This PDF file contains the front matter associated with SPIE Proceedings Volume 9454, including the Title Page, Copyright information, Table of Contents, Invited Panel Discussion, and Conference Committee listing.

On-site ground-penetrating radar (GPR) measurements with visual or acoustic real-time analysis cannot provide direct information whether or not GPR is suitable for the site at all. However, the knowledge of the limitations of a technique is of vital importance in the field in case of landmine, IED or UXO detection. For high-frequency (HF) GPR applications, various electromagnetic (EM) loss mechanisms in the soil play a crucial role. We investigated the EM properties of different soils using the coaxial transmission line (CTL) technique in the laboratory. We compared these results with measurements based on time-domain reflectometry (TDR) and direct current (DC) electrical conductivity measurements. We found that the absorption of EM energy in the soil cannot be described by DC electrical conductivity alone since dielectric relaxation mechanisms prevail at high frequencies. In order to predict the soil-dependent performance of GPR, we propose a conventional, relatively inexpensive, soil-moisture field sensor based on TDR as an alternative to the time consuming laboratory measurements. The TDR probe was calibrated by means of the CTL technique and measures the intrinsic attenuation as well as the relative dielectric constant. Comparisons between the GPR performance forecast carried out by on-site TDR measurements and the experimental GPR performance shows a promising correlation.

The need for advanced techniques to detect improvised explosive devices (IED) at stand-off distances greater than ten (10) meters has driven AMI Research and Development (AMI) to develop a solution to detect and identify the threat utilizing a forward looking Synthetic Aperture Radar (SAR) combined with our CW radar technology Nuclear Quadrupole Resonance (NQR) detection system. The novel features include a near-field sub-wavelength focusing antenna, a wide band 300 KHz to 300 MHz rapidly scanning CW radar facilitated by a high Q antenna/tuner, and an advanced processor utilizing Rabi transitions where the nucleus oscillates between states under the time dependent incident electromagnetic field and alternately absorbs energy from the incident field while emitting coherent energy via stimulated emission. AMI’s Sub-wavelength Focusing Wide Band Super Lens uses a Near-Field SAR, making detection possible at distances greater than ten (10) meters. This super lens is capable of operating on the near-field and focusing electromagnetic waves to resolutions beyond the diffraction limit. When applied to the case of a vehicle approaching an explosive hazard the methodologies of synthetic aperture radar is fused with the array based super resolution and the NQR data processing detecting the explosive hazard.

Mine countermeasures (MCM) missions entail planning and operations in very dynamic and uncertain operating environments, which pose considerable risk to personnel and equipment. Frequent schedule repairs are needed that consider the latest operating conditions to keep mission on target. Presently no decision support tools are available for the challenging task of MCM mission rescheduling. To address this capability gap, we have developed the CARPE system to assist operation planners. CARPE constantly monitors the operational environment for changes and recommends alternative repaired schedules in response. It includes a novel schedule repair algorithm called Case-Based Local Schedule Repair (CLOSR) that automatically repairs broken schedules while satisfying the requirement of minimal operational disruption. It uses a case-based approach to represent repair strategies and apply them to new situations. Evaluation of CLOSR on simulated MCM operations demonstrates the effectiveness of case-based strategy. Schedule repairs are generated rapidly, ensure the elimination of all mines, and achieve required levels of clearance.

Data fusion is a powerful theory that often leads to significant performance gain and/or improved robustness of a given solution. In this article, we explore how fusion can be used to advance our previously established improved Evolutionary COnstructed (iECO) image descriptor framework. The goal of iECO is to learn a diverse set of individuals (variable length chromosome in a genetic algorithm). Each iECO individual encodes a unique composition of different low-level image transformations in the context of a high-level image descriptor. Herein, we investigate multiple kernel (MK) aggregation and MK learning (MKL) for “feature-level” fusion of iECO chromosomes. Specifically, we explore MKL group lasso (MKLGL) and we put forth a new way to directly assign kernel weights from a measure defined on the kernel matrices. The proposed work is presented in the context of buried explosive hazard detection (EHD) in forward looking (FL) imagery. Experiments are reported using receiver operating characteristic (ROC) curves on data from a U.S. Army test site that contains multiple target and clutter types, burial depths and times of day. We demonstrate that MK support vector machine (MKSVM) classification outperform single kernel SVM (SKSVM) classification and our weight assignment procedure generalizes well and outperforms MKLGL for EHD in FLIR.

In this paper, we demonstrate the feasibility of microwave holographic imaging application based on spintronic microwave sensors. By adapting the rapid phase detection technique, a magnetic tunnel junction can achieve a real-time measurement of both the amplitude and phase of the scattered microwave using a lock-in amplifier. The built system has a capability to detect not only the existence of the concealed objects but also their shapes, allowing the concealed threat to be distinguished along with other hidden objects. The system is also able to estimate the distance of target by a broadband measurement. Anticipating the phase-detection of the dielectric function of the targets, we have also carried out extensive density functional theory calculations for a number of condensed phase energetic materials to determine their dielectric response in the microwave range.

Differential Excitation Spectroscopy (DES) is a new pump-probe detection technique (patent-pending) which characterizes molecules based on a multi-dimensional parameterization of the rovibrational excited state structure, pump and probe interrogation frequencies, as well as the lifetimes of the excited states. Under appropriate conditions, significant modulation of the ground state can result. DES results provide a unique, simple mechanism to probe various molecules. In addition, the DES multi-dimensional parameterization provides an identification signature that is highly unique and has demonstrated high levels of immunity from interferents, providing significant practical value for high-specificity material identification. Ammonium nitrate (AN) and urea nitrate (UN) are both components commonly used in IEDs; the ability to reliably detect these chemicals is key to finding, identifying and defeating IEDs. AN and UN are complicated materials, having a number of different phases and because they are molecular crystals, there are a number of different types of interactions between the constituent atoms which must be characterized in order to understand their DES behavior. Ab initio calculations were performed on both AN and UN for various rovibrational states up to J’ ≤ 3 and validated experimentally, demonstrating good agreement between theory and experiment and the very specific responses generated.

A variety of techniques exist for enhancing or inferring the existence and characteristics of an obscured or partially concealed target. Targets, however, may be completely blocked from view, presenting nothing to enhance and no image area to extend inferentially. Despite the difficulty, concealed (particularly intentionally) targets may be the most important to detect. This paper proposes a technique for using a Blackboard Architecture or Expert system to infer a target’s existence from symptoms (maneuvers of other units, water and soil deformation, etc.) and discusses the differences between the two approaches (Blackboard Architecture and expert system) for doing so.

