This PDF file contains the front matter associated with SPIE Proceedings Volume 7664, including the Title Page, Copyright information, Table of Contents,and the Conference Committee listing.

The detection of unexploded ordnance (UXO) in the electromagnetic induction regime often suffers from a low signal to noise ratio due to the strong decay of the magnetic field. As a result, a deep UXO may be overshadowed by smaller yet shallower metal items which render the classification of the main target challenging. It is therefore desirable to have the ability to model the various sources of noise and to include them in a detection algorithm. Toward this effect, we investigate here Kalman and extended Kalman filters for the inversion of UXO polarizabilities and positions, respectively, within a dipole model approximation. Inherent to the method, our analysis is based on the assumption of Gaussian noise distribution, which is often reasonable. Results are shown on both synthetic and TEMTADS data which have been purposely corrupted with noise. In particular, the situation of a main target in the presence of dense clutter is investigated, whereby the clutter is composed of 16 nosepieces buried close to the sensor.

The Strategic Environmental Research and Development Program (SERDP) is administering benchmark blind tests of increasing realism to the UXO community. One of the latest took place at Aberdeen Proving Ground in Maryland: 214 cells, each one containing at most one buried target, were interrogated with the TEMTADS electromagnetic induction (EMI) sensor array. Each item could be one of six standard ordnance or could be harmless clutter such as shrapnel. The test called for singling out potentially dangerous items and classifying them. Our group divided the task into three steps: location, characterization, and classification. For the first step the HAP method was used. The method assumes a pure dipolar response from the target and finds the position and orientation using the measured field and its associated scalar potential, the latter computed using a layer of equivalent sources. For target characterization we used the NSMS model, which employs an ensemble of dipole sources arranged on a spheroidal surface. The strengths of these sources are normalized by the primary field that strikes them; their surface integral is an electromagnetic signature that can be used as a classifier. In this work we look into automating the classification step using a multi-category support vector machine (SVM). The algorithm runs binary SVMs for every combination of pairs of target candidates, apportions votes to the winners, and assigns unknown examples to the category with the most votes. We look for the feature combinations and SVM parameters that result in the most expedient and accurate classification.

Discrimination studies carried out on TEMTADS and Metal Mapper blind data sets collected at the San Luis Obispo UXO site are presented. The data sets included four types of targets of interest: 2.36" rockets, 60-mm mortar shells, 81-mm projectiles, and 4.2" mortar items. The total parameterized normalized magnetic source (NSMS) amplitudes were used to discriminate TOI from metallic clutter and among the different hazardous UXO. First, in object's frame coordinate, the total NSMS were determined for each TOI along three orthogonal axes from the training data provided by the Strategic Environmental Research and Development Program (SERDP) along with the referred blind data sets. Then the inverted total NSMS were used to extract the time-decay classification features. Once our inversion and classification algorithms were tested on the calibration data sets then we applied the same procedure to all blind data sets. The combined NSMS and differential evolution algorithm is utilized for determine the NSMS strengths for each cell. The obtained total NSMS time-decay curves were used to extract the discrimination features and perform classification using the training data as reference. In addition, for cross validation, the inverted locations and orientations from NSMS-DE algorithm were compared against the inverted data that obtained via the magnetic field, vector and scalar potentials (HAP) method and the combined dipole and Gauss-Newton approach technique. We examined the entire time decay history of the total NSMS case-by-case for classification purposes. Also, we use different multi-class statistical classification algorithms for separating the dangerous objects from non hazardous items. The inverted targets were ranked by target ID and submitted to SERDP for independent scoring. The independent scoring results are presented.

Current time-domain electromagnetic induction instruments generally only utilize data acquired after the cessation of the transmitted field. During this "off time", signals are dominated by induced eddy currents and magnetic surface modes, but do not fully capture the magnetostatic response of permeable and conductive metallic ordnance. In this paper, we investigate the response of EMI systems that measure signals during excitation of the primary magnetic field (the so-called "on-time"). Our analysis shows that on-time signals have great potential to yield useful information that is not often exploited in current EMI systems. We compare analytical models to data from state-of-the-art time-domain EM sensors that have the capability to sample receivers during the on-time. We present modeling results that represent the responses from different current ramps and on-time waveforms for objects and ground. We consider target and clutter objects and grounds having a range of material properties, shapes and sizes, and configurations and investigate signal processing and inversion methods for target detection and discrimination. Specifically, correlations between on-time and off-time signals are shown to be a powerful tool for discriminating ferrous and non-ferrous metallic objects.

State-of-the-art electromagnetic induction (EMI) arrays provide significant capability enhancement to landmine, unexploded ordnance (UXO), and buried explosives detection applications. Arrays that are easily configured for integration with a variety of mobile platforms offer improved safety and efficiency to personnel conducting detection operations including site remediation, explosive ordnance disposal, and humanitarian demining missions. We present results from an evaluation of two vehicle-based frequency domain EMI arrays. Our research includes implementation of a simple circuit model to estimate target location from sensor measurements of the scattered vertical magnetic field component. Specifically, we characterize any conductive or magnetic target using a set of parameters that describe the eddy current and magnetic polarizations induced about a set of orthogonal axes. Parameter estimations are based on the fundamental resonance mode of a series inductance and resistance circuit. This technique can be adapted to a variety of EMI array configurations, and thus offers target localization capabilities to a number of applications.

Discrimination between UXO and harmless objects is particularly difficult in highly contaminated sites where two or more objects are simultaneously present in the field of view of the sensor and produce overlapping signals. The first step in overcoming this problem is estimating the number of targets. In this work an orthonormalized volume magnetic source (ONVMS) approach is introduced for estimating the number of targets, along with their locations and orientations. The technique is based on the discrete dipole approximation, which distributes dipoles inside the computational volume. First, a set of orthogonal functions are constructed using fundamental solutions of the Helmholtz equations (i.e., Green's functions). Then, the scattered magnetic field is approximated as a series of these orthogonal functions. The magnitudes of the expansion coefficients are determined directly from the measurement data without solving an ill-posed inverse-scattering problem. The expansion coefficients are then used to determine the amplitudes of the responding volume magnetic dipoles. The algorithm's superior performance and applicability to live UXO sites are illustrated by applying it to the bi-static TEMTADS multi-target data sets collected by NRL personnel at the Aberdeen Proving Ground UXO teststand site.

The physically complete Normalized Surface Magnetic Source (NSMS) model and a variant of the simple dipole model are applied to new-generation electromagnetic induction (EMI) data. The main objective is to assess the NSMS and dipole models' capabilities to discriminate between UXO and clutter starting from scattered EMI signals. The discrimination contains two sets of parameters: (1) intrinsic parameters associated with the size, shape, and material composition of the target; and (2) extrinsic parameters related to the orientation and location of the anomaly. To discriminate UXO from clutter a mathematical model is fit to the geophysical data, after which both intrinsic and extrinsic parameters are extracted using an optimization technique. The inverted intrinsic parameters thus found are used to isolate objects of interest from non-hazardous items. The discrimination performance depends significantly on the mathematical model. In this work we present results of applying the single dipole, multi-dipole, and NSMS models to single- and multi-axis sensor data produced by new-generation EMI instruments such as MPV, TEMTADS, and MetalMapper, all of which are are time-domain systems. The MPV has a single transmitter and five tri-axial receivers, the TEMTADS array is a towed system featuring 25 transmitter/receiver pairs, and MetalMapper contains three rectangular transmitters and five tri-axial receivers distributed on a plane. The inversion and discrimination performance of the NSMS and single-dipole models are illustrated for the high-quality, well-located EMI data produced by these instruments. Specifically, we present comparisons between inverted intrinsic and extrinsic parameters, as determined from each model and compared with the ground truth.

An algorithm is proposed to enumerate, locate and characterize individual signal sources given observation of their combined signals. No a-priori estimate for the number of sources is required. We assume a forward model exists, and that superposition holds, i.e. coupling between sources is ignored. A system of linear equations y=Ax is set up in which columns of matrix A contain expected signals from a large number of hypothesized sources, and y contains the observed signal. Recently-developed solvers designed for linear systems with sparse non-negative solutions make this approach feasible even when large numbers of sources are involved. With each iteration, the collection of hypothesized sources is refined using a Harmony Search algorithm. Application is demonstrated on the problem of locating multiple buried conductors based on electromagnetic induction (EMI) signals observed at ground surface.

