVESD system.

In prior work, we describe multiple image space anomaly detection algorithms for the identification of buried explosive materials in forward looking long wave infrared imagery. That work is extended here and focus is placed on improved detection with respect to diurnal temperature variation. An ensemble of shape and size independent image space anomaly detection algorithms are investigated. Specifically, anomalies are identified according to change and blob detection. This anomaly evidence is aggregated and targets are found using an ensemble of trainable size-contrast filters and weighted mean shift clustering. In addition, the blob detector makes use of contrast-limited adaptive histogram equalization for image enhancement. Experimental results are shown based on field data measurements from a U.S. Army test site.

Defence R&D Canada - Suffield and Bubble Technology Industries have been developing thermal neutron activation (TNA) sensors for detection of buried bulk explosives since 1994. First generation sensors, employing an isotopic source and NaI(Tl) gamma ray detectors, were deployed by Canadian Forces in 2002 as confirmation sensors on the ILDS teleoperated, vehicle-mounted, multi-sensor anti-tank landmine detection systems. The first generation TNA could detect anti-tank mines buried 10 cm or less in no more than a minute, but deeper mines and those significantly displaced horizontally required considerably longer times. Mines as deep as 30 cm could be detected with long counting times (1000 s). The second generation TNA detector is being developed with a number of improvements aimed at increasing sensitivity and facilitating ease of operation. Among these are an electronic neutron generator to increase sensitivity for deeper and horizontally displaced explosives; LaBr₃(Ce) scintillators, to improve time response and energy resolution; improved thermal and electronic stability; improved sensor head geometry to minimize spatial response nonuniformity; and more robust data processing. This improved sensitivity can translate to either decreased counting times, decreased minimum detectable explosive quantities, increased maximum sensor-to-target displacement, or a trade off among all three. Experiments to characterize the performance of the latest generation TNA in detecting buried landmines and IEDs hidden in culverts were conducted during 2011. This paper describes the second generation system. The experimental setup and methodology are detailed and preliminary comparisons between the performance of first and second generation systems are presented.

Bulk explosives hidden in culverts pose a serious threat to the Canadian and allied armies. Culverts provide an opportunity to conceal insurgent activity, avoid the need for detectable surface disturbances, and limit the applicability of conventional sub-surface sensing techniques. Further, in spite of the large masses of explosives that can be employed, the large sensor{target separation makes detection of the bulk explosive content challeng- ing. Defence R&D Canada { Sueld and Bubble Technology Industries have been developing thermal neutron activation (TNA) sensors for detection of buried bulk explosives for over 15 years. The next generation TNA sensor, known as TNA2, incorporates a number of improvements that allow for increased sensor-to-target dis- tances, making it potentially feasible to detect large improvised explosive devices (IEDs) in culverts using TNA. Experiments to determine the ability of TNA2 to detect improvised explosive devices in culverts are described, and the resulting signal levels observed for relevant quantities of explosives are presented. Observations conrm that bulk explosives detection using TNA against a culvert-IED is possible, with large charges posing a detection challenge at least as dicult as that of a deeply buried anti-tank landmine. Because of the prototype nature of the TNA sensor used, it is not yet possible to make denitive statements about the absolute sensitivity or detection time. Further investigation is warranted.

Energy dispersive X-ray diffraction (EDXRD) is a technique which can be used to improve the detection and characterisation of explosive materials. This study has performed EDXRD measurements of various explosive compounds using a novel, X-ray sensitive, pixelated, energy resolving detector developed at the Rutherford Appleton Laboratory, UK (RAL). EDXRD measurements are normally performed at a fixed scattering angle, but the 80×80 pixel detector makes it possible to collect both spatially resolved and energy resolved data simultaneously. The detector material used is Cadmium Telluride (CdTe), which can be utilised at room temperature and gives excellent spectral resolution. The setup uses characteristics from both energy dispersive and angular dispersive scattering techniques to optimise specificity and speed. The purpose of the study is to develop X-ray pattern "footprints" of explosive materials based on spatial and energy resolved diffraction data, which can then be used for the identification of such materials hidden inside packages or baggage. The RAL detector is the first energy resolving pixelated detector capable of providing an energy resolution of 1.0-1.5% at energies up to 150 keV. The benefit of using this device in a baggage scanner would be the provision of highly specific signatures to a range of explosive materials. We have measured diffraction profiles of five explosives and other compounds used to make explosive materials. High resolution spectra have been obtained. Results are presented to show the specificity of the technique in finding explosives within baggage.

Nuclear quadrupole resonance is a promising technique for the detection of illicit substances. It relies on the magnetic properties of some specific nuclei, such as nitrogen and chlorine, widely spread among explosives, narcotics or counterfeit medicines. In the basic NQR experiment, the signal (Free Induction Decay (FID)) is generated by a single radio frequency pulse. Because of its small amplitude, the signal is enhanced by averaging several measurements. However, the excitation cannot be repeated until the spin system relaxes back towards equilibrium and this recovery depends on the spin-lattice relaxation time (T₁). This can be sorted out by using multi-pulse sequences. One type of multi-pulse sequence, Steady State Free Precession (SSFP), could be used when the spin-spin relaxation time (T₂) of the compound is of the same order as T₁. It has been claimed that SSFP is a more efficient acquisition sequence than the accumulation of ordinary FIDs. The present study will show, by using simulations and experimental data, that SSFP is a useful sequence for RDX measurements at 5.192 MHz, but is not more effective than a series of well-separated FIDs with a repetition rate lower than 1/T₁.