One of the modern problems arising in the detection and identification of substances is a development of criteria for the assessment of a presence of explosive (or other dangerous substance) fingerprints in THz signals transmitted through or reflected from a sample. Obviously, criteria depend on the method used for the substance detection and identification. Taking into account our previous experience, we use for a solution of this problem the SDA method (method of the spectral dynamics analysis). Essential restrictions of usually used THz TDS method for the detection and identification under real conditions (at long distance about 3.5 m and at a high relative humidity more than 50%) are demonstrated using the physical experiment with paper napkins and thick paper bag. We show also that the THz TDS method detects spectral features of dangerous substances even in the THz signals measured in laboratory conditions (at distance 30-40 cm from the receiver and at a low relative humidity less than 2%) with semiconductors of different types used as samples. However, the integral correlation criteria, based on SDA method, allows us to detect the absence of dangerous substances in semiconductors. In order to demonstrate the possibilities of the integral criteria for finding additional substances in the mixture with semiconductors, we modeled several mixtures of n-doped Silicon with neutral substance Soap in different ratio. The discussed algorithms show high probability of the substance identification and a reliability of realization in practice, especially for non-destructive testing and security applications.

Vehicle Mounted Metal Detector (VMMD) systems are widely used for detection of threat objects in humanitarian demining and military route clearance scenarios. Due to the diverse nature of such operational conditions, operational use of VMMD without a proper understanding of its capability boundaries may lead to heavy causalities. Multi-criteria fitness evaluations are crucial for determining capability boundaries of any sensor-based demining equipment. Evaluation of sensor based military equipment is a multi-disciplinary topic combining the efforts of researchers, operators, managers and commanders having different professional backgrounds and knowledge profiles. Information acquired through field tests usually involves uncertainty, vagueness and imprecision due to variations in test and evaluation conditions during a single test or series of tests. This report presents a fuzzy logic based methodology for experimental data analysis and performance evaluation of VMMD. This data evaluation methodology has been developed to evaluate sensor performance by consolidating expert knowledge with experimental data. A case study is presented by implementing the proposed data analysis framework in a VMMD evaluation scenario. The results of this analysis confirm accuracy, practicability and reliability of the fuzzy logic based sensor performance evaluation framework.

This paper explores the effectiveness of an anomaly detection algorithm for downward-looking ground penetrating radar (GPR) and electromagnetic inductance (EMI) data. Threat detection with GPR is challenged by high responses to non-target/clutter objects, leading to a large number of false alarms (FAs), and since the responses of target and clutter signatures are so similar, classifier design is not trivial. We suggest a method based on a Run Packing (RP) algorithm to fuse GPR and EMI data into a composite confidence map to improve detection as measured by the area-under-ROC (NAUC) metric. We examine the value of a multiple kernel learning (MKL) support vector machine (SVM) classifier using image features such as histogram of oriented gradients (HOG), local binary patterns (LBP), and local statistics. Experimental results on government furnished data show that use of our proposed fusion and classification methods improves the NAUC when compared with the results from individual sensors and a single kernel SVM classifier.

Typically, electromagnetic sensors are applied using one of two modalities to detect buried or obscured targets: 1) lower frequency Electromagnetic Induction (EMI) sensors that enable detection of targets in the near-field; and 2) higher frequency wave propagation sensors, such as Forward Looking or Ground Penetrating Radar (FL/GPR) that enable detection of targets in the far-field. Each modality has advantages and limitations. EMI sensors enable deep penetration of overburden or structures that may obscure a target; however, sensitivity is typically limited to high conductivity targets (i.e., metals) due to the relatively low frequency of operation. Wave propagation sensors, such as GPR, enable detection of both conductive and non-conductive targets as a result of inherent dielectric contrast sensitivity; however, penetration into ground or structures is limited due to rapid attenuation of the propagating wave through lossy materials. In this paper, we present a concept for enhancing the target range capabilities of EMI sensors to extend sensitivity to lower conductivity targets. This concept incorporates an efficient transmitter driver design that extends the range of EMI operation into the High Frequency (HF) band while providing high power output. This ability to produce high frequency, high power output provides a sensor modality that bridges the gap between traditional EMI and wave propagating modalities. This High Frequency Transmitter (HFTX) concept could enable sensitivity to low conductivity targets (i.e., non-metals) while maintaining effective penetration through soil overburden or other materials that would typically impede GPR wave propagation.

In this work, we explore the efficacy of two buried threat detectors on handheld data. The first algorithm is an energy-based algorithm, which computes how anomalous a given A-scan measurement after it is normalized according to its local statistics. It is based on a commonly used prescreener for the Husky Mounted Detection System (HMDS). In the HMDS setting measurements are sampled on a crosstrack-downtrack grid, and sequential measurements are at neighboring downtrack locations. In contrast, in the handheld setting sequential scans are often taken at neighboring crosstrack locations, and neighboring downtrack locations can be hundreds of scans away. In order to include both downtrack and crosstrack information, we compute local statistics over a much larger area than in the HMDS setting. The second algorithm is a shape-based algorithm. Shape Invariant Feature Transform (SIFT) features, which capture the gradient distributions of local patches, are extracted and used to train a non-linear Support Vector Machine (SVM). We found that in terms of AUC, the SIFT-SVM algorithm results in a 2.2% absolute improvement over the energy-based algorithm, with the greatest gains seen at lower false alarm rates.

A dictionary learning approach for subsurface object detection using handheld electromagnetic induction (EMI) data is presented. A large number of unsupervised and supervised dictionary learning methods have been developed in the literature. However, the majority of these methods require data point-specific labels during training. In the application to subsurface object detection, often the specific training data samples that correspond to target and non-target are not known and difficult to determine manually. In this paper, a dictionary learning method that addresses this issue using the multiple instance learning techniques is presented. Results are shown on real EMI data sets.