This paper reports vehicle based electromagnetic induction (EMI) array sensor data inversion and discrimination results. Recent field studies show that EMI arrays, such as the Minelab Single Transmitter Multiple Receiver (STMR), and the Geophex GEM-5 EMI array, provide a fast and safe way to detect subsurface metallic targets such as landmines, unexploded ordnance (UXO) and buried explosives. The array sensors are flexible and easily adaptable for a variety of ground vehicles and mobile platforms, which makes them very attractive for safe and cost effective detection operations in many applications, including but not limited to explosive ordnance disposal and humanitarian UXO and demining missions. Most state-of-the-art EMI arrays measure the vertical or full vector field, or gradient tensor fields and utilize them for real-time threat detection based on threshold analysis. Real field practice shows that the threshold-level detection has high false alarms. One way to reduce these false alarms is to use EMI numerical techniques that are capable of inverting EMI array data in real time. In this work a physically complete model, known as the normalized volume/surface magnetic sources (NV/SMS) model is adapted to the vehicle-based EMI array, such as STMR and GEM-5, data. The NV/SMS model can be considered as a generalized volume or surface dipole model, which in a special limited case coincides with an infinitesimal dipole model approach. According to the NV/SMS model, an object's response to a sensor's primary field is modeled mathematically by a set of equivalent magnetic dipoles, distributed inside the object (i.e. NVMS) or over a surface surrounding the object (i.e. NSMS). The scattered magnetic field of the NSMS is identical to that produced by a set of interacting magnetic dipoles. The amplitudes of the magnetic dipoles are normalized to the primary magnetic field, relating induced magnetic dipole polarizability and the primary magnetic field. The magnitudes of the NSMS are determined directly by minimizing the difference between measured and modeled data for any known object and any type of EMI sensor data. The EMI array data are inverted via the combined NV/SMS and differential evolution inversion method that uses a search scheme to estimate the location of the target. First, the applicability of the NV/SMS and DE algorithms to STMR and GEM-5 data sets is demonstrated by comparing the modeled data against the actual data, and finally the discrimination studies are conducted using as discrimination parameters the total NV/SMS and the principal axis of the induced magnetic polarizability tensor for each target.

Two vehicle mounted metal detector arrays are used in conjunction to perform object classification. The first array (Vallon VMV-16) contains small coils for detecting shallow targets. The second array (Minelab STMR II) contains receive coils of roughly the same size, but a single large transmitter for detecting deep targets. These two sensors are used together to classify objects as: "SHALLOW and LARGE","DEEP and LARGE", or "SHALLOW and SMALL". SHALLOW/DEEP implies the depth of the object; SMALL/LARGE implies the metal content. These object classes are further specified within the paper. An experiment is performed using unexploded ordnance (UXO) and shallow buried calibration objects. The UXO ranges in depth from flush buried to 48". The calibration targets consist of metallic cylinders ranging in depth from flush buried to 12". The strength of each sensor is described and a fusion algorithm is developed. A detection performance curve is shown illustrating the benefit of multi-sensor fusion for UXO detection.

In subsurface UXO sensing, single field (SF) data arises when an excitation field produces a single, spatially distributed response field that is sampled from a number of different locations. Measurements from traditional magnetometers furnish such data, for which the essentially invariant excitation is the earth's magnetic field. Upward continuation (UC) of SF data allows one to calculate signals that would be received at a higher elevation above the ground without actually raising the receiver. This is done without having to solve for the actual target characteristics or location. The technique is designed to smooth out the perturbations from irregularities and near-surface clutter. Applied recently to broader band electromagnetic induction (EMI) data of the the GAP SAM system, UC has shown distinct benefits in suppressing the strength of near surface clutter signals relative to those from a deeper UXO. Here we investigate possible application to the TEMTADS sensor. Preliminary results suggest that here too the method may bring out the signal of an underlying larger UXO relative to discrete clutter or smaller shallowers items, thereby aiding discrimination. In future work resolution issues must be addressed.

Frequency-domain electromagnetic induction (EMI) sensors have the ability to measure target signatures which enable discrimination of landmines from harmless clutter. In a model-based signal processing paradigm, 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. The basis function parameters can then be used as features for classification of the target as landmine or clutter. One of the challenges associated with effectively utilizing frequency-domain EMI sensor data within a model-based signal processing paradigm such as this is determining the correct model order for the measured data, as the number of basis functions intrinsic to the target under consideration is not known a priori. In this work, relevance vector machine (RVM) regression is applied to simultaneously determine both the number of parameterized basis functions and their relative contributions to the measured signal. The target may then be classified utilizing the basis function parameters as features within a statistical classifier. Results for data measured with a prototype frequency-domain EMI sensor at a standardized test site are presented. Preliminary results indicate that RVM regression followed by statistical classification utilizing the resulting model-based features provides an effective approach for classifying targets as landmine or clutter.

Broadband electromagnetic induction (EMI) sensors have been shown to be able to reduce false alarm rates and increase the probability of detecting landmines. To aid in the development of these sensors and associated detection algorithms, a testing facility and inversion technique have been developed to characterize the response of typical targets and clutter objects as a function of orientation and frequency. The models are simple sets of magnetic dipoles with discrete relaxation frequencies. Results will be presented for a range of targets such as shell casings, wire loops, and landmines. It is envisioned that the models derived in this work will be utilized to reduce false alarm rates and increase the probability of detection for EMI sensors through improvements in both the hardware and the processing algorithms used to detect and discriminate buried targets.

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 in the sum is unknown. In this paper, the estimation problem is reformulated into a linear system and model parameters are estimated through a modified Lp-regularized least squares algorithm for 0 <= p <= 1. Using tests on synthetic data and laboratory measurement of known targets the proposed method is shown to provide satisfactory and stable estimates. In addition, the proposed method does not require a good initial guess for convergence.

The spectral emissivity of soils in the region of thermal emission from 8 - 14 micrometers is a combination of the spectral emission of the mineral and other components in the soil, as well as their physical arrangement and the thermal state of the soil (presence of thermal gradients). In this paper, we will outline the procedure for producing a spectral model of a mixed soil, and show examples of model soils compared to measured soils with the two major soil constituents: quartz and clay. The predictions of this theory are then compared to field measurements made with a LWIR Spectrometer of disturbed and undisturbed soil.

The multi-optical mine detection system (MOMS) is a research project focused on the detection of surface laid mines. In the sensor suite, both passive and active sensors are included, such as IR as well as hyper- and multispectral cameras, and 3-D laser radar. Extensive field experiments have been conducted to collect data under various environmental conditions. Three seasons have been covered during the field campaigns: Spring, summer, and autumn. Furthermore, the mines have been arranged in three different types of vegetation scenarios. Also, a long term data collection effort has been conducted to collect diurnal and seasonal signature variations. Among the signal processing techniques considered, anomaly detection emerges as a key component in a system concept. The method is based on detecting small differences between the mine-sized object and a local background. The spectral features of the detected anomalies are further analyzed with respect to general commonalities in the scene and known spectral properties of mine-like objects. In this paper we present some of the results from the project.

In this paper, a study involving the detection of buried objects by fusing airborne Multi-Spectral Imagery (MSI) and ground-based Ground Penetrating Radar (GPR) data is investigated. The benefit of using the airborne sensor to cue the GPR, which will then search the area indicated by the MSI, is investigated and compared to results obtained via a purely ground-based system. State-of-the-art existing algorithms, such as hidden Markov models will be applied to the GPR data both in queued and non-queued modes. In addition, the ability to measure disturbed earth with the GPR sensor will be investigated. Furthermore, state-of-theart algorithms for the MSI system will be described. These algorithms require very high detection rates with acceptable false alarm rates in order to serve as an acceptable system. Results will be presented on data collected at outdoor testing and evaluation sites.

We present a comparative study involving five distinctly different polarimetric imaging platforms that are designed to record calibrated Stokes images (and associated polarimetric products) in either the MidIR or LWIR spectral regions. The data set used in this study was recorded during April 14-18, 2008, at the Russell Tower Measurement Facility, Redstone Arsenal, Huntsville, AL. Four of the five camera systems were designed to operate in the LWIR (approx. 8-12μm), and used either cooled mercury cadmium telluride (MCT) focal-plane-arrays (FPA), or a near-room temperature microbolometer. The lone MidIR polarimetric sensor was based on a liquid nitrogen (LN2) cooled indium antimonide (InSb) FPA, resulting in an approximate wavelength response of 3-5μm. The selection of cameras was comprised of the following optical designs: a LWIR "super-pixel," or division-of-focal plane (DoFP) sensor; two LWIR spinning-achromatic-retarder (SAR) based sensors; one LWIR division-of-amplitude (DoAM) sensor; and one MidIR division-of-aperture (DoA) sensor. Cross-sensor comparisons were conducted by examining calibrated Stokes images (e.g., S0, S1, S2, and degree-of-linear polarization (DOLP)) recorded by each sensor for a given target at approximately the same test periods to ensure that data sets were recorded under similar atmospheric conditions. Target detections are applied to the image set for each polarimetric sensor for further comparison, i.e., conventional receiver operating characteristic (ROC) curve analysis and an effective contrast ratio are considered.

