Recently developed feature extraction methods proposed in the explosive hazard detection community have yielded
many features that potentially provide complementary information for explosive detection. Finding the right
combination of features that is most effective in distinguishing targets from clutter, on the other hand, is extremely
challenging due to a large number of potential features to explore. Furthermore, sensors employed for mine and buried
explosive hazard detection are typically sensitive to environmental conditions such as soil properties and weather as well
as other operating parameters. In this work, we applied Bayesian cross-categorization (CrossCat) to a heterogeneous set
of features derived from electromagnetic induction (EMI) sensor time-series for purposes of buried explosive hazard
detection. The set of features used here includes simple, point-wise measurements such as the overall magnitude of the
EMI response, contextual information such as soil type, and a new feature consisting of spatially aggregated Discrete
Spectra of Relaxation Frequencies (DSRFs). Previous work showed that the DSRF characterizes target properties with
some invariance to orientation and position. We have developed a novel approach to aggregate point-wise DSRF
estimates. The spatial aggregation is based on the Bag-of-Words (BoW) model found in the machine learning and
computer vision literatures and aims to enhance the invariance properties of point-wise DSRF estimates. We considered
various refinements to the BoW model for purpose of buried explosive hazard detection and tested their usefulness as
part of a Bayesian cross-categorization framework on data collected from two different sites. The results show improved
performance over classifiers using only point-wise features.
Automatic Target Recognition (ATR) algorithm performance is highly dependent on the sensing conditions under which the input data is collected. Open-loop fly-bys often produce poor results due to less than ideal measurement conditions. In addition, ATR algorithms must be extremely complicated to handle the diverse range of inputs with a resulting reduction in overall performance and increase in complexity. Our approach, closed-loop ATR (CL-ATR), focuses on improving the quality of information input to the ATR algorithms by optimizing motion, sensor settings and team (vehicle-vehicle-human) collaboration to dramatically improve classification accuracy. By managing the data collection guided by predicted ATR performance gain, we increase the information content of the data and thus dramatically improve ATR performance with existing ATR algorithms. CL-ATR has two major functions; first, an ATR utility function, which represents the performance sensitivity of ATR produced classification labels as a function of parameters that correlate to vehicle/sensor states. This utility function is developed off-line and is often available from the original ATR study as a confusion matrix, or it can be derived through simulation without direct access to the inner working of the ATR algorithm. The utility function is inserted into our CLATR framework to autonomously control the vehicle/sensor. Second, an on-board planner maps the utility function into vehicle position and sensor collection plans. Because we only require the utility function on-board, we can activate any ATR algorithm onto a unmanned aerial vehicle (UAV) platform no matter how complex. This pairing of ATR performance profiles with vehicle/sensor controls creates a unique and powerful active perception behavior.
Fusion of imaging data with auxiliary signal such as EW data for multitarget classification poses daunting theoretical and practical challenges. The problem is exacerbated by issues such as asynchronous data flow, uneven feature quality and object occlusion. In our approach, we assign prior probabilities to image and signal feature elements to handle those practical issues in a unified manner. Current state and class probability distributions estimated from previous instances are fused with new outputs from individual classifiers immediate after the outputs become available to establish updated state and class probability distributions in a Bayesian framework. Results are presented that demonstrate joint segmentation and tracking, target classification using imaging data, and fusion of imaging data with noisy and asynchronous auxiliary EW information under realistic simulation scenarios.
Most of the research on developing automatic target recognition (ATR) algorithms for acoustic-seismic landmine detection platforms has been focused on using geometric features, such as size and shape, of anomaly to distinguish between mines and clutter. This approach has achieved some success especially in detecting larger anti-tank mines. However, for smaller anti-personnel landmines, the difference in geometric features between mines and clutter can be very small, if any. To improve the detection vs. false alarm rates, it is necessary to incorporate other features into the ATR process. It has been observed from the collected acoustic data that areas with buried mines reveal more complicated surface vibration structures, such as the ring-like pattern, at certain frequencies than what a one-dimensional lumped mass-spring-dashpot model can describe. In this paper, we utilize the distributed mine/soil interaction model developed by the University of Mississippi to describe the surface vibration patterns. We develop a modified Hankel transform to extract features from areas under interrogation. Under such transform, concentration of energy is closely related to an object's physical properties. The frequency at which the energy concentration occurs corresponds to the object's natural frequency, while the corresponding Bessel basis captures its mode shape. After de-noising the transformed data, we use the frequencies, Bessel bases, and magnitudes of the energy concentrations, together with other geometric features, to form the feature vectors. We tested these features on a dataset consisting of anti-tank and anti-personnel mines as well as blank areas and metallic and non-metallic clutter. Classifiers designed based on the combined geometric and model-based features perform significantly better than those based on the geometric features alone.