In this paper the phase response and reflection coefficient notch of a metal/anomaly detector design that operates in the high to very high frequency range is studied. This design uses a high-Q tuned loop antenna for metal/anomaly detection. By measuring the reflection coefficient or voltage standing wave ratio a frequency notch can be detected. Tuning to the optimal location for detector performance can be accomplished by monitoring both the depth of the notch and the phase response. It has been experimentally observed that there are three regions of interest relative to the notch and phase response of the detector. One is at the frequency where the phase response is on a near vertical line of substantial phase shift and the notch is near its deepest depth. The second and third are at slightly higher and lower frequencies, where the slope of the phase shift line is reduced and the notch is still deep, but slightly removed from the frequency of maximum depth. As would be expected, initial experimentation indicates that the region of maximum detection performance, in terms of relative change in phase response, occurs when the phase response is at the center of the near vertical phase shift response near the location of the deepest notch. However, there may be advantages to the other two regions, since the response is more stable and less prone to false alarms. Performance results for various combinations of phase response and notch depth will be shown.

Many remote sensing modalities have been developed for buried target detection, each one offering its own relative advantages over the others. As a result there has been interest in combining several modalities into a single detection platform that benefits from the advantages of each constituent sensor, without suffering from their weaknesses. Traditionally this involves collecting data continuously on all sensors and then performing data, feature, or decision level fusion. While this is effective for lowering false alarm rates, this strategy neglects the potential benefits of a more general system-level fusion architecture. Such an architecture can involve dynamically changing which modalities are in operation. For example, a large standoff modality such as a forward-looking infrared (FLIR) camera can be employed until an alarm is encountered, at which point a high performance (but short standoff) sensor, such as ground penetrating radar (GPR), is employed. Because the system is dynamically changing its rate of advance and sensors, it becomes difficult to evaluate the expected false alarm rate and advance rate. In this work, a probabilistic model is proposed that can be used to estimate these quantities based on a provided operating policy. In this model the system consists of a set of states (e.g., sensors employed) and conditions encountered (e.g., alarm locations). The predictive accuracy of the model is evaluated using a collection of collocated FLIR and GPR data and the results indicate that the model is effective at predicting the desired system metrics.

In recent years, the number of commercially available LADAR (also referred to as LIDAR) systems have grown with the increased interest in ground vehicle robotics and aided navigation/collision avoidance in various industries. With this increased demand the cost of these systems has dropped and their capabilities have increased. As a result of this trend, LADAR systems are becoming a cost effective sensor to use in a number of applications of interest to the US Army. One such application is the standoff detection of road-side hazards from ground vehicles. This paper will discuss detection of road-side hazards partially concealed by light to medium vegetation. Current algorithms using commercially available LADAR systems for detecting these targets will be presented, along with results from relevant data sets. Additionally, optimization of commercial LADAR sensors and/or fusion with Radar will be discussed as ways of increasing detection ability.

Characterization of hazardous lands using ground-based techniques can be very challenging. For this reason, airborne surveys are often preferred. The use of thermal infrared imaging represents an interesting approach as surveys can be carried out under various illumination conditions and that the presence of buried objects typically modifies the thermal inertia of their surroundings. In addition, the burial or presence of a buried object will modify the particle size, texture, moisture and mineral content of a small region around it. All these parameters may lead to emissivity contrasts which will make thermal contrast interpretation very challenging. In order to illustrate the potential of airborne thermal infrared hyperspectral imaging for buried object characterization, various metallic objects were buried in a test site prior to an airborne survey. Airborne hyperspectral images were recorded using the targeting acquisition mode, a unique feature of the Telops Hyper-Cam Airborne system which allows recording of successive maps of the same ground area. Temperatureemissivity separation (TES) was carried out on the hyperspectral map obtained upon scene averaging. The thermodynamic temperature map estimated after TES highlights the presence of hot spots within the investigated area. Mineral mapping was carried out upon linear unmixing of the spectral emissivity datacube obtained after TES. The results show how the combination of thermal information and mineral distribution leads to a better characterization of test sites containing buried objects.

A big challenge with forward looking (FL), versus downward looking, sensors mounted on a ground vehicle for explosive hazard detection (EHD) is they “see everything”, on and off road. Even if a technology such as road detection is used, we still have to find and subsequently discriminate targets versus clutter on the road and often road side. When designing an automatic detection system for FL-EHD, we typically make use of a prescreener to identify regions of interest (ROI) instead of searching for targets in an inefficient brute force fashion by extracting complicated features and running expensive classifiers at every possible translation, rotation and scale. In this article, we explore the role of genetic algorithms (GAs), specifically with respect to a new adaptive mutation operator, for learning the parameters of a FL-EHD prescreener in FL infrared (FLIR) imagery. The proposed extended adaptive mutation (eAM) algorithm is driven by fitness similarities in the chromosome population. Currently, our prescreener consists of many free parameters that are empirically chosen by a researcher. The parameters are learned herein using the proposed optimization technique and the performance of the system is measured using receiver operating characteristic (ROC) curves on data obtained from a U.S. Army test site that includes a variety of target types buried at varying depths and from different times of day. The proposed technique is also applied to numerous synthetic fitness landscapes to further assess the effectiveness of the eAM algorithm. Results show that the new adaptive mutation technique converges faster to a better solution than a GA with fixed mutation.

A major difficulty in designing an automatic explosive hazard detection (EHD) system in forward looking (FL) imagery is the robust and efficient detection of regions of interest (ROIs) that warrant further investigation. FL-EHD is particularly challenging, versus a downward looking technology, because a camera sees everything in the scene, on- and off-road. While off-road can be somewhat mitigated through various mechanisms, such as road masks or a road detector, on-road obstacles still have to be addressed. A brute force strategy is infeasible for this application as it requires advanced standoff capabilities, a low false alarm rate, and real-time processing to achieve a goal such as route clearance or target avoidance. Herein, we discuss the design of a new pre-screener based on shearlet filtering and image post-processing that lets us exploit important characteristics of targets in FL imagery identified by a maximally stable extremal region (MSER) keypoint detector. Results indicate that this approach performs as desired, i.e., identifies expected percentage of target ROIs at the defined acceptable FAR, without need for extensive parameter learning. Performance is assessed in the context of receiver operating characteristic curves on data from a U.S. Army test site that contains multiple target and clutter types, burial depths and times of day.