Disturbance of ground surfaces can arise from a variety of processes, both manmade and natural. Burying landmines, vehicle movement, and walking are representative examples of processes that disturb ground surfaces. The nature of the specific disturbance process can lead to the observables that can aid the detection and identification of that process. While much research has been conducted in this area, fundamental questions related to the remote detection and characterization of disturbed soil surfaces remain unanswered. Under the sponsorship of the Army Research Office (ARO), the Night Vision and Electronic Sensors Directorate (NVESD), and the U.S. Army Corps of Engineers (USACE) Engineering Research and Development Center (ERDC), Georgia Tech hosted a workshop to address Remote Sensing Methods for Disturbed Soil Characterization. The workshop was held January 15-17, 2008 in Atlanta. The primary objective of this workshop was to take a new look at the disturbed soil problem in general as well as its relation to buried explosive detection and other manmade disturbances. In particular, the participants sought to outline the basic science and technology questions that need to be addressed across the full spectrum of military applications to fully exploit this phenomenon. This presentation will outline the approach taken during the workshop and provide a summary of the conclusions.

In this paper a new detection method for sonar imagery is developed in K-distributed background clutter. The equation for the log-likelihood is derived and compared to the corresponding counterparts derived for the Gaussian and Rayleigh assumptions. Test results of the proposed method on a data set of synthetic underwater sonar images is also presented. This database contains images with targets of different shapes inserted into backgrounds generated using a correlated K-distributed model. Results illustrating the effectiveness of the K-distributed detector are presented in terms of probability of detection, false alarm, and correct classification rates for various bottom clutter scenarios.

An improved automatic target recognition (ATR) processing string has been developed. The overall processing string consists of pre-processing, subimage adaptive clutter filtering, detection, feature extraction, optimal subset feature selection, feature orthogonalization and classification processing blocks. The objects that are classified by three distinct ATR strings are fused using the classification confidence values and their expansions as features, and using "summing" or log-likelihood-ratio-test (LLRT) based fusion rules. These three ATR processing strings were individually developed and tuned by researchers from different companies. The utility of the overall processing strings and their fusion was demonstrated with an extensive side-looking sonar dataset. In this paper we describe a new processing improvement: six additional classification features are extracted, using primarily target shadow information and a feature extraction window whose length is now made variable as a function of range. This new ATR processing improvement resulted in a 3:1 reduction in false alarms. Two advanced fusion algorithms are subsequently applied: First, a nonlinear Volterra expansion (2nd order) feature-LLRT fusion algorithm is employed. Second, a repeated application of a subset Volterra feature selection / feature orthogonalization / LLRT fusion block is utilized. It is shown that cascaded Volterra feature- LLRT fusion of the ATR processing strings outperforms baseline "summing" and single-stage Volterra feature-LLRT fusion algorithms, yielding significant improvements over the best single ATR processing string results, and providing the capability to correctly call the majority of targets while maintaining a very low false alarm rate.

This paper describes an approach to utilize a multi-channel, multi-spectral electro-optic (EO) system for littoral zone characterization. Advanced Coherent Technologies, LLC (ACT) presents their EO sensor systems for the surf zone environmental assessment and potential surf zone target detection. Specifically, an approach is presented to determine a Surf Zone Index (SZI) from the multi-spectral EO sensor system. SZI provides a single quantitative value of the surf zone conditions delivering an immediate understanding of the area and an assessment as to how well an airborne optical system might perform in a mine countermeasures (MCM) operation. Utilizing consecutive frames of SZI images, ACT is able to measure variability over time. A surf zone nomograph, which incorporates targets, sensor, and environmental data, including the SZI to determine the environmental impact on system performance, is reviewed in this work. ACT's electro-optical multi-channel, multi-spectral imaging system and test results are presented and discussed.

In order to demonstrate the possibility of identifying the material within the objects on the sea floor we have performed tests with the 14 MeV sealed tube neutron generator incorporated inside a small submarine, SURVEYOR, submerged in the test basin filled with sea water. The materials inside the investigated objects were identified by measuring the gamma ray spectra and by using the window on the measured alpha- gamma time spectrum. In addition, we describe our field test facility and measurements done at this location in the framework of the EU FP7 project UNCOSS. The existence of a data base of potentially explosive devices on the floor of coastal sea, especially ports and waterways, is of paramount importance. However, the sea floor is littered by number of different objects and the water is not very transparent on such locations. This makes the identification of objects extremely difficult even on known locations. We discuss how to position the SURVEYOR when the object investigated for the presence of explosive has been identified by other sensors (camera, sonar, magnetometer, etc.).

Over the past decade, notable progress has been made in the performance of airborne geophysical systems for mapping and detection of unexploded ordnance in terrestrial and shallow marine environments. For magnetometer systems, the most significant improvements include development of denser magnetometer arrays and vertical gradiometer configurations. In prototype analyses and recent Environmental Security Technology Certification Program (ESTCP) assessments using new production systems the greatest sensitivity has been achieved with a vertical gradiometer configuration, despite model-based survey design results which suggest that dense total-field arrays would be superior. As effective as magnetometer systems have proven to be at many sites, they are inadequate at sites where basalts and other ferrous geologic formations or soils produce anomalies that approach or exceed those of target ordnance items. Additionally, magnetometer systems are ineffective where detection of non-ferrous ordnance items is of primary concern. Recent completion of the Battelle TEM-8 airborne time-domain electromagnetic system represents the culmination of nearly nine years of assessment and development of airborne electromagnetic systems for UXO mapping and detection. A recent ESTCP demonstration of this system in New Mexico showed that it was able to detect 99% of blind-seeded ordnance items, 81mm and larger, and that it could be used to map in detail a bombing target on a basalt flow where previous airborne magnetometer surveys had failed. The probability of detection for the TEM-8 in the blind-seeded study area was better than that reported for a dense-array total-field magnetometer demonstration of the same blind-seeded site, and the TEM-8 system successfully detected these items with less than half as many anomaly picks as the dense-array total-field magnetometer system.

We assess the noise level caused by marine environments in underwater UXO discrimination studies. Underwater UXO detection and discrimination is subject to additional noise sources, which are not present in land-based scenarios. Particularly, we study the effects of water surface roughness on the diffusion of EMI (electromagnetic induction) fields through the air-water interface and the interaction effects between an underwater conducting object and its surrounding conductive medium. Numerical simulations are conducted using the 3-dimensional setup of the Method of Auxiliary Sources suitable for low-frequency regime. Water surface roughness is modeled as an interference pattern between a finite number of surface waves with varying amplitudes, wavelengths and propagation directions. The results indicate that the perturbations in diffused and scattered EMI fields due to water surface roughness are negligible (although they depend on the shape of water surface) and that these perturbations decay with distance from the interface. Thus, the conducting water body may be assumed to represent a half-space in subsequent calculations for UXO detection. Finally, it is shown that there is some interaction between a conducting object and its surrounding conductive environment for frequencies above 100 kHz. This interaction is attenuated if the object is surrounded by an insulating shell, but is amplified if the shell is conducting. This non-negligible effect needs to be taken into account for the purposes of UXO detection and discrimination.

The ability to reliably detect targets having signatures comprised of bright pixels (highlight) and dark pixels (shadow) is challenging when the background texture of the imagery also possesses bright and dark characteristics. This is especially difficult when the background contains large bright and dark areas that can mask target signatures. Detection and classification algorithms would benefit from an adaptive denoising algorithm that would remove or mitigate such background artifacts. This paper presents a Fourier-based denoising algorithm. The large support of the Fourier basis is used to capture and remove large-scale artifacts while leaving the smaller target-size features nearly unchanged. Datadriven soft thresholds allow the algorithm to automatically adapt to changing backgrounds. Preliminary investigations have demonstrated excellent performance. The algorithm is computationally fast and suitable for real-time application. The denoising algorithm is general in nature and can be applied to many types of high-resolution gray-scale imagery; e.g., side-looking sonar and SAR.

An approach to simulate synthetic aperture sonar (SAS) images with known autocorrelation functions (ACF) and single-point statistics is presented. ACF models for generating textures with and without periodicities are defined and explained. Simulated textures of these models are compared visually with real SAS image textures. Distortion and degradation of the synthetic textures are examined for various simulation parameter choices.

Raytheon has extensively processed high-resolution sidescan sonar images with its Automatic Target Recognition (ATR) algorithms to classify target-like objects (TLOs) in a variety of underwater environments. The ATR algorithm segments the image into candidate highlight and shadow regions of interest (ROIs), and extracts and scores features from these ROIs. The TLOs are classified by thresholding an overall classification score, formed by summing the individual feature scores. The algorithm performs reliably against TLOs that exhibit highlight and shadow regions that are both distinct relative to the ambient background. However, the sonar images for many real-world undersea environments can contain a significant percentage of TLOs exhibiting either "weak" highlight or shadow regions. Robust performance in these environments is achieved by tailoring the individual feature scoring algorithms to optimize the separation between the corresponding highlight or shadow feature scores of targets and non-targets. This study examines modifications to a previously presented alternate approach that employs Fisher fusion principles to generate optimal weighting coefficients which are applied to the individual feature scores before final classification processing. Results from processing of at-sea data sets demonstrate the performance benefits obtained from the modifications.