Nuclear Quadrupole Resonance (NQR) and Nuclear Magnetic Resonance (NMR) are very prospective methods of the bulk detection of explosives and illicit substances. Both methods are based on use of apparatus, which are very similar technically and in some cases could be applied simultaneously. We report our experimental works on NQR/NMR techniques for explosives detection. In addition of classical single-frequency NMR/NQR we also explored a potential of double resonance (NMR/NQR) and multifrequency NQR approaches as well as magnetic resonance imaging (MRI) techniques. Multifrequency (two/three) NQR technique involves various (two or three) transitions in the three energy level system of ¹⁴N nuclei. It is shown that this kind of NQR technique allows filtering spurious signal after radiofrequency pulses and increases the sensitivity of NQR detection. On the other hand, various liquids can be detected using NMR. We shown that reliable discrimination among extended set of liquids reveal a need in use of additional NMR parameters or complimentary techniques. It is demonstrated that MRI is also feasible method for detection of explosive/illicit liquids.

In order to demonstrate the possibility of Ground Penetrating Radar (GPR) for detection of small buried objects such as landmine and UXO, conducted demonstration tests by using the 3DGPR system, which is a GPR system combined with high accuracy positing system using a commercial laser positioning system (iGPS). iGPS can provide absolute and better than centimetre precise x,y,z coordinates to multiple mine sensors at the same time. The developed " 3DGPR" system is efficient and capable of high-resolution 3D shallow subsurface scanning of larger areas (25 m² to thousands of square meters) with irregular topography . Field test by using a 500MHz GPR system equipped with 3DGPR system was conducted. PMN-2 and Type-72 mine models have been buried at the depth of 5-20cm in sand. We could demonstrate that the 3DGPR can visualize each of these buried land mines very clearly.

Ground Penetrating Radar (GPR) has been applied for several years to the problem of detecting both anti-personnel and anti-tank landmines. One major challenge for reliable mine detection using GPR is removing the response from the ground. When the ground is flat this is a straightforward process. For the NIITEK GPR, the flat ground will show up as one of the largest responses and will be consistent across all the channels, making the surface simple to detect and remove. Typically, the largest responses from each channel, assumed to be the surface, are aligned in range and then zeroed out. When the ground is not flat, the response from the ground becomes more complicated making it no longer possible to just assume the largest response is from the ground. Also, certain soil surface features can create responses that look very similar to those of mines. To further complicate the ground removal process, the motion of the GPR antenna is not measured, making it impossible to determine if the ground or antenna is moving from just the GPR data. To address surface clutter issues arising from uneven ground, NVESD investigated profiling the soil surface with a LIDAR. The motion of both the LIDAR and GPR was tracked so the relative locations could be determined. Using the LIDAR soil surface profile, GPR data was modeled using a simplified version of the Physical Optics model. This modeled data could then be subtracted from the measured GPR data, leaving the response without the soil surface. In this paper we present a description and results from an experiment conducted with a NIITEK GPR and LIDAR over surface features and buried landmines. A description of the model used to generate the GPR response from the soil and the algorithm that was used to subtract the two provided. Mine detection performances using both GPR only and GPR with LIDAR algorithms are compared.

Ground Penetrating Radar (GPR) has been applied for several years to the problem of detecting both anti-personnel and anti-tank landmines. One area of research is using Forward Looking GPR (FLGPR) to detect mines. While FLGPR has the advantage of standoff versus downward looking GPR, the responses from buried targets generally decrease while the responses from clutter increase. One source of clutter is from sidelobes and grating lobes caused by off-road clutter. As it is not possible to get a narrow beamwidth at the low frequencies required to get ground penetration, FLGPR receives responses from both on and off the road. Off-road clutter responses are often much stronger than the responses from buried mines. These off-road clutter objects can produce sidelobes that overlap with and obscure the responses from inroad targets. This becomes especially problematic if the antenna array spacing is not fine enough and grating lobes are formed. To reduce both the sidelobes and grating lobes, a technique using L1-norm minimization was tested. One advantage of this technique is it only requires a single aperture. The resulting image retains phase information which allows the images to be then coherently summed, resulting in better quality images. In this paper a description of the algorithm is provided. The algorithm was applied to a FLGPR data set to show its ability to reduce both sidelobes and grating lobes. Resulting images are shown.

Standoff detection of mines and improvised explosive devices by ground penetrating radar has advantages in terms of safety and efficiency. However, the reflected signals from buried targets are often disturbed by those from the ground surface, which vary with the antennas angle, making it more difficult to detect at a safe distance. An understanding of the forward and backward scattering wave is thus essential for improving standoff detection capability. We present some experimental results from using our measurement system for such an analysis.

Improvised Explosive Devices (IEDs) triggered by pressure-plates are a serious threat in current theatres of operation. X-ray backscatter imaging (XBI) is a potential method for detecting buried pressure-plates. Monte-Carlo simulation code was developed in-house and has been used to study the potential of XBI for pressure-plate detection. It is shown that pressure-plates can be detected at depths up to 7 cm with high photon energies of 350 keV with reasonable speeds of 1 to 10 km/h. However, spatial resolution is relatively low due to multiple scattering.