Implanted mines and improvised devices are a persistent threat to Warfighters. Current Army countermine missions for route clearance need on-the-move standoff detection to improve the rate of advance. Vehicle-based forward looking sensors such as electro-optical and infrared (EO/IR) devices can be used to identify potential threats in near real-time (NRT) at safe standoff distance to support route clearance missions. The MOVERS (Micro-Cloud for Operational, Vehicle-Based EO-IR Reconnaissance System) is a vehicle-based multi-sensor integration and exploitation system that ingests and processes video and imagery data captured from forward-looking EO/IR and thermal sensors, and also generates target/feature alerts, using the Video Processing and Exploitation Framework (VPEF) “plug and play” video processing toolset. The MOVERS Framework provides an extensible, flexible, and scalable computing and multi-sensor integration GOTS framework that enables the capability to add more vehicles, sensors, processors or displays, and a service architecture that provides low-latency raw video and metadata streams as well as a command and control interface. Functionality in the framework is exposed through the MOVERS SDK which decouples the implementation of the service and client from the specific communication protocols.

The scope of the Micro-Cloud for Operational, Vehicle-Based EO-IR Reconnaissance System (MOVERS) development effort, managed by the Night Vision and Electronic Sensors Directorate (NVESD), is to develop, integrate, and demonstrate new sensor technologies and algorithms that improve improvised device/mine detection using efficient and effective exploitation and fusion of sensor data and target cues from existing and future Route Clearance Package (RCP) sensor systems. Unfortunately, the majority of forward looking Full Motion Video (FMV) and computer vision processing, exploitation, and dissemination (PED) algorithms are often developed using proprietary, incompatible software. This makes the insertion of new algorithms difficult due to the lack of standardized processing chains. In order to overcome these limitations, EOIR developed the Government off-the-shelf (GOTS) Video Processing and Exploitation Framework (VPEF) to be able to provide standardized interfaces (e.g., input/output video formats, sensor metadata, and detected objects) for exploitation software and to rapidly integrate and test computer vision algorithms. EOIR developed a vehicle-based computing framework within the MOVERS and integrated it with VPEF. VPEF was further enhanced for automated processing, detection, and publishing of detections in near real-time, thus improving the efficiency and effectiveness of RCP sensor systems.

In this paper, we present a vehicular buried threat detection approach developed over the past several years, and its latest implementation and integration in VPEF environment. Buried threats have varying signatures under different operation environment. To reliably detect the true targets and minimizing the number of false alarms, a suite of false alarm mitigators (FAMs) have been developed to process the potential targets identified by the baseline module. A vehicle track can be formed over a number of frames and targets are further analyzed both spatially and temporally. Algorithms have been implemented in C/C++ as GStreamer plugins and are suitable for vehicle mounted, on-the-move realtime exploitation.

A sliding window based prescreening algorithm, utilizing multi-scale histogram of oriented gradient (MS-HOG) features and a linear support vector machine (SVM) classifier, for detection of buried explosive hazards in forward-looking infrared (FL-IR) and forward-looking ground penetrating radar (FL-GPR) data is presented. This algorithm is compared to previously published FL-IR and FL-GPR prescreening algorithms. The MS-HOG prescreening approach has higher computational complexity, but improves overall detection rates, especially for low-contrast and obscured target signatures. Results are presented on several data sets collected at US Army test sites. These collections span several days, and the FL-IR collections include imagery from both long-wave and mid-wave infrared cameras at multiple standoff distances captured at different hours of the day and different times of the year.

For road implanted explosive hazard detection, detecting the road at distance is critical for the classification algorithms and for sensor positioning to maintain road view during turns. In this paper, we propose the use of the Log-Gabor Filter (LGF) to enhance our road detection system. The LGF can be used to suppress the road-like pixels in the image. By filtering the unpaved road images with varying scales and orientations of the LGF and a combination of basic image processing techniques, evidence images of the road are created. Each evidence image is a binary image where value one at any pixel represents evidence of the road at that pixel. Otherwise the value will be zero. However, the maximum distance for generating evidence of the road varies for each image. Therefore, additionally, a road model is utilized. Using the least squares algorithm, the road model is optimized to fit the support of the road presented in each image. By specifying the length of the road on the optimized model, the distance of road detection can also be specified. Thus, utilizing the LGF and the road model allows our system to detect poorly defined dirt roads as far as forty meters as shown for a winding road at an arid U.S. Army test site.

A method for detecting ultralow quantities of explosives in air with use a state-of-the-art picosecond chip Nd³⁺:YAG laser has been developed. The method combines field asymmetric ion mobility spectrometry (FAIMS) with laser ionization of examined air samples. Radiation of λ = 266nm, τ_pulse = 300ps, E_pulse = 30–150μJ, ν = 20–300Hz was used. Processes in the ion source for the use both picosecond and nanosecond ionization modes were analyzed. Parameters of the laser ion source have been specially optimized. The dependences on frequency, pulse energy, peak intensity, and average power for trinitrotoluene (TNT) and cyclotrimethylenetrinitramine (RDX) were obtained. It was shown that the optimal peak intensity should be no less 3·10⁶ W/cm². The detected ion signals for all explosives were shown to be threefold higher for picosecond excitation in comparison with use a nanosecond laser of the same average power. The estimated detection threshold of the prototype equals 1. 10^-15 g/cm³. The results are promising for the development of a highly sensitive, portable laser explosive detector.

The requirement to detect hazardous materials (i.e., chemical, biological, and explosive) on a host of materials has led to the development of hazard detection systems. These new technologies and their capabilities could have immediate uses for the US military, national security agencies, and environmental response teams in efforts to keep people secure and safe. In particular, due to the increasing use by terrorists, the detection of common explosives and improvised explosive device (IED) materials have motivated research efforts toward detecting trace (i.e., particle level) quantities on multiple commonly encountered surfaces (e.g., textiles, metals, plastics, natural products, and even people). Non-destructive detection techniques can detect trace quantities of explosive materials; however, it can be challenging in the presence of a complex chemical background. One spectroscopic technique gaining increased attention for detection is Raman. One popular explosive precursor material is ammonium nitrate (AN). The material AN has many agricultural applications, however it can also be used in the fabrication of IEDs or homemade explosives (HMEs). In this paper, known amounts of AN will be deposited using an inkjet printer into several different common material surfaces (e.g., wood, human hair, textiles, metals, plastics). The materials are characterized with microscope images and by collecting Raman spectral data. In this report the detection and identification of AN will be demonstrated.