The Autonomous Mine Detection System (AMDS) program is developing a landmine and explosive hazards standoff detection, marking, and neutralization system for dismounted soldiers. The AMDS Capabilities Development Document (CDD) has identified the requirement to deploy three payload modules for small robotic platforms: mine detection and marking, explosives detection and marking, and neutralization. This paper addresses the neutralization payload module. There are a number of challenges that must be overcome for the neutralization payload module to be successfully integrated into AMDS. The neutralizer must meet stringent size, weight, and power (SWaP) requirements to be compatible with a small robot. The neutralizer must be effective against a broad threat, to include metal and plastic-cased Anti-Personnel (AP) and Anti-Tank (AT) landmines, explosive devices, and Unexploded Explosive Ordnance (UXO.) It must adapt to a variety of threat concealments, overburdens, and emplacement methods, to include soil, gravel, asphalt, and concrete. A unique neutralization technology is being investigated for adaptation to the AMDS Neutralization Module. This paper will describe review this technology and how the other two payload modules influence its design for minimizing SWaP. Recent modeling and experimental efforts will be included.

NIITEK (Non-Intrusive Inspection Technology, Inc) develops and fields vehicle-mounted mine and buried threat detection systems. Since 2003, the NIITEK has developed and tested a remote robot-mounted mine detection system for use in the NVESD AMDS program. This paper will discuss the road map of development since the outset of the program, including transition from a data collection platform towards a militarized field-ready system for immediate use as a remote countermine and buried threat detection solution with real-time autonomous threat classification. The detection system payload has been integrated on both the iRobot Packbot and the Foster-Miller Talon robot. This brief will discuss the requirements for a successful near-term system, the progressive development of the system, our current real-time capabilities, and our planned upgrades for moving into and supporting field testing, evaluation, and ongoing operation.

Landmines have been laid in conflicts around the world and continue to maim or kill civilians and soldiers. Metal detectors (MD) have been used successfully to detect mines, but have difficulty detecting mines with little or no metal content. Ground penetrating radar (GPR) systems have successfully been used to supply detection capabilities where metal detectors fail. Handheld devices using such sensors have historically been used in battle but they can put the user at high risk under direct fire from the enemy while exposed during some operations. We describe a robotic, explosive hazard, anti-personnel/anti-tank mine detection system featuring dual-sensor GPR/MD capability for enhanced mine detection and for removing the soldier from the mine field. The MD is a broadband electromagnetic induction sensor to help discriminate between buried landmines and metal clutter. The sensor operates in the frequency domain and collects data at 21 logarithmically spaced frequencies from 300 Hz to 90 kHz. The GPR is a broadband stepped frequency continuous wave (SFCW) sensor operating from 700 MHz to 4 GHz in 10 MHz steps. The GPR employs an array of low cross section inverted V-dipoles swept over the scene. The GPR data will also support 3D synthetic aperture radar (SAR) imagery to aid in user target verification.

This paper shall demonstrate the results of a prototype system to detect explosive objects and obscured contaminated targets. By combining a high volume sampling nozzle with an inline 2-stage preconcentrator and a Fido, greater standoff is achieved than with the Fido alone. The direct application of this system is on the Autonomous Mine Detection System (AMDS) but could be deployed on a large variety of robotic platforms. It is being developed under the auspices of the U.S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate, Countermine Division. This device is one of several detection tools and technologies to be used on the AMDS. These systems will have multiple, and at times, overlapping objectives. One objective is trace detection on the surface of an unknown potential target. By increasing the standoff capabilities of the detector, the fine manipulation of the robot deploying the detector is less critical. Current detectors used on robotic systems must either be directly in the vapor plume or make direct contact with the target. By increasing the standoff, detection is more easily and quickly achieved. The end result detector must overcome cross-contamination, sample throughput, and environmental issues. The paper will provide preliminary results of the prototype system to include data, and where feasible, video of testing results.

The increased use of Unmanned Ground Vehicles (UGVs) drives the need for new lightweight, low cost sensors. Microelectromechanical System (MEMS) based microcantilever sensors are a promising technology to meet this need, because they can be manufactured at low cost on a mass scale, and are easily integrated into a UGV platform for detection of explosives and other threat agents. While the technology is extremely sensitive, selectivity is a major challenge and the response modes are not well understood. This work summarizes advances in characterizing ultrasensitive microcantilever responses, sampling considerations, and sensor design and cantilever coating methodologies consistent with UGV point detector needs.

QinetiQ North America (QNA) has approximately 27 years experience in the mine/countermine mission area. Our expertise covers mine development, detection, and neutralization and has always been intertwined with deployment of remote robotic systems. Our countermine payload systems have been used to detect limpet mines on ship hulls, antiassault mines in shallow water and littoral zones and currently for clearance and render safe of land-based routes. In our talk, we will address the challenges encountered in addressing the ongoing countermine mission over a diverse range of operational scenarios, environmental conditions and strategic priorities.

Mine detection is a dangerous and physically demanding task that is very well-suited for robotic applications. In the experiment described in this paper, we try to determine whether a remotely-operated robotic mine detection system equipped with a hand-held mine detector can match the performance of a human equipped with a hand-held mine detector. To achieve this objective, we developed the Robotic Mine Sweeper (RMS). The RMS platform is capable of accurately sweeping and mapping mine lanes using common detectors, such as the Minelab F3 Mine Detector or the AN/PSS-14. The RMS is fully remote controlled from a safe distance by a laptop via a redundant wireless connection link. Data collected from the mine detector and various sensors mounted on the robot are transmitted and logged in real-time to the remote user interface and simultaneously graphically displayed. In addition, a stereo color camera mounted on top of the robot sends a live picture of the terrain. The system plays audio feedback from the detector to further enhance the user's situational awareness. The user is trained to drag and drop various icons onto the user interface map to locate mines and non-mine clutter objects. We ran experiments with the RMS to compare its detection and false alarm rates with those obtained when the user physically sweeps the detectors in the field. The results of two trials: one with the Minelab F3, the other with the Cyterra AN/PSS-14 are presented here.

CMMAD is a risk reduction effort for the AMDS program. As part of CMMAD, multiple instances of semi autonomous robotic mine detection systems were created. Each instance consists of a robotic vehicle equipped with sensors required for navigation and marking, countermine sensors and a number of integrated software packages which provide for real time processing of the countermine sensor data as well as integrated control of the robotic vehicle, the sensor actuator and the sensor. These systems were used to investigate critical interest functions (CIF) related to countermine robotic systems. To address the autonomy CIF, the INL developed RIK was extended to allow for interaction with a mine sensor processing code (MSPC). In limited field testing this system performed well in detecting, marking and avoiding both AT and AP mines. Based on the results of the CMMAD investigation we conclude that autonomous robotic mine detection is feasible. In addition, CMMAD contributed critical technical advances with regard to sensing, data processing and sensor manipulation, which will advance the performance of future fieldable systems. As a result, no substantial technical barriers exist which preclude - from an autonomous robotic perspective - the rapid development and deployment of fieldable systems.

During field trials, operator usability data were collected in support of lane clearing missions and hazard sensing for two robot platforms with Robot Intelligence Kernel (RIK) software and sensor scanning payloads onboard. The tests featured autonomous and shared robot autonomy levels where tasking of the robot used a graphical interface featuring mine location and sensor readings. The goal of this work was to provide insights that could be used to further technology development. The efficacy of countermine and hazard systems in terms of mobility, search, path planning, detection, and localization were assessed. Findings from objective and subjective operator interaction measures are reviewed along with commentary from soldiers having taken part in the study who strongly endorse the system.

The countermine mission (CM) is a compelling example of what autonomous systems must address to reduce risks that Soldiers take routinely. The list of requirements is formidable and includes autonomous navigation, autonomous sensor scanning, platform mobility and stability, mobile manipulation, automatic target recognition (ATR), and systematic integration and control of components. This paper compares and contrasts how the CM is done today against the challenges of achieving comparable performance using autonomous systems. The Soldier sets a high standard with, for example, over 90% probability of detection (Pd) of metallic and low-metal mines and a false alarm rate (FAR) as low as 0.05/m². In this paper, we suggest a simplification of the semi-autonomous CM by breaking it into three components: sensor head maneuver, robot navigation, and kill-chain prosecution. We also discuss the measurements required to map the system's physical and state attributes to performance specifications and note that current Army countermine metrics are insufficient to the guide the design of a semi-autonomous countermine system.

Payloads for small robotic platforms have historically been designed and implemented as platform and task specific solutions. A consequence of this approach is that payloads cannot be deployed on different robotic platforms without substantial re-engineering efforts. To address this issue, we developed a modular countermine payload that is designed from the ground-up to be platform agnostic. The payload consists of the multi-mission payload controller unit (PCU) coupled with the configurable mission specific threat detection, navigation and marking payloads. The multi-mission PCU has all the common electronics to control and interface to all the payloads. It also contains the embedded processor that can be used to run the navigational and control software. The PCU has a very flexible robot interface which can be configured to interface to various robot platforms. The threat detection payload consists of a two axis sweeping arm and the detector. The navigation payload consists of several perception sensors that are used for terrain mapping, obstacle detection and navigation. Finally, the marking payload consists of a dual-color paint marking system. Through the multimission PCU, all these payloads are packaged in a platform agnostic way to allow deployment on multiple robotic platforms, including Talon and Packbot.