In recent years NVESD has been investigating laser-based neutralization of buried mines and minelike targets. This paper covers the most recent efforts in this area. A field-test was conducted to demonstrate the state-of-the-art capability for standoff laser neutralization of surface and buried mines. The neutralization laser is a Ytterbium fiber laser with a nominal power output of 10 kW and a beam quality of M² ≈ 1.8 at maximum power. Test trials were conducted at a standoff range of 50 meters with a 20° angle of attack. The laser was focused to a submillimeter spot using a Cassegrain telescope with a 12.5 inch diameter primary mirror. The targets were 105 mm artillery rounds with a composition B explosive fill. Three types of overburden were studied: sand, soil, and gravel. Laser neutralization capability was demonstrated under these conditions for live rounds buried under 7 cm of dry sand, 4 cm of soil, and 2 cm of gravel.

Dual sensor is one of the most promising sensors for humanitarian demining operations. Conventional landmine detection depends on highly trained and focused human operators manually sweeping 1m² plots with a metal detector and listening for characteristic audio signals indicating the presence of AP (Anti-personnel) landmines. In order to reduce the time of plodding detected objects, metal detectors need to be combined with a complimentary subsurface imaging sensor. i.e., GPR(Ground Penetrating Radar). The demining application requires real-time imaging results with centimetre resolution in a highly portable package. We are currently testing a dual sensor ALIS which is a real-time sensor tracking system based on a CCD camera and image processing. In this paper we introduce ALIS systems which we have developed for detection of buried antipersonnel mines and small size explosives. The performance of ALIS has been tested in Cambodia since 2009. More than 80 anti-personnel mines have been detected and removed from local agricultural area. ALIS has cleared more than 70,000 m² area and returned it to local farmers.

Ground penetrating radar (GPR) is a commonly employed sensing modality for landmine detection. It has been successfully deployed in vehicular systems, and is also being integrated into handheld systems. Handheld mine detection systems are typically deployed in situations where either the terrain or mission renders a vehicular-based system less effective. Handheld systems are often more compact and maneuverable, but quality of the sensor data may also be more dependent on the operators experience with and technique in using the system. In particular, the sensor height with respect to the air-ground interface may be more variable than with a vehicular-based system. This variation in sensor height above the air-ground interface may have the potential to adversely affect mine detection performance with the GPR sensing modality. In this work, the effects of operator technique on handheld sensor data quality is investigated, and ground alignment is explored as a potential approach to reducing variability in the sensor data quality due to operator technique. Results for data measured with a standard GPR/EMI handheld sensor at a standardized test site are presented.

We provide an evaluation of spectral features extracted from the signal return of a forward-looking ground penetrating radar to improve the detection performance of buried explosive hazards. The evaluations are performed on data collected at two different lanes at a government test site. The performance of the one-dimensional (1D), two-dimensional (2D) and multiple (ML) spectral features will be contrasted through lane-based cross-validation for training and testing. Additional features to characterize the spectral behaviors of the forward-looking radar return will also be examined.

This paper proposes an effective anomaly detection algorithm for forward-looking ground-penetrating radar (FLGPR). The challenges in detecting explosive hazards with FLGPR are that there are multiple types of targets buried at different depths in a highly-cluttered environment. A wide array of target and clutter signatures exist, which makes classifier design difficult. Recent work in this application has focused on fusing the classifier results from multiple frequency subband images. Each sub-band classifier is trained on suites of image features, such as histogram of oriented gradients (HOG) and local binary patterns (LBP). This prior work fused the sub-band classifiers by, first, choosing the top-ranked feature at each frequency sub-band in the training data and then accumulating the sub-band results in a confidence map. We extend this idea by employing multiple kernel learning (MKL) for feature-level fusion. MKL fuses multiple sources of information and/or kernels by learning the weights of a convex combination of kernel matrices. With this method, we are able to utilize an entire suite of features for anomaly detection, not just the top-ranked feature. Using FLGPR data collected at a US Army test site, we show that classifiers trained using MKL show better explosive hazard detection capabilities than single-kernel methods.

Improvements to an automatic detection system for locating buried explosive hazards in forward-looking longwave infrared (FL-LWIR) imagery, as well as the system's application to detection in confidence maps and forwardlooking ground penetrating radar (FL-GPR) data, are discussed. The detection system, described in previous work, utilizes an ensemble of trainable size-contrast filters and the mean-shift algorithm in Universal Transverse Mercator (UTM) coordinates. Improvements of the raw detection algorithm include weighted mean-shift within the individual size-contrast filters and a secondary classification step which exacts cell structured image space features, including local binary patterns (LBP), histogram of oriented gradients (HOG), edge histogram descriptor (EHD), and maximally stable extremal regions (MSER) segmentation based shape information, from one or more looks and classifies the resulting feature vector using a support vector machine (SVM). FL-LWIR specific improvements include elimination of the need for multiple models due to diurnal temperature variation. The improved algorithm is assessed on FL-LWIR and FL-GPR data from recent collections at a US Army test site.

The UHF band in SAR has foliage penetration and limited ground penetration capability, while LIDAR scans are capable of providing elevation information of objects on the terrain. In this paper, we integrate the complementary strengths of these two different classes of sensors to locate buried objects with improved precision. The main underlying concept is that the buried targets are discernible only in UHF-SAR space while LIDAR is rich with above-ground False Alarm information. The LIDAR elevation information at the changes and anomalies are exploited to rule out above-ground false-alarms in the UHF-SAR domain, thereby isolating the buried IEDs. Definitive proof-of-concept validation is given for same-day/single-pass buried object detection capability using single-pass SAR anomaly detection with LIDAR fusion. We also demonstrate significant performance improvement with 2-pass SAR change detection with LIDAR integration. Detection performance is further enhanced via exploitation of multiple polarizations and multiple passes for SAR data. The proposed SAR-LIDAR fusion strategy is shown to detect emplaced buried objects with an order of magnitude improvement in detection performance, i.e., achieve higher PD at lower PFA when compared with SAR-only performance. The proof-of-concept research is demonstrated on simultaneous multisensor UHF-SAR/LIDAR data collected under JIEDDO's HALITE-1 program.