Raman spectroscopy is a powerful tool for obtaining molecular structure information of a sample. While Raman spectroscopy is a common laboratory based analytical tool, miniaturization of opto-electronic components has allowed handheld Raman analyzers to become commercially available. These handheld systems are utilized by Military and Bomb squad operators tasked with rapidly identifying explosives in the field, sometimes in clandestine laboratories. However, one limitation of many handheld Raman detection systems is strong interference caused by fluorescence of the sample or underlying surface which obscures the characteristic Raman signature of the target analyte. Homemade explosives (HMEs) are produced in clandestine laboratories, and the products under these conditions are typically contaminated with degradation products, contaminants, and unreacted precursors. These contaminations often will have strong fluorescence. In this work, Raman spectra of both commercial explosives and HMEs were collected using a handheld Raman spectrometer with a 1064 nm excitation laser. While Raman scattering generated by a 1064 nm laser is inherently less efficient than excitation at shorter wavelengths, high quality spectra were easily obtained due to significantly reduced fluorescence of HMEs.

Forward-looking ground penetrating radar (FLGPR) is a remote sensing modality that has recently been investigated for buried threat detection. FLGPR offers greater standoff than other downward-looking modalities such as electromagnetic induction and downward-looking GPR, but it suffers from high false alarm rates due to surface and ground clutter. A stepped frequency FLGPR system consists of multiple radars with varying polarizations and bands, each of which interacts differently with subsurface materials and therefore might potentially be able to discriminate clutter from true buried targets. However, it is unclear which combinations of bands and polarizations would be most useful for discrimination or how to fuse them. This work applies sparse structured basis pursuit, a supervised statistical model which searches for sets of bands that are collectively effective for discriminating clutter from targets. The algorithm works by trying to minimize the number of selected items in a dictionary of signals; in this case the separate bands and polarizations make up the dictionary elements. A structured basis pursuit algorithm is employed to gather groups of modes together in collections to eliminate whole polarizations or sensors. The approach is applied to a large collection of FLGPR data for data around emplaced target and non-target clutter. The results show that a sparse structure basis pursuits outperforms a conventional CFAR anomaly detector while also pruning out unnecessary bands of the FLGPR sensor.

Explosive hazards are one of the most deadly threats in modern conflicts. The U.S. Army is interested in a reliable way to detect these hazards at range. A promising way of accomplishing this task is using a forward-looking ground-penetrating radar (FLGPR) system. Recently, the Army has been testing a system that utilizes both L-band and X-band radar arrays on a vehicle mounted platform. Using data from this system, we sought to improve the performance of a constant false-alarm-rate (CFAR) prescreener through the use of a deep belief network (DBN). DBNs have also been shown to perform exceptionally well at generalized anomaly detection. They combine unsupervised pre-training with supervised fine-tuning to generate low-dimensional representations of high-dimensional input data. We seek to take advantage of these two properties by training a DBN on the features of the CFAR prescreener’s false alarms (FAs) and then use that DBN to separate FAs from true positives. Our analysis shows that this method improves the detection statistics significantly. By training the DBN on a combination of image features, we were able to significantly increase the probability of detection while maintaining a nominal number of false alarms per square meter. Our research shows that DBNs are a good candidate for improving detection rates in FLGPR systems.

A forward-looking and -moving ground-penetrating radar (GPR) acquires data that can be used for buried target detection. As the platform moves forward the sensor can acquire and form a sequence of images for a common spatial region. Due to the near-field nature of relevant collection scenarios, the point-spread function (PSF) varies significantly as a function of the spatial position, both within the scene and relative to the sensor platform. This variability of the PSF causes computational difficulties for matched-filter and related processing of the full video sequence. One approach to circumventing this difficulty is to coherently or incoherently integrate the video frames, and then perform detection processing on the integrated image. Here, averaging over the space- and motion-variant nature of the PSFs for each frame causes the PSF for the integrated image to appear less space-variant. Another alternative—and the one we investigate in this paper—is to transform each image from the conventional (range, cross-range) coordinate system to a (range, sine-angle) coordinate system in which the PSF is approximated as spatially invariant. The advantage of the (range, sine-angle) coordinate space is that methods that require space-invariance can be directly applied. Here we develop a multi-anodization approach, which results in a significantly improved image. To evaluate the relative advantages of this procedure, we will empirically measure the integrated side-lobe ratio, which represents the reduction in the side-lobes before and after applying the algorithm.

A novel quasi-monostatic system operating in a side-scan synthetic aperture acoustic (SAA) imaging mode is presented. This research project's objectives are to explore the military utility of outdoor continuous sound imaging of roadside foliage and target detection. The acoustic imaging method has several military relevant advantages such as being immune to RF jamming, superior spatial resolution as compared to 0.8-2.4 GHz ground penetrating radar (GPR), capable of standoff side and forward-looking scanning, and relatively low cost, weight and size when compared to GPR technologies. The prototype system's broadband 2-17 kHz LFM chirp transceiver is mounted on a manned all-terrain vehicle. Targets are positioned within the acoustic main beam at slant ranges of two to seven meters and on surfaces such as dirt, grass, gravel and weathered asphalt and with an intervening metallic chain link fence. Acoustic image reconstructions and signature plots result in means for literal interpretation and quantifiable analyses.

This paper proposes a machine learning algorithm for subsurface object detection on multiple-input-multiple-output (MIMO) forward-looking ground-penetrating radar (FLGPR). By detecting hazards using FLGPR, standoff distances of up to tens of meters can be acquired, but this is at the degradation of performance due to high false alarm rates. The proposed system utilizes an anomaly detection prescreener to identify potential object locations. Alarm locations have multiple one-dimensional (ML) spectral features, two-dimensional (2D) spectral features, and log-Gabor statistic features extracted. The ability of these features to reduce the number of false alarms and increase the probability of detection is evaluated for both co-polarizations present in the Akela MIMO array. Classification is performed by a Support Vector Machine (SVM) with lane-based cross-validation for training and testing. Class imbalance and optimized SVM kernel parameters are considered during classifier training.