This paper reports on a Soldier Experiment performed by the Army Research Lab's Human Research Engineering Directorate (HRED) Field Element located at the Maneuver Center of Excellence, Ft. Benning, and a Limited Use Assessment conducted by the Marine Corps Forces Pacific Command Experimentation Center (MEC) at Camp Pendleton evaluating the effectiveness of using speech commands to control an Unmanned Ground Vehicle. SPEAR, developed by Think-A-Move, Ltd., provides speech control of UGVs. SPEAR detects user speech in the ear canal with an earpiece containing an in-ear microphone. The system design provides up to 30 dB of passive noise reduction, enabling it to work well in high-noise environments, where traditional speech systems, using external microphones, fail; it also utilizes a proprietary speech recognition engine. SPEAR has been integrated with iRobot's PackBot 510 with FasTac Kit, and with Multi-Robot Operator Control Unit (MOCU), developed by SPAWAR Systems Center Pacific. These integrated systems allow speech to supplement the hand-controller for multi-modal control of different UGV functions simultaneously. HRED's experiment measured the impact of SPEAR on reducing the cognitive load placed on UGV Operators and the time to complete specific tasks. Army NCOs and Officer School Candidates participated in this experiment, which found that speech control was faster than manual control to complete tasks requiring menu navigation, as well as reducing the cognitive load on UGV Operators. The MEC assessment examined speech commands used for two different missions: Route Clearance and Cordon and Search; participants included Explosive Ordnance Disposal Technicians and Combat Engineers. The majority of the Marines thought it was easier to complete the mission scenarios with SPEAR than with only using manual controls, and that using SPEAR improved their situational awareness. Overall results of these Assessments are reported in the paper, along with possible applications to autonomous mine detection systems.

This paper discusses robot behaviors and an interaction scheme between the robot and the operator needed to facilitate both the near and long term future AMDS program goals. The behaviors necessary to meet the AMDS goals include guarded motion, shared control driving, ground scanning with height control, and mobile manipulation capabilities such as automated reaching, object scanning and particulate sampling. The Operator Control Unit (OCU) is also discussed together with innovative concepts for interface design including visualization and tasking tools. The paper also discusses how these behaviors can support the near-term manufacture and deployment of payloads to support dismounted combat missions.

Herein we purpose to utilize upconverting phosphors to detect explosives. To detect TNT, antibodies specific to TNT are conjugated to the surface. The role of the antibodies is two fold; to bind a quencher and to accept TNT. The quencher is a bifunctional molecule, with one end containing a TNT analog and the other end being a dark fluorescent quenching dye. The dye is chosen so that the luminescence from the phosphor will be absorbed preventing it from emitting, reducing luminescence from the phosphor. However, in the presence of TNT the quencher that is bound with DNT will be displaced. With the quencher displaced the phosphor will be able to emit light indicating TNT is present in the select area.

In an effort to produce a handheld explosives sensor the Xsense project has been initiated at the Technical University of Denmark in collaboration with a number of partners. Using micro- and nano technological approaches it will be attempted to integrate four detection principles into a single device. At the end of the project, the consortium aims at having delivered a sensor platform consisting of four independent detector principles capable of detecting concentrations of TNT at sub parts-per-billion (ppb) concentrations and with a false positive rate less than 1 parts-per-thousand. The specificity, sensitivity and reliability are ensured by the use of clever data processing , surface functionalisation and nanostructured sensors and sensor surfaces.

Research has been conducted since the 1950s on nuclear methods to confirm the presence of bulk explosives by detecting characteristic emitted radiation. In most practical situations, penetrating radiation is required, which restricts the problem to gamma rays and neutrons. The most successful reactions to date has been radiative thermal neutron capture (thermal neutron analysis) and prompt radiative emission following inelastic fast neutron scattering (fast neutron analysis). An alternative to these neutron-in, gamma ray-out reactions is photoneutron production. A gamma ray whose energy exceeds the threshold for neutron production in a particular atomic nucleus can cause a neutron to be emitted. For a given isotope and assuming monoenergetic photons, the emitted neutrons will have a spectrum consisting of one or more discrete energies. If the gamma ray source and neutron spectrometer are appropriately chosen, the neutron spectrum can be used as a fingerprint to identify the isotope. This photoneutron spectroscopy method has a number of potential advantages over thermal and fast neutron analysis, such as generally simpler spectra and low inherent natural neutron background. It also has drawbacks, such as possible induced neutron background and a present lack of suitable fieldable photon sources. This paper will describe the method and preliminary simulations and calculations to examine its feasibility. Possible sources, detectors and geometries will be discussed.

A technique for stand-off detection of trace explosives using infrared (IR) photo-thermal (PT) imaging, remote explosives detection (RED), is under development at the Naval Research Laboratory. In this approach, compact IR quantum cascade lasers (QCLs) tuned to strong absorption bands of trace explosives illuminate a surface of interest. An IR focal plane array is used to image the surface and detect any small increase in the thermal emission upon laser illumination. The technique has been previously demonstrated at several meters of stand-off distance indoors and in field tests with sensitivity to explosive traces as small as a single grain (~1 ng), while operating the lasers below the eye-safe intensity limit (100 mW/cm²) at the tested wavelengths. By varying the incident wavelength slightly, selectivity between TNT and RDX has been achieved. A complete test and analysis can be performed in less than 1 second. This manuscript critically examines components used with RED and demonstrates several improvements. These include QCL drive electronics for narrower spectral emission linewidth, fixed wavelength QCL packaging that optimizes spectral and spatial output, fiber-optic coupling for QCL beam steering and spatial filtering, cooled IR sensors that increase sensitivity and speed, tunable QCL sources that increase selectivity and extend the library of possible analytes, and dynamic PT signal processing that can increase sensitivity and speed. When considered in combination with the capabilities previously demonstrated for RED, and its capability to operate within eye-safety limits, this technology offers the potential for a wide area of applications relating to the detection of trace explosives on surfaces in both non-contact and stand-off configurations.

This paper gives a brief overview on our latest progress in the area of standoff detection. Standoff Raman measurements from 200 m and 470 m distance have been performed on bulk amounts of TATP and AN respectively, the former through a double sided window, the latter under heavy rain. Resonance Raman measurements on TNT, DNT and NM vapors in the ppm concentration regime are presented, showing resonance enhancement in the range of 2 200 (NM) to 57 000 (TNT) as compared to 532 nm Raman cross sections. Finally, the application of hyper spectral Raman imaging is described, exemplified by the resolution of four different samples (sulphur, AN, DNT, and TNT) in the form of 5 mm wide discs in one single image.

TNT is released into the soil from many different sources, especially from military and mining activities, including buried land mines. Vegetation may absorb explosive residuals, causing stress and by understanding how plants respond to energetic compounds, we may be able to develop non-invasive techniques to detect soil contamination. The objectives of our study were to examine the physiological response of plants grown in TNT contaminated soils and to use remote sensing methods to detect uptake in plant leaves and canopies in both laboratory and field studies. Differences in physiology and light-adapted fluorescence were apparent in laboratory plants grown in N enriched soils and when compared with plants grown in TNT contaminated soils. Several reflectance indices were able to detect TNT contamination prior to visible signs of stress, including the fluorescence-derived indices, R740/R850 and R735/R850, which may be attributed to transformation and conjugation of TNT metabolites with other compounds. Field studies at the Duck, NC Field Research Facility revealed differences in physiological stress measures, and leaf and canopy reflectance when plants growing over suspected buried UXOs were compared with reference plants. Multiple reflectance indices indicated stress at the suspected contaminated sites, including R740/R850 and R735/R850. Under natural conditions of constant leaching of TNT into the soil, TNT uptake would be continuous in plants, potentially creating a distinct signature from remotely sensed vegetation. We may be able to use remote sensing of plant canopies to detect TNT soil contamination prior to visible signs.

The principal possibility to recognize liquid explosives and their components in various glass and plastic containers with different transparency in visible spectral range was demonstrated. Acetone was used as a target, as alone and mixed with traditional liquids. The advantage of gated Raman spectroscopy over the CW was proved. It was found that using 532 nm, 6 ns laser pulses any real target with characteristic Raman spectrum with intensity similar to those for acetone may be detected in 100 % of glass and in 80 % of plastic containers. The mixing with different liquid makes detection more difficult and acetone was detected in 55 % of studied cases. The main reasons for detection difficulties are intrinsic Raman and luminescence of plastic containers and liquids relevant to airport passengers. In case of strong luminescence the advantages of red light excitation over green light was demonstrated.

Soil moisture conditions have an impact upon virtually all aspects of Army activities and are increasingly affecting its systems and operations. Soil moisture conditions affect operational mobility, detection of landmines and unexploded ordinance, military engineering activities, blowing dust and sand, watershed responses, and flooding. This study explores a novel method for high-resolution (2.7 m) soil moisture mapping using remote satellite optical imagery that is readily available from Landsat and QuickBird. The soil moisture estimations are needed for the evaluation of sensors for Improvised Explosive Devices (IEDs) using the Countermine Simulation Test Bed in regions where access is denied. The method has been tested in Helmand Province, Afghanistan, using a Landsat7 and a QuickBird image of April 23 and 24, 2009, respectively. The first implementation of the method yielded promising results.