KEYWORDS: Mining, General packet radio service, Image segmentation, Feature extraction, Data modeling, Land mines, Sensors, Principal component analysis, Soil science, Mahalanobis distance
Designing robust landmine detection algorithms for ground penetrating radar (GPR) remains a challenging task due to variations of environmental conditions and diverse clutter objects in the soil, among others. The problem is aggravated for handheld systems by introducing operator motion and by the position uncertainty. Even though aggregating consecutive GPR samples to form multi-sample features seems to be an intuitively sensible approach to improve Pd/Pf, determining multi-sample features that are robust to the operator motion and position uncertainty is a formidable task. In this paper, we propose an ATR method to identify mines based on handheld GPR data collected for regional processing, where systematic operator motion is required and perhaps some position information is collected along with the data. The regional processing is intended to be conducted after other initial detection methods have identified an area for further interrogation. In this study, we will use GPR data that were collected by a robotic arm. In order for the developed ATR method to be applicable to data collected by human operators, which have greater position uncertainty, we focus on features that can still be used either directly or with minor modification when accurate sensor positions are not available. We tested two classes of classifiers, Support Vector Machines (SVM) and Gaussian Mixtures (GM). For both classifiers, less complex forms of the classifiers outperform those with more complicated structures. The reason is that the training set is relatively small compared to the diversity of the mines and the clutter objects in the training set.
KEYWORDS: Mining, Land mines, Acoustics, Velocity measurements, Laser Doppler velocimetry, Systems modeling, Signal detection, Doppler effect, System identification, Target detection
The acoustic-seismic mine detection concept is based on the principle that an area with a buried object shows different dynamic response to acoustic excitation from that of soil. In this paper, we attempt to model and identify the dynamic behavior of a landmine under acoustic excitation for the purpose of automatic mine detection. A linear distributed model is used to model the two-dimensional vibration patterns of landmines. According to modal analysis of the model, it is shown that locations of the poles remain invariant throughout the area where a mine is buried underneath, and can be used as important features for distinguishing mines from clutter. A time-domain method that utilizes the acoustic pressure measured by a microphone as the input and the ground velocity measured by a laser Doppler vibrometer (LDV) as the output was employed to identify the model parameters including the poles. Based on the invariant property of the poles, the identified poles from neighboring measurements were combined to separate any area that show features in the spatial-spectral domain that correspond to presence of a mine.
KEYWORDS: Mining, Land mines, Acoustics, Solids, Laser Doppler velocimetry, Velocity measurements, Systems modeling, Night vision, Sensors, Signal detection
Acoustic-seismic coupling mine detection offers an alternative approach to distinguishing mines from clutter. The approach is based on the principle that an area with a buried object shows a different response to acoustic excitation from that of the surrounding soil. Prior research shows that the response in the low frequency range can be captured using simple physically based models under certain conditions. According to the models, areas with buried mines exhibit natural frequencies that can be determined from mine types and buried depths. In this paper, we argue that not only are the natural frequencies useful for the purpose of mine detection, but the locations of the transmission zeros are important as well. Under certain conditions, the locations of the transmission zeros are also less sensitive to changes in physical properties of mines. We take advantage of this characteristic and offer a method to improve signal-to-clutter ratio for the purpose of automatic mine detection.
KEYWORDS: Land mines, Sensors, Mining, Data modeling, Detection and tracking algorithms, General packet radio service, Principal component analysis, Image segmentation, Stochastic processes, Autoregressive models
A major difficulty in automatic mine detection arises from the fact that the physical properties of background soil can vary significantly from one location to another. This in turns alters the sensor signals of the buried mines. Hence, a robust ATR algorithm for mine detection requires that the algorithm be adaptable to environmental changes. Moreover, mine features used for detection should be invariant to background variation. We have developed an ATR algorithm that uses only background soil data during the training phase and mine features that are less affected by soil changes. Since the algorithm uses only the background data for training, not only is it much easier to tailor the algorithm to a minefield but the algorithm can also be adapted in real-time during operation. This further improves robustness of the process. The algorithm demonstrated good performance when tested on ground penetrating radar data acquired from U.S. Army test lanes.