Based on the lab-on-a-fiber (LOF) concept we proposed before, we further optimize its architecture while preserving its capability in fluorescent signal collection and excitation stray light rejection. This LOF device is a short fiber taper with a TNT sensory film overlay at one end of a 400 μm core fiber which is approximately 50 mm long. The optimized LOF also lowers the system cost, eases the fiber replacement and maintenance, which are enabled by a reusable 3-leg bifurcated fiber bundle with SMA connectors to connect LOF, an excitation light source and a spectrometer. This LOF device occupies only a Φ0.4 mm × 1 mm space.

Multispectral imaging Raman spectroscopy is a novel technique for detecting and identifying explosive residues, e.g. explosives particles which are left on surfaces after handling or manufacturing of explosives. By imaging a suspect surface using the imaging Raman technique, explosives particles at stand-off distances can be identified and displayed using color coding1. In this paper we present an attempt to determine a limit of detection for imaging Raman spectroscopy by analyzing holes of various sizes in aluminum plates filled with four different substances; 2,4-dinitrotoulene (DNT), ammonium nitrate (AN), sulfur, and 2,4,6-trinitrotoulene (TNT). The detection time in the presented experiments has not been optimized, instead more effort has been invested in order to reduce false alarms. The detection system used is equipped with a green second harmonic Nd:YAG laser with an average power of 2 W, a 200 mm telescope and a liquid crystal tunable filter to scan the wavenumbers. The distance to the target was 10 m and the imaged area was 28 mm × 28 mm. The measured multi-spectral data cubes were evaluated using least square fitting to distinguish between DNT, AN,S, TNT and the background. The detection limit has been determined to be sub microgram using the current setup.

This paper presents the ongoing development of a laser ionization mass spectrometric system to be applied for screening for security related threat substances, specifically explosives. The system will be part of a larger security checkpoint system developed and demonstrated within the FP7 project EFFISEC to aid border police and customs at outer border checks. The laser ionization method of choice is SPI (single photon ionization), but the system also incorporates optional functionalities such as a cold trap and/or a particle concentrator to facilitate detection of minute amounts of explosives. The possibility of using jet-REMPI as a verification means is being scrutinized. Automated functionality and user friendliness is also considered in the demo system development.

Changes in the fluorescence of semiconductor nanocrystals were explored as a potential sensing mechanism for the detection of chemicals associated with landmines, IEDs and HME materials. A series of quantum dots (QDs) with fluorescence emissions spanning the visible spectrum was investigated using the Stern-Volmer relationship, specifically measuring the effect of quencher concentration on QD fluorescence intensity and photo-excited lifetime. The series of QDs was investigated with respect to their ability to donate excited-state electrons to an electronwithdrawing explosive related compound (ERC). Electron transfer was monitored by observing the steady-state fluorescence signal and the excited-state lifetimes of the QDs in the presence of ERC1. Increased sensitivities of QDs towards ERC1 were observed as the size and emission wavelength of the QDs decreased. As the QDs size decreased, the Stern-Volmer quenching constants increased. The larger QD exhibited the lowest Ksv and is thought to be quenched by a purely static quenching mechanism. As QD size decreased, an additional collisional quenching mechanism was introduced, denoted by a non-linearity in the quenching-vs-concentration Stern-Volmer plot. Increases in quenching efficiency were due to increased excited-state lifetimes, and the introduction of a collisional quenching mechanism. The quenching constant for the smallest QD was approximately an order of magnitude higher than those of similarly evaluated commercially available fluorescent polymers, suggesting that QDs could be exploited to develop sensitive detectors for electron-withdrawing compounds such as nitroaromatics.

This paper details the use of a genetic algorithm (GA) as a method to preselect spectral feature variables for chemometric algorithms, using spectroscopic data gathered on explosive threat targets. The GA was applied to laserinduced breakdown spectroscopy (LIBS) and ultraviolet Raman spectroscopy (UVRS) data, in which the spectra consisted of approximately 10000 and 1000 distinct spectral values, respectively. The GA-selected variables were examined using two chemometric techniques: multi-class linear discriminant analysis (LDA) and support vector machines (SVM), and the performance from LDA and SVM was fed back to the GA through a fitness function evaluation. In each case, an optimal selection of features was achieved within 20 generations of the GA, with few improvements thereafter. The GA selected chemically significant signatures, such as oxygen and hydron peaks from LIBS spectra and characteristic Raman shifts for AN, TNT, and PETN. Successes documented herein suggest that this GA approach could be useful in analyzing spectroscopic data in complex environments, where the discriminating features of desired targets are not yet fully understood.

During foreign operations, Improvised Explosive Devices (IEDs) are one of major threats that soldiers may unfortunately encounter along itineraries. Based on a vehicle-mounted camera, we propose an original approach by image comparison to detect signicant changes on these roads. The classic 2D-image registration techniques do not take into account parallax phenomena. The consequence is that the misregistration errors could be detected as changes. According to stereovision principles, our automatic method compares intensity proles along corresponding epipolar lines by extrema matching. An adaptive space warping compensates scale dierence in 3D-scene. When the signals are matched, the signal dierence highlights changes which are marked in current video.