ALIS is a hand-held dual sensor developed by Tohoku University, Japan since 2002. Dual sensor is a general name of sensor for humanitarian demining, which are equipped with metal detector and GPR. ALIS is only one hand-held dual sensor, which can record the sensor position with sensor signals. Therefore, the data can be processed after data acquisition, and can increase the imaging capability. ALIS has been tested in some mine affected courtiers including Afghanistan (2004), Egypt(2005), Croatia(2006-) and Cambodia(2007-). Mine fields at each country has different conditions and soil types. Therefore testes at the real mine fields are very important. ALIS has detected more than 30 AP-Mines in evaluation test in Cambodia held in 2009.

In this paper hand-held dual sensor detector development requirements are considered dedicated to buried object detection. Design characteristics of such a system are categorized and listed. Hardware and software structures, ergonomics, user interface, environmental and EMC/EMI tests to be applied and performance test issues are studied. Main properties of the developed system (SEZER) are presented, which contains Metal Detector (MD) and Ground Penetrating Radar (GPR). The realized system has ergonomic structure and can detect both metallic and non-metallic buried objects. Moreover classification of target is possible if it was defined to the signal processing software in learning phase.

Hybrid ground penetrating radar (GPR) and electromagnetic induction (EMI) sensors have advanced landmine detection far beyond the capabilities of a single sensing modality. Both probability of detection (PD) and false alarm rate (FAR) are impacted by the algorithms utilized by each sensing mode and the manner in which the information is fused. Algorithm development and fusion will be discussed, with an aim at achieving a threshold probability of detection (PD) of 0.98 with a low false alarm rate (FAR) of less than 1 false alarm per 2 square meters. Stochastic evaluation of prescreeners and classifiers is presented with subdivisions determined based on mine type, metal content, and depth. Training and testing of an optimal prescreener on lanes that contain mostly low metal anti-personnel mines is presented. Several fusion operators for pre-screeners and classifiers, including confidence map multiplication, will be investigated and discussed for integration into the algorithm architecture.

Ground penetrating radar (GPR) and electromagnetic induction (EMI) sensors provide complementary capabilities in detecting buried targets such as landmines, suggesting that the fusion of GPR and EMI modalities may provide improved detection performance over that obtained using only a single modality. This paper considers both pre-screening and the discrimination of landmines from non-landmine objects using real landmine data collected from a U.S. government test site as part of the Autonomous Mine Detection System (AMDS) landmine program. GPR and EMI pre-screeners are first reviewed and then a fusion pre-screener is presented that combines the GPR and EMI prescreeners using a distance-based likelihood ratio test (DLRT) classifier to produce a fused confidence for each pre-screener alarm. The fused pre-screener is demonstrated to provide substantially improved performance over the individual GPR and EMI pre-screeners. The discrimination of landmines from non-landmine objects using feature-based classifiers is also considered. The GPR feature utilized is a pre-processed, spatially filtered normalized energy metric. Features used for the EMI sensor include model-based features generated from the AETC model and a dipole model as well as features from a matched subspace detector. The EMI and GPR features are then fused using a random forest classifier. The fused classifier performance is superior to the performance of classifiers using GPR or EMI features alone, again indicating that performance improvements may be obtained through the fusion of GPR and EMI sensors. The performance improvements obtained both for pre-screening and for discrimination have been verified by blind test results scored by an independent U.S. government contractor.

The U. S. Army Night Vision and Electronic Sensors Directorate (NVESD) recently tested an explosive-hazards detection vehicle that combines a pulsed FLGPR with a visible-spectrum color camera. Additionally, NVESD tested a human-in-the-loop multi-camera system with the same goal in mind. It contains wide field-of-view color and infrared cameras as well as zoomable narrow field-of-view versions of those modalities. Even though they are separate vehicles, having information from both systems offers great potential for information fusion. Based on previous work at the University of Missouri, we are not only able to register the UTM-based positions of the FLGPR to the color image sequences on the first system, but we can register these locations to corresponding image frames of all sensors on the human-in-the-loop platform. This paper presents our approach to first generate libraries of multi-sensor information across these platforms. Subsequently, research is performed in feature extraction and recognition algorithms based on the multi-sensor signatures. Our goal is to tailor specific algorithms to recognize and eliminate different categories of clutter and to be able to identify particular explosive hazards. We demonstrate our library creation, feature extraction and object recognition results on a large data collection at a US Army test site.

Forward-looking ground-penetrating radar (FLGPR) has received a significant amount of attention for use in explosivehazards detection. A drawback to FLGPR is that it results in an excessive number of false detections. This paper presents our analysis of the explosive-hazards detection system tested by the U.S. Army Night Vision and Electronic Sensors Directorate (NVESD). The NVESD system combines an FLGPR with a visible-spectrum color camera. We present a target detection algorithm that uses a locally-adaptive detection scheme with spectrum-based features. The remaining FLGPR detections are then projected into the camera imagery and image-based features are collected. A one-class classifier is then used to reduce the number of false detections. We show that our proposed FLGPR target detection algorithm, coupled with our camera-based false alarm (FA) reduction method, is effective at reducing the number of FAs in test data collected at a US Army test facility.

Metal detectors and ground penetrating radar have become the standard sensors for buried landmine and UXO detection. Joint systems have existed since the late 90s. Recent system development has again led to the placement of MD and GPR sensors on ground vehicles for detection of in-road landmine and UXO objects. In this work, two prominent systems - one GPR and one metal detector - are operated on a test site populated with landmine and deep buried UXO. The strength of the GPR is the ability to detect plastic cased landmines while the strength of the metal detector is to detect deep buried UXO. The sensors' capabilities overlap in regards to metal cased landmines. A simple fusion approach is used to show how these two sensors can be used together to create a platform that carries the strengths of both sensors. The final alarm list averages the confidence values produced by each sensor. ROC curves are used to quantify the performance. Curves are presented for each sensor standing alone and for their fusion performance.

Using laser Doppler vibrometers (LDVs) to find buried land mines has been shown to have a high probability of detection coupled with a low probability of false alarms. Equally good results have been achieved using a 16-beam LDV. Time division multiplexing (TDM) of this multiple-beam LDV has also been investigated as a means of increasing the scanning speed and potentially allowing the sensor to move down the road at speeds faster than that allowed using stop-and-stare LDVs. A moving platform induces Doppler shifts in the LDV beams that are not perpendicular to the motion vector. This shift can be much greater than the modulation bandwidth of a stationary LDV signal; therefore, the demodulation must allow for the shift either by increasing the processing bandwidth, which increases the system noise or by tracking the Doppler offset and adjusting a band pass filter's center frequency. A method has been developed to track the carrier frequency to compensate for the Doppler offset for each of the 16 channels caused by the moving platform and then adjusting the center frequency of a digital band pass filter. This paper will present the basic filter structure and compare the noise statistics from two different carrier tracking methods that were investigated.

A novel outdoor synthetic aperture acoustic (SAA) system consists of a microphone and loudspeaker traveling along a 6.3-meter rail system. This is an extension from a prior indoor laboratory measurement system in which selected targets were insonified while suspended in air. Here, the loudspeaker and microphone are aimed perpendicular to their direction of travel along the rail. The area next to the rail is insonified and the microphone records the reflected acoustic signal, while the travel of the transceiver along the rail creates a synthetic aperture allowing imaging of the scene. Ground surfaces consisted of weathered asphalt and short grass. Several surface-laid objects were arranged on the ground for SAA imaging. These included rocks, concrete masonry blocks, grout covered foam blocks; foliage obscured objects and several spherical canonical targets such as a bowling ball, and plastic and metal spheres. The measured data are processed and ground targets are further analyzed for characteristics and features amenable for discrimination. This paper includes a description of the measurement system, target descriptions, synthetic aperture processing approach and preliminary findings with respect to ground surface and target characteristics.

When available, GPS is the quick and easy solution to localizing a robot. However, because it is often not available (e.g. indoors) or not reliable enough, other techniques, using laser range finders or cameras have been developed that offer better performance. For 2D localization,lLaser range finders are far more precise and easier to work with than cameras. We report here on the performance of several implementations of the main class of localization algorithms that use a laser, Simultaneous Localization And Mapping (SLAM) on the RAWSEEDS benchmark. SRI International's SLAM system has an RMS error in XY of 0.32m (0.22%). This is the best reported performance on this benchmark.