Ground Penetrating Radar (GPR) is considered as one of the promising technologies to address the challenges of detecting buried threat objects. However, the success rate of the GPR systems are limited by operational conditions and the robustness of automatic target recognition (ATR) algorithms embedded with the systems. In this paper an alternate ATR algorithm applicable to GPR is developed by combining image pre-processing and machine learning techniques. The aim of this research was to design a potential solution for detection of threat alarms using GPR data and reducing the number of false alarms through classification into one of the predefined categories of target types. The proposed ATR algorithm has been validated using a data set acquired by a vehicle-mounted GPR array. The data set utilized in this investigation involved greyscale GPR images of threat objects (both conventional and improvised) commonly found in realistic operational scenarios. Target based summaries of the algorithm performance are presented in terms of the probability of detection, false alarm rate, and confidence of allocating detections to a predefined target class.

The Canadian Armed Forces (CAF) identified a requirement for inert simulants to act as improvised, or homemade, explosives (IEs) when training on, or evaluating, ground penetrating radar (GPR) systems commonly used in the detection of buried landmines and improvised explosive devices (IEDs). In response, Defence R and D Canada (DRDC) initiated a project to develop IE simulant formulations using commonly available inert materials. These simulants are intended to approximate the expected GPR response of common ammonium nitrate-based IEs, in particular ammonium nitrate/fuel oil (ANFO) and ammonium nitrate/aluminum (ANAl). The complex permittivity over the range of electromagnetic frequencies relevant to standard GPR systems was measured for bulk quantities of these three IEs that had been fabricated at DRDC Suffield Research Centre. Following these measurements, published literature was examined to find benign materials with both a similar complex permittivity, as well as other physical properties deemed desirable - such as low-toxicity, thermal stability, and commercial availability - in order to select candidates for subsequent simulant formulation. Suitable simulant formulations were identified for ANFO, with resulting complex permittivities measured to be within acceptable limits of target values. These IE formulations will now undergo end-user trials with CAF operators in order to confirm their utility. Investigations into ANAl simulants continues. This progress report outlines the development program, simulant design, and current validation results.

Symmetric and asymmetric buried explosive hazards (BEHs) present real, persistent, deadly threats on the modern battlefield. Current approaches to mitigate these threats rely on highly trained operatives to reliably detect BEHs with reasonable false alarm rates using handheld Ground Penetrating Radar (GPR) and metal detectors. As computers become smaller, faster and more efficient, there exists greater potential for automated threat detection based on state-of-the-art machine learning approaches, reducing the burden on the field operatives. Recent advancements in machine learning, specifically deep learning artificial neural networks, have led to significantly improved performance in pattern recognition tasks, such as object classification in digital images. Deep convolutional neural networks (CNNs) are used in this work to extract meaningful signatures from 2-dimensional (2-D) GPR B-scans and classify threats. The CNNs skip the traditional “feature engineering” step often associated with machine learning, and instead learn the feature representations directly from the 2-D data. A multi-antennae, handheld GPR with centimeter-accurate positioning data was used to collect shallow subsurface data over prepared lanes containing a wide range of BEHs. Several heuristics were used to prevent over-training, including cross validation, network weight regularization, and “dropout.” Our results show that CNNs can extract meaningful features and accurately classify complex signatures contained in GPR B-scans, complementing existing GPR feature extraction and classification techniques.

This paper investigates the application of Robust Principal Component Analysis (RPCA) to ground penetrating radar as a means to improve GPR anomaly detection. The method consists of a preprocessing routine to smoothly align the ground and remove the ground response (haircut), followed by mapping to the frequency domain, applying RPCA, and then mapping the sparse component of the RPCA decomposition back to the time domain. A prescreener is then applied to the time-domain sparse component to perform anomaly detection. The emphasis of the RPCA algorithm on sparsity has the effect of significantly increasing the apparent signal-to-clutter ratio (SCR) as compared to the original data, thereby enabling improved anomaly detection. This method is compared to detrending (spatial-mean removal) and classical principal component analysis (PCA), and the RPCA-based processing is seen to provide substantial improvements in the apparent SCR over both of these alternative processing schemes. In particular, the algorithm has been applied to both field collected impulse GPR data and has shown significant improvement in terms of the ROC curve relative to detrending and PCA.

Ground penetrating radar (GPR) devices use sensors to capture one-dimensional representations, or A-scans, of the soil and buried properties at each sampling point. Previous work uses reciprocal pointer chains (RPCs) to find one-dimensional layers in two-dimensional data (B-scans). We extend this work to find two-dimensional layers in three-dimensional data. We explore the application and differences of our technique when applied to vehicular mounted systems versus handheld systems and their distinct detection sequences. Not only can this work be used to display subsurface structure to a system operator, but we can also use changes in the subsurface structure of a local region to help identify buried objects within the data. We propose distinguishing buried objects from layers can reduce false alarm rates and may help increase probability of detection.

Buried threat detection algorithms in Ground Penetrating Radar (GPR) measurements often utilize a statistical classifier to model target responses. There are many different target types with distinct responses and all are buried in a wide range of conditions that distort the target signature. Robust performance of this classifier requires it to learn the distinct responses of target types while accounting for the variability due to the physics of the emplacement. In this work, a method to reduce certain sources of excess variation is presented that enables a linear classifier to learn distinct templates for each target type’s response despite the operational variability. The different target subpopulations are represented by a Gaussian Mixture Model (GMM). Training the GMM requires jointly extracting the patches around target responses as well as learning the statistical parameters as neither are known a priori. The GMM parameters and the choice of patches are determined by variational Bayesian methods. The proposed method allows for patches to be extracted from a larger data-block that only contain the target response. The patches extracted from this method improve the ROC for distinguishing targets from background clutter compared to the patches extracted using other patch extraction methods aiming to reduce the operational variability.

This paper investigates an algorithm for forming 3D images of the subsurface using stepped-frequency GPR data. The algorithm is specifically designed for a handheld GPR and therefore accounts for the irregular sampling pattern in the data and the spatially-variant air-ground interface by estimating an effective “ground-plane” and then registering the data to the plane. The algorithm efficiently solves the 4th-order polynomial for the Snell reflection points using a fully vectorized iterative scheme. The forward operator is implemented efficiently using an accelerated nonuniform FFT (Greengard and Lee, 2004); the adjoint operator is implemented efficiently using an interpolation step coupled with an upsampled FFT. The imaging is done as a linearized version of the full inverse problem, which is regularized using a sparsity constraint to reduce sidelobes and therefore improve image localization. Applying an appropriate sparsity constraint, the algorithm is able to eliminate most the surrounding clutter and sidelobes, while still rendering valuable image properties such as shape and size. The algorithm is applied to simulated data, controlled experimental data (made available by Dr. Waymond Scott, Georgia Institute of Technology), and government-provided data with irregular sampling and air-ground interface.