Recent research sponsored by the Army, Navy and DARPA has significantly advanced the sensor technologies for mine detection. Several innovative sensor systems have been developed and prototypes were built to investigate their performance in practice. Most of the research has been focused on hardware design. However, in order for the systems to be in wide use instead of in limited use by a small group of well-trained experts, an automatic process for mine detection is needed to make the final decision process on mine vs. no mine easier and more straightforward. In this paper, we describe an automatic mine detection process consisting of three stage, (1) signal enhancement, (2) pixel-level mine detection, and (3) object-level mine detection. The final output of the system is a confidence measure that quantifies the presence of a mine. The resulting system was applied to real data collected using radar and acoustic technologies.
During the last two decades IR Goodman, HT Nguyen and others have shown that several basic aspects of expert-systems theory-fuzzy logic, Dempster-Shafer evidence theory, and rule-based inference-can be subsumed within a completely probabilistic framework based on random set theory. In addition, it has been shown that this body of research can be rigorously integrated with multisensor, multitarget filtering and estimation using a special case of random set theory called 'finite-set statistics' (FISST). In particular, FISST allows the basis for standard tracking and ID algorithms-nonlinear filtering theory and estimation theory; to be extended to the case when evidence can be highly 'ambiguous' because of extended operating conditions, e.g. when images are corrupted by effects such as dents, mud etc. This paper extends those results by showing that the technique is relatively insensitive to the uncertainty model used to construct the ambiguous likelihood function.
KEYWORDS: Mining, General packet radio service, Land mines, Algorithm development, Principal component analysis, Detection and tracking algorithms, Ferroelectric LCDs, Distance measurement, Ground penetrating radar, Data mining
We describe several automatic mine detection algorithms in this paper. These methods were tested on real Ground Penetrating Radar (GPR) data and showed dramatic improvement in terms of probability of detection and false alarm rates compared to energy based techniques. The main contributions of this paper are as follows. (1) Only background clutter data, instead of mine data, are needed for the development of the algorithms, which makes collection of data for training and adapting the algorithms to new environment much easier than methods requiring both clutter and mine data for training. (2) The mine detection algorithms are developed in a fairly general form, and thus can be ported to other sensor platforms or future generations of mine detection hardware with little modification. (3) The algorithms require little on-line computation. (4) Adaptation of the algorithms to new environment or mine-fields is done automatically, which reduces human resources and the cost of training.
During the last two decades I.R. Goodman, H.T. Nguyen and others have shown that several basic aspects of expert- systems theory-fuzzy logic, Dempster-Shafer evidence theory, and rule-based inference-can be subsumed within a completely probabilistic framework based on random set theory. In addition, it has been shown that this body of research can be rigorously integrated with multisensor, multitarget filtering and estimation using a special case of random set theory called `Finite-Set Statistics' (FISST). In particular, FISST allows the basis for standard tracking and I.D. algorithms--nonlinear filtering theory and estimation theory--to be extended to the case when evidence can be highly `ambiguous' (imprecise, vague, contingent, etc.). This paper summarizes preliminary results in applying the FISST filtering approach to the problem of identifying ground targets from Synthetic Aperture Radar data that is `ambiguous' because of Extended Operating Conditions, e.g. when images are corrupted by effects such as dents, mud, etc.
In this paper, we describe the automatic damage assessment process we developed to assess the extent of battle damage based on laser rdar (Ladar) images taken before and after a strike. The process is composed of three modules, the image registration module, the damage isolation module, and the damage assessment module. In the image registration module, distortion in the raw range data was first compensated to obtain the actual heights of the objects. Then, Ladar images taken before and after the strike were aligned according to their pixel intensities. In the damage isolation module, changes between the two sets of images were compared to isolate the locations of the actual damage. Factors such as sensor noise, sensor perspective difference, debris, and movement of vehicles, which all contribute to changes in the images, were automatically discounted in this module. The approximate location of the damage, if it existed, was passed to the damage assessment module to determine the extent of the damage using the region growing technique. This process enables fast and accurate evaluation of a strike with as little human supervision as possible.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.