The Geotechnical and Structures Laboratory at the US Army Corps of Engineers, Engineer Research and Development Center (ERDC) has developed a near-surface properties laboratory to provide complete characterization of soil. Data from this laboratory is being incorporated into a comprehensive database, to enhance military force projection and protection by providing physical properties for modelers and designers of imaging and detection systems. The database will allow cross-referencing of mineralogical, electromagnetic, thermal, and optical properties to predict surface and subsurface conditions. We present an example data set from recent collection efforts including FTIR in the Near-IR, MWIR, and LWIR bands, magnetic susceptibility (500 Hz to 8 GHz), and soil conductivity and complex permittivity (10 μHz to 8 GHz) measurements. X-ray data is presented along with a discussion of site geology, sample collection and preparation methods, and mineralogy. This type of data-collection effort provides useful constraint information of soil properties for use in modeling and target detection. By establishing real ranges for critical soil properties, we are able to improve algorithms to define anomalies that can indicate the presence of land mines, unexploded ordnance (UXOs), improvised explosive devices (IEDs), tunnels, and other visually obscured threats.

The Common IED Exploitation Target Set (CIEDETS) ontology provides a comprehensive semantic data model for capturing knowledge about sensors, platforms, missions, environments, and other aspects of systems under test. The ontology also includes representative IEDs; modeled as explosives, camouflage, concealment objects, and other background objects, which comprise an overall threat scene. The ontology is represented using the Web Ontology Language and the SPARQL Protocol and RDF Query Language, which ensures portability of the acquired knowledge base across applications. The resulting knowledge base is a component of the CIEDETS application, which is intended to support the end user sensor test and evaluation community. CIEDETS associates a system under test to a subset of cataloged threats based on the probability that the system will detect the threat. The associations between systems under test, threats, and the detection probabilities are established based on a hybrid reasoning strategy, which applies a combination of heuristics and simplified modeling techniques. Besides supporting the CIEDETS application, which is focused on efficient and consistent system testing, the ontology can be leveraged in a myriad of other applications, including serving as a knowledge source for mission planning tools.

Cadmium Zinc Telluride (CZT) continues to progress in quality and cost as a material for the detection of hard X-ray and gamma-ray photons with excellent spatial and energy resolutions. We are developing large-volume (0.5×3.9×3.9 cm³) cross-strip CZT detectors with the objective to combine the excellent performance achieved so far only with pixelated CZT detectors with a reduced number of readout channels. In this contribution, we discuss the spectroscopic performance of large volume CZT detectors from the company Orbotech when contacted as pixelated detectors. Subsequently, we present results obtained when the same substrates where contacted with cross-strip contacts. Finally, we use the results from a simulation study to discuss the optimization of the design of the strip contacts and the readout electronics.

The thermal imaging cameras can see the heat signature of people, boats, and vehicles in total darkness as well as through smoke, haze, and light fog, but not through the forest canopy. This study develops a novel algorithm to help detecting obscure targets underneath forest canopy and mitigate the vegetation problem for those bare ground point extraction filters as well. By examining our automatically processed results with actual LiDAR data, the forest canopy is successfully removed where all obscure vehicles or buildings underneath canopy can now be easily seen. Besides, the occluded rate of forest canopy and the detailed underneath x-y point distribution can be easily obtained accordingly. This will be very useful for predicting the performance of occluded target detection with respect to various object locations.

This paper develops algorithms for the detection of interesting and abnormal objects in color and infrared imagery taken from cameras mounted on a moving vehicle, observing a fixed scene. The primary purpose of detection is to cue a human-in-the-loop detection system. Algorithms for direct detection and change detection are investigated, as well as fusion of the two. Both methods use temporal information to reduce the number of false alarms. The direct detection algorithm uses image self-similarity computed between local neighborhoods to determine interesting, or unique, parts of an image. Neighborhood similarity is computed using Euclidean distance in CIELAB color space for the color imagery, and Euclidean distance between grey levels in the infrared imagery. The change detection algorithm uses the affine scale-invariant feature transform (ASIFT) to transform multiple background frames into the current image space. Each transformed image is then compared to the current image, and the multiple outputs are fused to produce a single difference image. Changes in lighting and contrast between the background run and the current run are adjusted for in both color and infrared imagery. Frame-to-frame motion is modeled using a perspective transformation, the parameters of which are computed using scale-invariant feature transform (SIFT) keypoint correspondences. This information is used to perform temporal accumulation of single frame detections for both the direct detection and change detection algorithms. Performance of the proposed algorithms is evaluated on multiple lanes from a data collection at a US Army test site.

Spectral, shape and texture features of the detected targets are used to model the likelihood of detections to be potential mines in a minefield. However, a large number of these potential mines can be false alarms due to the similarity of the mine signatures with natural and other manmade clutter objects which significantly affects the overall detection performance. In addition to the spectral features, spatial distribution of the detected targets can be used to improve the minefield detection performance. In this paper, spectral features and spatial distributions are used simultaneously for minefield detection. We use nearest neighbor distances of the detected targets to capture the spatial characteristics of the minefields. We investigate the spatial distributions and evaluate minefield performance for both patterned and scatterable minefields in a cluttered environment where the number of detected mines is many times less than the number of false alarms. For patterned minefields, performance for minefields with different number of rows at different mine false alarm rates is evaluated. For scatterable minefields, we evaluate the performance of minefields where potential mines are randomly and regularly distributed. In all cases, the false alarms are assumed to be spatially randomly distributed. The performance of the proposed detection algorithm is compared to the baseline algorithm using extensive simulated minefield data.

On-board real-time processing is highly desirable in airborne detection applications. As the data processing involved here is computationally expensive, typically high power multi-rack system is required to achieve real-time detection. Use of such hardware on-board is often not feasible in airborne applications due to space and power constraints. Recently, there has been a lot of interest in the use of Graphics Processing Units (GPUs) for real-time image processing because of their highly parallel architecture, low cost, and compact size. With the introduction of high level languages like C/CUDA (Nvidia), CTM (ATI), OpenCL, etc., GPUs are enjoying a manifold increase in their adoption for general purpose computation. In this paper we present GPU bound implementations of image registration and multiband RX anomaly detector. We identify the sub-problems, namely band-to-band registration, phase correlation, feature detection, feature tracking and image transformation, that can be efficiently parallelized on the SIMD architecture of the GPU. The results from experiments using these implementation are compared against existing implementation written in Matlab and C++.

Multiple instance learning (MIL) is a technique used for identifying a target pattern within sets of data. In MIL, a learner is presented with sets of samples; whereas in standard techniques, a learner is presented with individual samples. The MI scenario is encountered given the nature of landmine detection in GPR data, and therefore landmine detection results should benefit from the use of multiple instance techniques. Previously, a random set framework for multiple instance learning (RSF-MIL) was proposed which utilizes random sets and fuzzy measures to model the MIL problem. An improved version C-RSF-MIL was recently developed showing a increase in learning and classification performance. This new approach is used to learn and characterize features of landmines within GPR imagery for the purposes of classification. Experimental results show the benefits of using C-RSF-MIL for landmine detection in GPR imagery.

Previous work has introduced a framework for information-based sensor management that is capable of tasking multiple sensors searching for targets among a set of discrete objects or in a cell grid. However, in many real-world scenarios-- such as detecting landmines along a lane or road--an unknown number of targets are present in a continuous spatial region of interest. Consequently, this paper introduces a grid-free sensor management approach that allows multiple sensors to be managed in a sequential search for targets in a grid-free spatial region. Simple yet expressive Gaussian target models are introduced to model the spatial target responses that are observed by the sensors. The sensor manager is then formulated using a Bayesian approach, and sensors are directed to make new observations that maximize the expected information gain between the posterior density on the target parameters after a new observation and the current posterior target parameter density. The grid-free sensor manager is applied to a set of real landmine detection data collected with ground penetrating radar (GPR) and electromagnetic induction (EMI) sensors at a U.S. government test site. Results are presented that compare the performance of the sensor manager with the performance of an unmanaged joint pre-screener that fuses individual GPR and EMI pre-screeners. The sensor manager is demonstrated to provide improved detection performance while requiring substantially fewer sensor observations than are made with the unmanaged joint pre-screening approach.

A factor that could affect the performance of ground penetrating radar for landmine detection is self-signature. The radar self-signature is created by the internal coupling of the radar itself and it appears constant in different scans. Although not varying much, the radar self-signature can create hyperbolic shape or anomaly pattern after ground alignment and thereby increasing the amount of false detections. This paper examines the effect of radar self-signature on the performance of the subspace spectral feature landmine detection algorithm. Experimental results in the presence of strong radar self-signatures will be given and performance comparison with the pre-screener that is based on anomaly detection will be made.

In this paper we describe a method for generating cues of possible abnormal objects present in the field of view of an infrared (IR) camera installed on a moving vehicle. The proposed method has two steps. In the first step, for each frame, we generate a set of possible points of interest using a corner detection algorithm. In the second step, the points related to the background are discarded from the point set using an one class classifier (OCC) trained on features extracted from a local neighborhood of each point. The advantage of using an OCC is that we do not need examples from the "abnormal object" class to train the classifier. Instead, OCC is trained using corner points from images known to be abnormal object free, i.e., that contain only background scenes. To further reduce the number of false alarms we use a temporal fusion procedure: a region has to be detected as "interesting" in m out of n, m<n, consecutive frames in order to be reported as abnormal. To choose the best classifier for our task, we compare the performance of three OCCs: nearest neighbor (OCNN), SVM (OC-SVM) and Gaussian mixture (OC-GM). The comparison is performed using a set of about 900 background point neighborhoods for training and 400 for testing. The best performing OCC is then used to detect abnormal objects in a set of IR video sequences obtained on a 1 mile long country road.