In this paper, we propose a system to detect buried disk-shaped landmines from ground penetrating radar (GPR) and forward-looking long wave infrared (FL-LWIR) data. The data is collected from a test area of 500m², which was prepared at the IPA Defence, Ankara, Turkey. This test area was divided into four lanes, each of size 25m length by 4m width and 1m depth. Each lane was first carefully cleaned of stones and clutter and then filled with different soil types, namely fine-medium sand, course sand, sandy silt loam and loam mix. In all lanes, various clutter objects and landmines were buried at different depths and at 1meter intervals. In the proposed approach, IR data is used as a pre-screener. Then possible target regions are further analyzed using the GPR data. IR data processing is done in three steps such as preprocessing, target detection, and postprocessing. In the pre-processing stage, bilateral noise reduction filtering is performed. The target detection stage finds circular targets by a radial transformation algorithm. The proposed approach is compared with the RX algorithm used widely for anomaly detection. The suspicious regions are further analyzed using Histogram of Oriented Gradient (HOG) features that are extracted from GPR images and classified by SVM. The same approach can also be applied in a parallel way where the results are combined using decision level fusion. The results of the proposed approach are given on different scenarios including different weather temperature and depth of buried targets.

The ground penetrating radar (GPR) signal for a deeply buried non-metal object is weak and often does not have a hyperbolic signature, making it difficult to detect with high confidence. This paper takes a blind source separation approach by using non-negative matrix factorization (NMF) to improve the detection of deeply buried non-metal objects. The proposed approach interprets the GPR signal return as the sum of two independent components from two different sources, the background and the object. NMF enables the separation of the object signal component from the composite and thereby improves the detection performance. Preliminary results from a test site in the United States indicate that the probability of detecting these objects is improved by more than 20% compared to the pre-screener, at a false alarm rate of 0.003/m².

Improved performance in the discrimination of buried threats using Ground Penetrating Radar (GPR) data has recently been achieved using features developed for applications in computer vision. These features, designed to characterize local shape information in images, have been utilized to recognize patches that contain a target signature in two-dimensional slices of GPR data. While these adapted features perform very well in this GPR application, they were not designed to specifically differentiate between target responses and background GPR data. One option for developing a feature specifically designed for target differentiation is to manually design a feature extractor based on the physics of GPR image formation. However, as seen in the historical progression of computer vision features, this is not a trivial task. Instead, this research evaluates the use of convolutional neural networks (CNNs) applied to two-dimensional GPR data. The benefit of using a CNN is that features extracted from the data are a learned parameter of the system. This has allowed CNN implementations to achieve state of the art performance across a variety of data types, including visual images, without the need for expert designed features. However, the implementation of a CNN must be done carefully for each application as network parameters can cause performance to vary widely. This paper presents results from using CNNs for object detection in GPR data and discusses proper parameter settings and other considerations.

This paper develops an anomaly detection algorithm for subsurface object detection using the handheld ground penetrating radar. The algorithm is based on the Mahalanobis distance measure with adaptive update of the background statistics. It processes the data sequentially for each data sample in a causal manner to generate detection confidences. The algorithm is applied to process the data from two different radars, an impulse and a step-frequency, for performance evaluation.

Ground-penetrating radar (GPR) technology has proven capable of detecting buried threats. The system relies on a binary classifier that is trained to distinguish between two classes: a target class, encompassing many types of buried threats and their components; and a nontarget class, which includes false alarms from the system prescreener. Typically, the training process involves a simple partition of the data into these two classes, which allows for straightforward application of standard classifiers. However, since training data is generally collected in fully controlled environments, it includes auxiliary information about each example, such as the specific type of threat, its purpose, its components, and its depth. Examples from the same specific or general type may be expected to exhibit similarities in their GPR data, whereas examples from different types may differ greatly. This research aims to leverage this additional information to improve overall classification performance by fusing classifier concepts for multiple groups, and to investigate whether structure in this information can be further utilized for transfer learning, such that the amount of expensive training data necessary to learn a new, previously-unseen target type may be reduced. Methods for accomplishing these goals are presented with results from a dataset containing a variety of target types.

A goal of ground penetrating radar (GPR) preprocessing is to distinguish background from data containing explosive threats. This is commonly achieved by performing depth-dependent mean and standard deviation normalization, where the mean and standard deviation are computed on background data. Under the assumption that data with explosive threats have different statistical characteristics than the background/clutter, after normalization explosive threat data will have larger absolute normalized scores than the background/clutter. An underlying problem is determining which data to compute the background mean and standard deviation statistics over. Often the background statistics are computed over a moving window, which is centered at the location of interest and has a predetermined guard band, a region of data that is ignored. However, buried explosive threats vary considerably in their shapes and more importantly sizes subsequently, the size of the GPR responses from these objects are considerably varied. We examine a number of additional detection methods that utilize Robust Principal Component Analysis (RPCA), where RPCA decomposes the data into low-rank and sparse components. Intuitively, the low-rank component should capture the background data and the sparse should capture the anomalous explosive threat response. We find that detection performance using energy- and shape-based detection algorithms improves when using RPCA preprocessing.

The work is devoted to illegal materials detection via tagged neutron method (TNM). The detection of hazardous substances is based on recording of gamma radiation from a neutron-irradiated object and analysis of its elemental composition. As against other neutron radiation methods the TNM enables to obtain 3D distribution of elements in the inspected area. The results of experimental part of the research show operational capabilities (probabilities of missing and false alarm) of a portable TNM inspection system when inspecting small hand-luggage-type objects.

The ability to perform remote forensics in situ is an important application of autonomous undersea vehicles (AUVs). Forensics objectives may include remediation of mines and/or unexploded ordnance, as well as monitoring of seafloor infrastructure. At JHU/APL, digital holography is being explored for the potential application to underwater imaging and integration with an AUV. In previous work, a feature-based approach was developed for processing the holographic imagery and performing object recognition. In this work, the results of the image processing method were incorporated into a Bayesian framework for autonomous path planning referred to as information surfing. The framework was derived assuming that the location of the object of interest is known a priori, but the type of object and its pose are unknown. The path-planning algorithm adaptively modifies the trajectory of the sensing platform based on historical performance of object and pose classification. The algorithm is called information surfing because the direction of motion is governed by the local information gradient. Simulation experiments were carried out using holographic imagery collected from submerged objects. The autonomous sensing algorithm was compared to a deterministic sensing CONOPS, and demonstrated improved accuracy and faster convergence in several cases.