Roadside explosive threats continue to pose a significant risk to soldiers and civilians in conflict areas around the world. These objects are easy to manufacture and procure, but due to their ad hoc nature, they are difficult to reliably detect using standard sensing technologies. Although large roadside explosive hazards may be difficult to conceal in rural environments, urban settings provide a much more complicated background where seemingly innocuous objects (e.g., piles of trash, roadside debris) may be used to obscure threats. Since direct detection of all innocuous objects would flag too many objects to be of use, techniques must be employed to reduce the number of alarms generated and highlight only a limited subset of possibly threatening regions for the user. In this work, change detection techniques are used to reduce false alarm rates and increase detection capabilities for possible threat identification in urban environments. The proposed model leverages data from multiple video streams collected over the same regions by first applying video aligning and then using various distance metrics to detect changes based on image keypoints in the video streams. Data collected at an urban warfare simulation range at an Eastern US test site was used to evaluate the proposed approach, and significant reductions in false alarm rates compared to simpler techniques are illustrated.

The detection and matching of robust features in images is an important step in many computer vision applications. In this paper, the importance of the keypoint detection algorithms and their inherent parameters in the particular context of an image-based change detection system for IED detection is studied. Through extensive application-oriented experiments, we draw an evaluation and comparison of the most popular feature detectors proposed by the computer vision community. We analyze how to automatically adjust these algorithms to changing imaging conditions and suggest improvements in order to achieve more exibility and robustness in their practical implementation.

Many effective buried threat detection systems rely on close proximity and near vertical deployment over subsurface objects before reasonable performance can be obtained. A forward-looking sensor configuration, where an object can be detected from much greater distances, allows for safer detection of buried explosive threats, and increased rates of advance. Forward-looking configurations also provide an additional advantage of yielding multiple perspectives and looks at each subsurface area, and data from these multiple pose angles can be potentially exploited for improved detection. This work investigates several aspects of detection algorithms that can be applied to forward-looking imagery. Previous forward-looking detection algorithms have employed several anomaly detection algorithms, such as the RX algorithm. In this work the performance of the RX algorithm is compared to a scale-space approach based on Laplcaian of Gaussian filtering. This work also investigates methods to combine the detection output from successive frames to aid detection performance. This is done by exploiting the spatial colocation of detection alarms after they are mapped from image coordinates into world coordinates. The performance of the resulting algorithms are measured on data from a forward-looking vehicle mounted optical sensor system collected over several lanes at a western U.S. test facility. Results indicate that exploiting the spatial colocation of detections made in successive frames can yield improved performance.

An ROV constructed for the inspection of objects lying on the coastal sea floor has been described. In order to establish if an object on the sea floor contains some sort of threat material (explosives, chemical agent), a system using a neutron sensor installed within ROV has been developed. We describe the maritime properties of ROV and show that the measured gamma spectra for commonly found ammunition charged with TNT explosives are dominated by C, O and Fe peaks enabling the determination of the presence of explosives inside an ammunition shell.

Automatic detection of oating mines by passive sensing is of major interest, yet remains a hard problem. In this paper, we propose an algorithm to detect them in infrared sequences, based on their geometry, provided by spatial derivatives. In infrared images, oating mines contrast with the sea due to the dierence of emissivity at low incidence angles: they form bright elliptical areas. Using the available data and the geometry of our camera, we rst determine the scales of interest, which represent the possible size of mines in number of pixels. Then, we use a temporal and a morphological lter to perform smoothing in the time dimension and contrast enhancement in the space dimensions, at the selected scales, and calculate for every pixel the Hessian matrix, composed of the second order derivatives, which are estimated in the classical scale-space framework, by convolving the image with derivatives of Gaussian. Based on the eigenvalues of the Hessian matrix, representing the curvatures along the principal directions of the image, we dene two parameters describing the eccentricity of an elliptical area and the contrast with sea, and propose a measure of mine-likeliness" that will be high for bright elliptical regions with selected eccentricy. At the end, we only retain pixels with high mine-likeliness, stable in time, as potential mines. Using a dataset of 10 sequences with ground truth, we evaluated the performance and stability of our algorithm, and obtained a precision between 80% and 100%, and a per-frame recall between 30% and 100%, depending on the diculty of the scenarios.

When using flat windows between an air medium and one with higher index of refraction, the surface becomes optically active and a number of aberrations are induced. One affecting the optical control of a remotely-piloted underwater vehicle is the apparent pincushion distortion resulting from Snell's law at the interface. Small wide-angle lenses typically have the opposite problem, a barrel distortion caused by limitations in the number of lens surfaces and the constraints of cost. An experimental calibration is described in which the barrel distortion of the lens compensated for most of the inherent pincushion of the change in medium. ZEMAX^TM models will be used to elucidate this phenomenon with a published lens design.* With careful selection of the lens and additional corrector, the resultant image can be made almost rectilinear, thus easing steering control and automatic target recognition.

