Wen-Hsiung Lee

A Large-Scale Systematic Evaluation of Algorithms Using Ground-Penetrating Radar for Landmine Detection and Discrimination

A variety of algorithms for the detection of landmines and discrimination between landmines and clutter objects have been presented. We discuss four quite different approaches in using data collected by a vehicle-mounted ground-penetrating radar sensor to detect landmines and distinguish them from clutter objects. One uses edge features in a hidden Markov model; the second uses geometric features in a feed-forward order-weighted average network; the third employs spectral features as its basis; and the fourth clusters edge histograms. We present the results of a large-scale cross-validation evaluation that uses a diverse set of data collected over 41 807.57 m2 of ground, including 1593 mine encounters. Finally, we discuss the results of that ranking and what one can conclude concerning the performance of these four algorithms in various settings.

Detecting landmines with ground-penetrating radar using feature-based rules, order statistics, and adaptive whitening

An approach to detecting landmines using ground-penetrating radar (GPR) based on feature-based rules, order statistics, and adaptive whitening (FROSAW) is described. FROSAW relies on independent adaptation of whitening statistics in different depths and combining feature-based methods with anomaly detection using rules. Constant false alarm rate (CFAR) detectors are used for anomaly detection on the depth-dependent adaptively whitened data. A single CFAR confidence measure is computed via a function of order statistics. Anomalies are detected at locations with high CFAR confidence. Depth-dependent features are computed on regions containing anomalies. Rules based on the features are used to reject alarms that do not exhibit mine-like properties. The utility of combining the CFAR and feature-based methods is evaluated. The algorithms and analysis are applied to data acquired from tens of thousands of square meters from several outdoor test sites with a state-of-the-art array of GPR sensors.

Optimizing the Area Under a Receiver Operating Characteristic Curve With Application to Landmine Detection

A common approach to training neural network classifiers in a supervised learning setting is to minimize the mean-square error (mse) between the network output for each labeled training sample and some desired output. In the context of landmine detection and discrimination, although the performance of an algorithm is correlated with the mse, it is ultimately evaluated by using receiver operating characteristic (ROC) curves. In general, the larger the area under the ROC curve (AUC), the better. We present a new method for maximizing the AUC. Desirable properties of the proposed algorithm are derived and discussed that differentiate it from previously proposed algorithms. A hypothesis test is used to compare the proposed algorithm to an existing algorithm. The false alarm rate achieved by the proposed algorithm is found to be less than that of the existing algorithm with 95% confidence

Feature analysis for the NIITEK ground-penetrating radar using order-weighted averaging operators for landmine detection

An automated methodology for combining Ground Penetrating Radar features from different depths is presented and analyzed. GPR data from the NIITEK system are processed by a depth-dependent, adaptive whitening algorithm. Shape and contrast features, including compactness, solidity, eccentricity, and relative area are computed at the different depths. These features must be combined to make a decision as to the presence of a landmine at a specific location. Since many of the depths contain no useful information and the depths of the mines are unknown, a strategy based on sorting is used. In a previous work, sorted features were combined via a rule-based system. In the current paper, an automated algorithm that builds a decision rule from sets of Ordered Weighted Average (OWA) operators is described. The OWA operator sorts the feature values, weights them, and performs a weighted sum of the sorted values, resulting in a nonlinear combination of the feature values. The weights of the OWA operators are trained off-line in combination with those of a decision-making network and held fixed during testing. The combination of OWA operators and decision-making network is called a FOWA network. The FOWA network is compared to the rule-based method on real data taken from multiple collections at two outdoor test sites.

The efficacy of human observation for discrimination and feature identification of targets measured by the NIITEK ground-penetrating radar

Recently, blind tests of several automated detection algorithms operating on the NIITEK ground penetrating radar data (GPR) have resulted in quite promising performance results. Anecdotally, human observers have also shown notable skill in detecting landmines and rejecting false alarms in this same data; however, the basis of human performance has not been studied in depth. In this study, human observers are recruited from the undergraduate and graduate student population at Duke University and are trained to visually detect landmines in the NIITEK GPR data. Subjects are then presented with GPR responses associated with blanks, clutter items (including emplaced clutter), and landmines in a blind test scenario. Subjects are asked to make the decision as to whether they are viewing a landmine response or a false alarm, and their performance is scored. A variety of landmines, measured at several test sites, are presented to determine the relative difficulty in detecting each mine type. Subject performance is compared to the performance of two automated algorithms already under development for the NIITEK radar system: LMS and FROSAW. In addition, subjects are given a subset of features for each alarm from which they may indicate the reason behind their decision. These last data may provide a basis for the design of an automated algorithm that takes advantage of the most useful of the observed features.