This paper addresses the planning of multiple collaborative searchers that are seeking to find hidden objects (i.e. mines) in environments where the sensor detection process is prone to false alarms. In such situations it is anticipated that collaboration between searchers that are examining the same sub-regions may be used to mitigate the impact of false alarms. A standard Receiver Operator Characteristic (ROC) analysis is conducted and the mapping between a single search pass ROC curve and an equivalent multiple search pass representation within a cumulative probability space is discussed. This mapping produces an analogous family of ROC curves for an increasing number of search passes using either a first detection or multiple occurrence performance criteria. The migration of ROC operating points is analyzed as additional search passes are included within a search plan and suggests the need to coordinate search effort with operating point selection. The mapping from waiting time event probabilities to a total error performance criterion weighted according to the cumulative probabilities of missed detection and false alarm is developed. Details of its application for threshold optimization within search planning is discussed and numerical results are provided to demonstrate the usefulness of the models in evaluating performance trade-offs.

Typically, the detection of an object of interest improves as we view the object from multiple angles. For cases where viewing angle matters, object detection can be improved further by optimally selecting the relative angles of multiple views. This motivates the search for viewing angles that maximize the expected probability of detection. Although our work is motivated by applications in subsea sensing, our fundamental analysis is easily adapted for other classes of applications. The specific challenge that motivates our work is the selection of optimal viewing angles for subsea sensing in which sonar is used for bathymetric imaging.

This paper proposes a possibilistic context identification approach for synthetic aperture sonar (SAS) imagery. SAS seabed imagery can display a variety of textures that can be used to identify seabed types such as sea grass, sand ripple and hard-packed sand, etc. Target objects in SAS imagery often have varying characteristics and features due to changing environmental context. Therefore, methods that can identify the seabed environment can be used to assist in target classification and detection in an environmentally adaptive or context-dependent approach. In this paper, a possibilistic context identification approach is used to identify the seabed contexts. Alternative methods, such as crisp, fuzzy or probabilistic methods, would force one type of context on every sample in the imagery, ignoring the possibility that the test imagery may include an environmental context that has not yet appeared in the training process. The proposed possibilistic approach has an advantage in that it can both identify known contexts as well as identify when an unknown context has been encountered. Experiments are conducted on a collection of SAS imagery that display a variety of environmental features.

There is a desire in the Mine Counter Measure community to develop a systematic method to predict and/or estimate the performance of Automatic Target Recognition (ATR) algorithms that are detecting and classifying mine-like objects within sonar data. Ideally, parameters exist that can be measured directly from the sonar data that correlate with ATR performance. In this effort, two metrics were analyzed for their predictive potential using high frequency synthetic aperture sonar (SAS) images. The first parameter is a measure of contrast. It is essentially the variance in pixel intensity over a fixed partition of relatively small size. An analysis was performed to determine the optimum block size for this contrast calculation. These blocks were then overlapped in the horizontal and vertical direction over the entire image. The second parameter is the one-dimensional K-shape parameter. The K-distribution is commonly used to describe sonar backscatter return from range cells that contain a finite number of scatterers. An Ada-Boosted Decision Tree classifier was used to calculate the probability of classification (P_c) and false alarm rate (FAR) for several types of targets in SAS images from three different data sets. ROC curves as a function of the measured parameters were generated and the correlation between the measured parameters in the vicinity of each of the contacts and the ATR performance was investigated. The contrast and K-shape parameters were considered separately. Additionally, the contrast and K-shape parameter were associated with background texture types using previously labeled high frequency SAS images.

The art gallery problem (AGP) asks the question: “How can we place a small set of sensors to provide maximum coverage of an observed environment?” The watchman route problem (WRP) operates in conjunction with the AGP by asking the question “How do we create the shortest route between AGP-solving positions?” The objective of this work is to provide a means of assessing where to place both static and mobile sensors in order to solve the AGP and WRP, respectively, while adapting subsequent AGP/WRP-solutions in anticipation of future events. We can fulfill this objective by 1) extracting a 3D point cloud representation of the item of interest (IOI) to be surveiled in a video frame, 2) determine highest probability anticipated behavior by the IOI based upon training data and 3) incorporate the information gained from items 1 and 2 in order to obtain approximate solutions to the AGP and WRP using the respective Sensor Placement Optimization via Queries (SPOQ) and the Photon-mapping-Informed active-Contour Route Designator (PICRD) algorithms. In this paper, we show how to obtain the requirements embodied in items 1, 2 and 3 and thus fulfill our objective.

In practice, environmental information about an ocean bottom area to be searched using SONAR is often known a priori to some coarse level of resolution. The SONAR search sensor then typically has a different performance characterization function for each environmental classification. Large ocean bottom surveys using search SONAR can pose some difficulties when the environmental conditions vary significantly over the search area because search planning tools cannot adequately segment the area into sub-regions of homogeneous search sensor performance. Such segmentation is critically important to unmanned search vehicles; homogenous bottom segmentation will result in more accurate predictions of search performance and area coverage rate. The Naval Surface Warfare Center, Panama City Division (NSWC PCD) has developed an automated area segmentation algorithm that subdivides the mission area under the constraint that the variation of the search sensor’s performance within each sub-mission area cannot exceed a specified threshold, thereby creating sub-regions of homogeneous sensor performance. The algorithm also calculates a new, composite sensor performance function for each sub-mission area. The technique accounts for practical constraints such as enforcing a minimum sub-mission area size and requiring sub-mission areas to be rectangular. Segmentation occurs both across the rows and down the columns of the mission area. Ideally, mission planning should consider both segmentation directions and choose the one with the more favorable result. The Automated Area Segmentation Algorithm was tested using two a priori bottom segmentations: rectangular and triangular; and two search sensor configurations: a set of three bi-modal curves and a set of three uni-modal curves. For each of these four scenarios, the Automated Area Segmentation Algorithm automatically partitioned the mission area across rows and down columns to create regions with homogeneous sensor performance. The testing results indicated that the algorithm correctly segmented the rectangular a priori regions. For the triangular a priori segmentation, the algorithm created reasonable rectangular sub-areas.