In the underground inspection problem, signature of a big target at a certain depth may give equivalent information to the signature of a smaller target at shallower depth, unless depth information is not used. This results in a difficulty in the identification process. Therefore, depth information is coming into prominence in the classification step to increase the identification performance. In this study, we propose a burial depth estimation method on GPR data. In our work, discrete wavelet transform is used in the preprocessing step. After this stage, statistical hypothesis tests are utilized to detect the statistical discrepancies in the returning signals at different depth levels.

The legacy AN/PSS-14 (Army-Navy Portable Special Search-14) Handheld Mine Detecting Set (also called HSTAMIDS for Handheld Standoff Mine Detection System) has proven itself over the last 7 years as the state-of-the-art in land mine detection, both for the US Army and for Humanitarian Demining groups. Its dual GPR (Ground Penetrating Radar) and MD (Metal Detection) sensor has provided receiver operating characteristic curves (probability of detection or Pd versus false alarm rate or FAR) that routinely set the mark for such devices. Since its inception and type-classification in 2003 as the US (United States) Army standard, the desire for use of the AN/PSS-14 against alternate threats - such as bulk explosives - has recently become paramount. To this end, L-3 CyTerra has developed and tested bulk explosive detection and discrimination algorithms using only the Stepped Frequency Continuous Wave (SFCW) Ground Penetrating Radar (GPR) portion of the system, versus the fused version that is used to optimally detect land mines. Performance of the new bulk explosive algorithm against representative zero-metal bulk explosive target and clutter emplacements is depicted, with the utility to the operator also described.

This paper proposes an effective anomaly detection algorithm for a forward-looking ground-penetrating radar (FLGPR). One challenge for threat detection using FLGPR is its high dynamic range in response to different kinds of targets and clutter objects. The application of a fixed threshold for detection often yields a large number of false alarms. We propose a locally-adaptive detection method that adjusts the detection criteria automatically and dynamically across different spatial regions, which improves the detection of weak scattering targets. The paper also examines a spectrum-based classifier. This classifier rejects false alarms (FAs) by classifying each alarm location based on its spatial frequency-spectrum. Experimental results for the improved detection techniques are demonstrated by field data measurements from a US Army test site.

Syntactic pattern recognition is being used to detect and classify non-metallic landmines in terms of their range impedance discontinuity profile. This profile, extracted from the ground penetrating radar's return signal, constitutes a high-range-resolution and unique description of the inner structure of a landmine. In this paper, we discuss two preprocessing steps necessary to extract such a profile, namely, inverse filtering (deconvolving) and binarization. We validate the use of an inverse filter to effectively decompose the observed composite signal resulting from the different layers of dielectric materials of a landmine. It is demonstrated that the transmitted radar waveform undergoing multiple reflections with different materials does not change appreciably, and mainly depends on the transmit and receive processing chains of the particular radar being used. Then, a new inversion approach for the inverse filter is presented based on the cumulative contribution of the different frequency components to the original Fourier spectrum. We discuss the tradeoffs and challenges involved in such a filter design. The purpose of the binarization scheme is to localize the impedance discontinuities in range, by assigning a '1' to the peaks of the inverse filtered output, and '0' to all other values. The paper is concluded with simulation results showing the effectiveness of the proposed preprocessing technique.

Recently, there has been considerable interest in the development of robust, cost-effective and high performance non-metallic landmine detection systems using ground penetrating radar (GPR). Many of the available solutions try to discriminate landmines from clutter by extracting some form of statistical or geometrical information from the raw GPR data, and oftentimes, it is difficult to assess the performance of such systems without performing extensive field experiments. In our approach, a landmine is characterized by a binary-valued string corresponding to its impedance discontinuity profile in the depth direction. This profile can be detected very quickly utilizing syntactic pattern recognition. Such an approach is expected to be very robust in terms of probability of detection (P_d) and low false alarm rates (FAR), since it exploits the inner structure of a landmine. In this paper, we develop a method to calculate an upper bound on the FAR, which is the probability of false alarm per unit area. First, we parameterize the number of possible mine patterns in terms of the number of impedance discontinuities, dither and noise. Then, a combinatorial enumeration technique is used to quantify the number of admissible strings. The upper bound on FAR is given as the ratio of an upper bound on the number of possible mine pattern strings to the number of admissible strings per unit area. The numerical results show that the upper bound is smaller than the FAR reported in the literature for a wide range of parameter choices.

Time domain ground penetrating radar (GPR) has been shown to be a powerful sensing phenomenology for detecting buried objects such as landmines. Landmine detection with GPR data typically utilizes a feature-based pattern classification algorithm to discriminate buried landmines from other sub-surface objects. In high-fidelity GPR, the time-frequency characteristics of a landmine response should be indicative of the physical construction and material composition of the landmine and could therefore be useful for discrimination from other non-threatening sub-surface objects. In this research we propose modeling landmine time-domain responses with a nonparametric Bayesian time-series model and we perform clustering of these time-series models with a hierarchical nonparametric Bayesian model. Each time-series is modeled as a hidden Markov model (HMM) with autoregressive (AR) state densities. The proposed nonparametric Bayesian prior allows for automated learning of the number of states in the HMM as well as the AR order within each state density. This creates a flexible time-series model with complexity determined by the data. Furthermore, a hierarchical non-parametric Bayesian prior is used to group landmine responses with similar HMM model parameters, thus learning the number of distinct landmine response models within a data set. Model inference is accomplished using a fast variational mean field approximation that can be implemented for on-line learning.

Context-dependent classification techniques applied to landmine detection with ground-penetrating radar (GPR) have demonstrated substantial performance improvements over conventional classification algorithms. Context-dependent algorithms compute a decision statistic by integrating over uncertainty in the unknown, but probabilistically inferable, context of the observation. When applied to GPR, contexts may be defined by differences in electromagnetic properties of the subsurface environment, which are due to discrepancies in soil composition, moisture levels, and surface texture. Context-dependent Feature Selection (CDFS) is a technique developed for selecting a unique subset of features for classifying landmines from clutter in different environmental contexts. In past work, context definitions were assumed to be soil moisture conditions which were known during training. However, knowledge of environmental conditions could be difficult to obtain in the field. In this paper, we utilize an unsupervised learning algorithm for defining contexts which are unknown a priori. Our method performs unsupervised context identification based on similarities in physics-based and statistical features that characterize the subsurface environment of the raw GPR data. Results indicate that utilizing this contextual information improves classification performance, and provides performance improvements over non-context-dependent approaches. Implications for on-line context identification will be suggested as a possible avenue for future work.

We propose a landmine detection algorithm that uses ensemble discrete hidden Markov models with context dependent training schemes. We hypothesize that the data are generated by K models. These different models reflect the fact that mines and clutter objects have different characteristics depending on the mine type, soil and weather conditions, and burial depth. Model identification is based on clustering in the log-likelihood space. First, one HMM is fit to each of the N individual sequence. For each fitted model, we evaluate the log-likelihood of each sequence. This will result in an N x N log-likelihood distance matrix that will be partitioned into K groups. In the second step, we learn the parameters of one discrete HMM per group. We propose using and optimizing various training approaches for the different K groups depending on their size and homogeneity. In particular, we will investigate the maximum likelihood, and the MCE-based discriminative training approaches. Results on large and diverse Ground Penetrating Radar data collections show that the proposed method can identify meaningful and coherent HMM models that describe different properties of the data. Each HMM models a group of alarm signatures that share common attributes such as clutter, mine type, and burial depth. Our initial experiments have also indicated that the proposed mixture model outperform the baseline HMM that uses one model for the mine and one model for the background.

The Edge Histogram Detector (EHD) is a landmine detection algorithm that has been developed for ground penetrating radar (GPR) sensor data. It has been tested extensively and has demonstrated excellent performance. The EHD consists of two main components. The first one maps the raw data to a lower dimension using edge histogram based feature descriptors. The second component uses a possibilistic K-Nearest Neighbors (pK-NN) classifier to assign a confidence value. In this paper we show that performance of the baseline EHD could be improved by replacing the pK-NN classifier with model based classifiers. In particular, we investigate two such classifiers: Support Vector Regression (SVR), and Relevance Vector Machines (RVM). We investigate the adaptation of these classifiers to the landmine detection problem with GPR, and we compare their performance to the baseline EHD with a pK-NN classifier. As in the baseline EHD, we treat the problem as a two class classification problem: mine vs. clutter. Model parameters for the SVR and the RVM classifiers are estimated from training data using logarithmic grid search. For testing, soft labels are assigned to the test alarms. A confidence of zero indicates the maximum probability of being a false alarm. Similarly, a confidence of one represents the maximum probability of being a mine. Results on large and diverse GPR data collections show that the proposed modification to the classifier component can improve the overall performance of the EHD significantly.