This paper presents a discussion of U.S. naval mine countermeasures (MCM) theory modernization in light of advances in the areas of autonomy, tactics, and sensor processing. The unifying theme spanning these research areas concerns the capability for in situ adaptation of processing algorithms, plans, and vehicle behaviors enabled through run-time situation assessment and performance estimation. Independently, each of these technology developments impact the MCM Measures of Effectiveness¹ [MOE(s)] of time and risk by improving one or more associated Measures of Performance² [MOP(s)]; the contribution of this paper is to outline an integrated strategy for realizing the cumulative benefits of these technology enablers to the United States Navy's minehunting capability. An introduction to the MCM problem is provided to frame the importance of the foundational research and the ramifications of the proposed strategy on the MIW community. We then include an overview of current and future adaptive capability research in the aforementioned areas, highlighting a departure from the existing rigid assumption-based approaches while identifying anticipated technology acceptance issues. Consequently, the paper describes an incremental strategy for transitioning from the current minehunting paradigm where tactical decision aids rely on a priori intelligence and there is little to no in situ adaptation or feedback to a future vision where unmanned systems³, equipped with a representation of the commander's intent, are afforded the authority and ability to adapt to environmental perturbations with minimal human-in-the-loop supervision. The discussion concludes with an articulation of the science and technology issues which the MCM research community must continue to address.

A sensor system has been constructed that is capable of detecting and discriminating between various explosives presented in ocean water with detection limits at the 10-100 parts per trillion level. The sensor discriminates between different compounds using a biologically-inspired fluorescent polymer sensor array, which responds with a unique fluorescence quenching pattern during exposure to different analytes. The sensor array was made from commercially available fluorescent polymers coated onto glass beads, and was demonstrated to discriminate between different electron-withdrawing analytes delivered in salt water solutions, including the explosives 2,4,6-trinitrotoluene (TNT) and tetryl, the explosive hydrolysis products 2-amino-4,6-dinitrotoluene and 4-amino-2,6-dinitrotoluene, as well as other explosive-related compounds and explosive simulants. Sensitivities of 10-100 parts per trillion were achieved by employing a preconcentrator (PC) upstream of the sensor inlet. The PC consists of the porous polymer Tenax, which captures explosives from contaminated water as it passes through the PC. As the concentration of explosives in water decreased, longer loading times were required to concentrate a detectable amount of explosives within the PC. Explosives accumulated within the PC were released to the sensor array by heating the PC to 190 C. This approach yielded preconcentration factors of up to 100-1000x, however this increased sensitivity towards lower concentrations of explosives was achieved at the expense of proportionally longer sampling times. Strategies for decreasing this sampling time are discussed.

Vehicle-mounted ground-penetrating radar (GPR) has proved to be a valuable technology for buried threat detection, especially in the area of military route clearance. However, detection performance may be degraded in very rough terrain or o-road conditions. This is because the signal processing approaches for target detection in GPR rst identify the ground re ection in the data, and then align the data in order to remove the ground re ection. Under extremely rough terrain, antenna bounce and multipath eects render nding the ground re ection a dicult task, and errors in ground localization can lead to data alignment that distorts potential target signatures and/or creates false alarms. In this work, commercial-o-the-shelf light detection and ranging (LIDAR), global positioning system (GPS), and inertial measurement unit (IMU) were integrated with a GPR into a prototype route clearance system. The LIDAR provided high-resolution measurements of the ground surface prole, and the GPS/IMU recorded the vehicle's position and orientation. Experiments investigated the applicability of the integrated system for nding the ground re ection in GPR data and decoupling vehicle motion from the rough surface response. Assessment of ground-tracking performance was based on an experiment involving three prepared test lanes, each with dierent congurations of buried targets and terrain obstacles. Several algorithms for target detection in GPR were applied to the data, both with traditional preprocessing and incorporating the LIDAR and IMU. Experimental results suggest that the LIDAR and IMU may be valuable components for ground tracking in next-generation GPR systems.

Ground penetrating radar (GPR) systems have been successfully developed and deployed for buried object detection in real-world environments. Currently, the detection algorithms using GPR assume a flat earth model and incoming data is manipulated so the air ground interface conforms to this requirement. This manipulation of data occasionally results in missed detections and false alarms. Knowing the exact location of the air-ground interface could be used to relax the flat earth assumption and alleviate some of the associated detection problems. This paper describes a method of using the Microsoft Kinect sensor to track the ground for use in the buried object detection algorithms.

When processing ground penetrating radar (GPR) data for the detection of subsurface objects it is common to align the data based on the location of the air-ground interface in order to eliminate the effects of antenna motion. This practice assumes that the ground is mostly flat and that variations in the measured ground locations are primarily due to antenna motion. In practice this assumption is often false so ground alignment will cause true ground contours to be flattened, potentially distorting signatures from subsurface objects. In this paper we investigate extracting edge histogram descriptor (EHD) features from GPR data with varying degrees of alignment: unaligned, fully aligned and aligned only in the downtrack direction, where the effects of antenna motion are most prevalent. One problem with not performing ground alignment is that features generated from the ground surface or subsurface layers that follow the contour of the ground may cause false alarms. To address this problem we also consider employing background subtraction prior to feature extraction on aligned data, independent of the alignment method used for feature extraction. We compare the detection performance of algorithms using each of these feature extraction approaches.

In landmine detection using vehicle-mounted ground-penetrating radar (GPR) systems, ground tracking has proven to be an eective pre-processing step. Identifying the ground can aid in the correction of distortions in downtrack radar data, which can result in the reduction of false alarms due to ground anomalies. However, the air-ground interface is not the only layer boundary detectable by GPR systems. Multiple layers can exist within the ground, and these layers are of particular importance because they give rise to anomalous signatures below the ground surface, where target signatures will typically reside. In this paper, an ecient method is proposed for performing multiple ground layer-identication in GPR data. The method is an extension of the dynamic programming-based Viterbi algorithm, nding not only the globally optimal path, which can be associated with the ground surface, but also locally optimal paths that can be associated with distinct layer boundaries within the ground. In contrast with the Viterbi algorithm, this extended method is uniquely suited to detecting not only multiple layers that span the entire antenna array, but also layers that span only a subset of the channels of the array. Furthermore, it is able to accomplish this while retaining the ecient nature of the original Viterbi scheme.