Continuous Choquet integrals with respect to random sets with applications to landmine detection

A hit-miss transform is defined using Choquet integrals with respect to random sets. In this context, random sets represent random shapes defined on the plane. Random sets are characterised by their capacity functional. Capacity functional is fuzzy measures. Choquet integrals with respect to random sets are interpreted as stochastic morphological operations. Specifically, the integrals represent the average probability that sets either intersect or are contained in the random sets. Formulas are defined for random erosions and dilations of disks and annuli with Gaussian radii. Applications to landmine detection are given.

Region Processing of Ground Penetrating Radar and Electromagnetic Induction for Handheld Landmine Detection

An analysis of the utility of region-based processing of Ground Penetrating Radar (GPR) and Electromagnetic Induction (EMI) is presented. Algorithms for re-sampling GPR data acquired over non-rectangular and non-regular grids are presented. Depth-dependent whitening is used to form GPR images as functions of depth bins. Shape, size, and contrast-based features are used to distinguish mines from non-mines. The processing is compared to point-wise processing of the same data. Comparisons are made to GPR data collected by machine and by humans. Evaluations are performed on calibration data, for which the ground truth is known to the algorithm developers, and blind data, for which the ground truth is not known to the algorithm developers.

Landmine detection using frequency domain features from GPR measurements and their fusion with time domain features

We present in this paper the use of frequency domain features deduced from the energy density spectrum to improve the detection of landmines. The energy density spectrum is obtained from the GPR measurements at an alarm location, and a method to estimate the energy density spectrum is proposed. The energy density spectrum is shown to reveal distinct characteristics between landmine and clutter objects and therefore can be explored for their discrimination. The robustness and consistency of the frequency domain features are demonstrated through two different GPRs. The fusion of frequency domain features and time-domain features is also examined. Experimental results at several test sites confirm the advantages of frequency domain features to better discriminate between mine targets and clutter objects, and the effectiveness of fusion in frequency and time domain features to improve detection performance.

Renyi entropy with respect to Choquet capacities

Entropy terms are analyzed for use in a heterogeneous network involving ordered weighted averaging (OWA) operators and feed-forward neural networks, referred to as FOWA networks. The OWA operators are used to aggregate feature information and were previously used successfully in a landmine application. Weight vectors of OWA operators are special cases of Choquet capacities for which the Shannon entropy has been defined. One can also consider the quadratic Renyi entropy for OWA weight vectors. A term is added to the objective function of the FOWA network involving these entropy terms. An analysis of the derivatives used in gradient descent is conducted.

Fusion of acoustic/seismic and GPR detection algorithms

A variety of sensors have been investigated for the purpose of detecting buried landmines in outdoor environments. Mines with little or no metal are very difficult to detect with traditional mine detection systems. Ground Penetrating Radar (GPR) sensors have shown great promise in detecting low metal mines and can easily detect metal mines. Unfortunately, it can still be difficult to detect low-metal mines with GPR due to very low contrast between the mine and the surrounding medium. Acoustic-seismic systems were proposed by Sabatier et.al. and have also shown great promise in detecting low metal mines. There are now a wealth of references that discuss these systems and algorithms for processing data from these systems. Therefore, they will not be discussed in detail here. In fact, low-metal mines are easier to detect than metal mines with this acoustic-seismic systems. Low metal mines that are difficult for a GPR to detect can be quite easy to detect with acoustic-seismic systems. Sensor fusion with these sensors is of interest since together they can find a broader range of mines with relative ease. The algorithmic challenge is to determine a strategy for combining the multi-sensor information in a way that can increase the probability of detection without increasing the false alarm rate significantly. In this paper, we investigate fusion of information obtained from GPR and acoustic-seismic on real data measured from a mine lane containing three types of buried landmines and also areas containing no landmines. Algorithms are applied to data acquired from each sensor and confidence values are assigned to each location at which a measurement is made by each sensor. The GPR is used as a primary sensor. At each location at which the GPR algorithm declares an alarm, a modified likelihood-based approach is used to increase the GPR derived confidence if the likelihood that a mine is present, defined by the acoustic-based confidence, is larger than the likelihood that no mine is present. If the acoustic-derived confidence is very high, then a declaration is made even if there is no GPR declaration. The experiments were conducted using data acquired from the sensors at different times. The acoustic-seismic system collected data over a subset of the region at which the GPR collected data. Results are given only over those regions for which both sensors collected data.