The intent of this research is to align and compare an array of GPR (Ground Penetrating Radar) data with historical data taken over similar pathways, separated in time, with soft positioning accuracies. The objective is to develop an overlap of the two or more data runs to reduce the false alarm load on the operator and automatically reveal new alarms showing up in the new data (change detection). Data is taken with a GPR system. GPS (Global Positioning System) coordinates are stamped on the data but are very inaccurate, therein rests the registration problem. Two approaches have been taken to align the data sets: 1) Alarm registration through 2D correlation methods. 2) Image registration of ground contours using 2D correlation methods. Radar data are displayed and analyzed within the context of the above algorithms. Data displays are shown in 2D formats, with alarm registration displaying cross track vs. down track and ground contour registration displaying down track vs. ground depth. Preliminary results indicate a positive benefit from these registration processes, including: Rapid detection of new alarms. Reduction of the overall FAR (False Alarm Rate) load on the operator. These processes have application to the support of radar systems in operational scenarios by decreasing the load on operators and producing a more rapid ROA (Rate of Advance).

Image keypoints are widely used in computer vision for object matching and recognition, where they provide the best solution for matching and instance recognition of complex objects within cluttered images. Most matching algorithms operate by rst nding interest points, or keypoints, that are expected to be common across multiple views of the same object. A small area, or patch, around each keypoint can be represented by a numerical descriptor that describes the structure of the patch. By matching descriptors from keypoints found in 2-D data to keypoints of known origin, matching algorithms can determine the likelihood that any particular patch matches a pre-existing template. The objective in this research is to apply these methods to two-dimensional slices of Ground Penetrating Radar (GPR) data in order to distinguish between landmine and non-landmine responses. In this work, a variety of established object matching algorithms have been tested and evaluated to examine their application to GPR data. In addition, GPR specic keypoint and descriptor methods have been developed which better suit the landmine detection task within GPR data. These methods improve on the performance of standard image processing techniques, and show promise for future work involving translations of technologies from the computer vision eld to landmine detection in GPR data.

Unprocessed ground penetrating radar (GPR) imagery often suffers from horizontal background striations owing to internal system noise and/or ground layers. These striations adversely affect the ability to identify buried objects, either via visual inspection of the imagery or by automatic target detection techniques. Singular value decomposition (SVD) is one of the most common techniques for removing these background striations, but it is hindered in real-time implementations due to its computational overhead. This paper proposes and demonstrates an alternative technique. The resulting background removal process based on weighted principal component analysis runs faster, preserves more of the target information, and removes a greater percentage of the background compared to standard SVD-based techniques.

Ground Penetrating Radar (GPR) has been extensively employed as a technology for the detection of subsurface buried threats. Although vehicular mounted GPRs generate data in three dimensions, alarm declarations are usually only available in the form of 2-D spatial coordinates. The uncertainty in the depth of the target in the three dimensional volume of data, and the difficulties associated with automatically localizing objects in depth, can adversely impact feature extraction and training in some detection algorithms. In order to mitigate the negative impact of uncertainty in target depth, several algorithms have been developed that extract features from multiple depth regions and utilize these feature vectors in classification algorithms to perform final mine/nonmine decisions. However, the uncertainty in object depth significantly complicates learning since features at the correct target depth are often significantly different from features at other depths but in the same volume. Multiple Instance Learning (MIL) is a type of supervised learning approach in which labels are available for a collection of feature vectors but not for individual samples, or in this application, depths. The goal of MIL is to classify new collections of vectors as they become available. This set-based learning method is applicable in the landmine detection problem because features that are extracted independently from several depth bins can be viewed as a set of unlabeled feature vectors, where the entire set either corresponds to a buried threat or a false alarm. In this work, a novel generative Dirichlet Process Gaussian mixture model for MIL is developed that automatically infers the number of mixture components required to model the underlying distributions of mine/non-mine signatures and performs classification using a likelihood ratio test. In this work, we show that the performance of the proposed approach for discriminating targets from non-targets in GPR data is promising.

The Husky Mine Detection System (HMDS) detects and alerts operators to potential threats observed in groundpenetrating RADAR (GPR) data. In the current system architecture, the classifiers have been trained using available data from multiple training sites. Changes in target types, clutter types, and operational conditions may result in statistical differences between the training data and the testing data for the underlying features used by the classifier, potentially resulting in an increased false alarm rate or a lower probability of detection for the system. In the current mode of operation, the automated detection system alerts the human operator when a target-like object is detected. The operator then uses data visualization software, contextual information, and human intuition to decide whether the alarm presented is an actual target or a false alarm. When the statistics of the training data and the testing data are mismatched, the automated detection system can overwhelm the analyst with an excessive number of false alarms. This is evident in the performance of and the data collected from deployed systems. This work demonstrates that analyst feedback can be successfully used to re-train a classifier to account for variable testing data statistics not originally captured in the initial training data.

Ground-penetrating radar (GPR) is a very useful technology for buried threat detection applications which is capable of identifying both metallic and non-metallic objects with moderate false alarm rates. Several pattern classication algorithms have been proposed and evaluated which enable GPR systems to achieve robust per- formance. However, comparisons of these algorithms have shown that their relative performance varies with respect to the environmental context under which the GPR is operating. Context-dependent fusion has been proposed as a technique for algorithm fusion and has been shown to improve performance by exploiting the dierences in algorithm performance under dierent environmental and operating conditions. Early approaches to context-dependent fusion clustered observations in the joint condence space of all algorithms and applied fusion rules within each cluster (i.e., discriminative learning). Later approaches exploited physics-based fea- tures extracted from the background data to leverage more environmental information, but decoupled context learning from algorithm fusion (i.e., generative learning). In this work, a Bayesian inference technique which combines the generative and discriminative approaches is proposed for physics-based context-dependent fusion of detection algorithms for GPR. The method uses a Dirichlet process (DP) mixture as a model for context, and relevance vector machines (RVMs) as models for algorithm fusion. Variational Bayes is used as an approximate inference technique for joint learning of the context and fusion models. Experimental results compare the pro- posed Bayesian discriminative technique to generative techniques developed in past work by investigating the similarities and dierences in the contexts learned as well as overall detection performance.

Prony's Method with a Polynomial Model (PMPM) is a novel way of doing classification. Given a number of training samples with features and labels, it assumes a Gaussian mixture model for each feature, and uses Prony's method to determine a method of moments solution for the means and priors of the Gaussian distributions in the Gaussian mixture model. The features are then sorted in descending order by their relative performance. Based on the Gaussian mixture model of the first feature, training samples are partitioned into clusters by determining which Gaussian distribution each training sample is most likely from. Then with the training samples in each cluster, a new Gaussian mixture model is built for the next most powerful feature. This process repeats until a Gaussian mixture model is built for each feature, and a tree is thus grown with the training data partitioned into several final clusters. A "leaf" model for each final cluster is the weighted least squares solution (regression) for approximating a polynomial function of the features to the truth labels. Testing consists of determining for each testing sample a likelihood that the testing sample belongs to each cluster, and then regressions are weighted by their likelihoods and averaged to produce the test confidence. Evaluation of PMPM is done by extracting features from data collected by both Ground Penetrating Radar and Metal Detector of a robot-mounted land-mine detection system, training PMPM models, and testing in a cross-validation fashion.

Characterization and removal of unwanted artifacts in ground penetrating radar (GPR) imagery is a non-trivial task. The many factors affecting the presence, magnitude, and duration of such artifacts include their origin (man-made or naturally occurring), location (above ground or in-ground), dielectric constant, and moisture content, to name a few. It is of significant benefit to anomaly detection systems to remove such artifacts to reduce false alarm rates and increase threat alarm confidences. Man-made artifacts are typically a result of secondary reflections from the radar emitting surface and its related hardware. These "self-signatures" are manifested as artifacts below the ground surface that tend to be visible for all scans. However, when the sensor height is not held constant above the ground, the position (in time) and magnitude of the reflections become variable and difficult to predict. Naturally occurring artifacts include ground layers, sub-surface water layers, etc.

Hidden Markov Models (HMM) have proved to be eective for detecting buried land mines using data collected by a moving-vehicle-mounted ground penetrating radar (GPR). The general framework for a HMM-based landmine detector consists of building a HMM model for mine signatures and a HMM model for clutter signatures. A test alarm is assigned a condence proportional to the probability of that alarm being generated by the mine model and inversely proportional to its probability in the clutter model. The HMM models are built based on features extracted from GPR training signatures. These features are expected to capture the salient properties of the 3-dimensional alarms in a compact representation. The baseline HMM framework for landmine detection is based on gradient features. It models the time varying behavior of GPR signals, encoded using edge direction information, to compute the likelihood that a sequence of measurements is consistent with a buried landmine. In particular, the HMM mine models learns the hyperbolic shape associated with the signature of a buried mine by three states that correspond to the succession of an increasing edge, a at edge, and a decreasing edge. Recently, for the same application, other features have been used with dierent classiers. In particular, the Edge Histogram Descriptor (EHD) has been used within a K-nearest neighbor classier. Another descriptor is based on Gabor features and has been used within a discrete HMM classier. A third feature, that is closely related to the EHD, is the Bar histogram feature. This feature has been used within a Neural Networks classier for handwritten word recognition. In this paper, we propose an evaluation of the HMM based landmine detection framework with several feature extraction techniques. We adapt and evaluate the EHD, Gabor, Bar, and baseline gradient feature extraction methods. We compare the performance of these features using a large and diverse GPR data collection.

This paper proposes a correlation method for detecting super-regenerative RF receivers via stimulation. Long PN sequences are used as to stimulate the unintended emissions from the RF receivers. High correlation between known PN sequence and stimulated unintended emissions from RF receivers helps improving the detection range compared to passive detection and power detection methods. Although RF receivers generate unintended emissions from their nonlinear devices, without stimulation, the power of these unintended emission is usually lower than --70dBm, as per the FCC regulations. Direct detection (passive detection) of these emissions is a challenging task specially in noisy conditions. When a stimulation signal is transmitted from distance, superregenerative receivers generate unintended emissions that contain the stimulation signal and its harmonics. Excellent correlation property of PN sequence enables us to improve the range and accuracy of detecting the super-regenerative receivers through stimulation method even in noisy conditions. The experiment involves detection of wireless doorbell, a commercially available super-regenerative receiver. USRP is used for transmitting the stimulant signal and receiving unintended stimulated emissions from the doorbell. Experiments show that the detection range of the proposed method with long PN sequences is much larger than passive detection and power detection methods.