Brandon Smock

Wideband EMI pre-screening for landmine detection

This paper considers the use of data from a wideband electromagnetic induction (EMI) sensor in a prescreener for a landmine detection system employing both ground-penetrating radar (GPR) and EMI sensors. The paper looks at a unique EMI prescreening strategy based on the use of prototypes derived from a training set of landmines. We show that this prescreener is robust to a wide range of induced energy levels in sensed objects. We also compare properties of the receiver operating characteristics (ROC) curve of this prescreener on a varied collection of targets to the properties of a GPR prescreener, identifying performance difference with respect to target object classes.

Comparison of algorithms for finding the air-ground interface in ground penetrating radar signals

In using GPR images for landmine detection it is often useful to identify the air-ground interface in the GPR signal for alignment purposes. A number of algorithms have been proposed to solve the air-ground interface detection problem, including some which use only A-scan data, and others which track the ground in B-scans or C-scans. Here we develop a framework for comparing these algorithms relative to one another and we examine the results. The evaluations are performed on data that have been categorized in terms of features that make the air-ground interface difficult to find or track. The data also have associated human selected ground locations, from multiple evaluators, that can be used for determining correctness. A distribution is placed over each of the human selected ground locations, with the sum of these distributions at the algorithm selected location used as a measure of its correctness. Algorithms are also evaluated in terms of how they affect the false alarm and true positive rates of mine detection algorithms that use ground aligned data.

Reciprocal pointer chains for identifying layer boundaries in ground-penetrating radar data

Identifying the ground surface in ground-penetrating radar (GPR) data is useful and can be done efficiently and accurately using the Viterbi algorithm. This involves representing the radar image as a trellis graph and solving for the optimal path. To identify multiple layer boundaries in a radar image in this manner, it is necessary to find multiple disjoint paths through the trellis. Two main types of algorithms currently exist that find the k best disjoint paths whose aggregate sum is minimized. However, this criterion has drawbacks. Instead, we propose a novel criterion for choosing multiple disjoint paths in a trellis that we call the reciprocal pointer chain. This criterion has both a nice intuitive and theoretical justification, and leads to an algorithm with better qualitative results and significantly lower computational complexity than any of the methods previously proposed.

Efficient multiple layer boundary detection in ground-penetrating radar data using an extended Viterbi algorithm

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

DynaMax+ ground-tracking algorithm

In this paper, we propose a new method for performing ground-tracking using ground-penetrating radar (GPR). Ground-tracking involves identifying the air-ground interface, which is usually the dominant feature in a radar image but frequently is obscured or mimicked by other nearby elements. It is an important problem in landmine detection using vehicle-mounted systems because antenna motion, caused by bumpy ground, can introduce distortions in downtrack radar images, which ground-tracking makes it possible to correct. Because landmine detection is performed in real-time, any algorithm for ground-tracking must be able to run quickly, prior to other, more computationally expensive algorithms for detection. In this investigation, we first describe an efficient algorithm, based on dynamic programming, that can be used in real-time for tracking the ground. We then demonstrate its accuracy through a quantitative comparison with other proposed ground-tracking methods, and a qualitative comparison showing that its ground-tracking is consistent with human observations in challenging terrain.

Optimal fusion of alarm sets from multiple detectors using dynamic programming

In a standard target detection approach, data is collected, points of interest called alarms are identified, and detection algorithms determine the confidence that a target is present at each point. Receiver operating characteristic (ROC) curves can be used to evaluate the performance of each detector and choose operating thresholds. The use of multiple sensors can improve the probability of detection of a diverse set of targets. It is difficult to properly assess the performance of a system of detectors and choose the best joint set of operating thresholds if confidence values from different detectors do not compare meaningfully. Fusion methods can be used to improve the joint performance of a set of detectors. However, in the case where different detectors do not operate on the same points of interest, typical fusion methods cannot be used to improve the binary decisions on individual alarms. In this paper, we propose a new fusion method that maps the confidence outputs from different detectors to a shared range where they compare meaningfully, and optimizes the joint performance of multiple detectors even when their alarm sets are disjoint. Our method uses dynamic programming to monotonically map the confidence output from each detector onto a shared range in such a way that we maximize the area-under-the-curve (AUC) of the ROC curve corresponding to the joint set of alarms. This joint ROC curve can be used to determine the operational thresholds for each individual detector to maximize their joint performance.

Improved landmine detection through context-dependent score calibration

Algorithms developed for the detection of landmines are tasked with discriminating a wide variety of targets in a diverse array of environmental conditions. However, the potential performance of a detection algorithm may be underestimated by evaluating it in batch on a large, diverse dataset. This is because environmental, or in general, contextual, factors may contribute significant variance to the output of a detection algorithm across different contexts. One way to view this is as a problem of miscalibration: within each context, the output scores of a detection algorithm can be seen as miscalibrated relative to the scores produced in the other contexts. As a result of this miscalibration, the observed receiver operating characteristic (ROC) curve for a detector can have a sub-optimal area-under-the-curve (AUC). One solution, then, is to re-calibrate the detector within each context. In this work, we identify multiple sets of contexts in which different landmine detection algorithms exhibit significant output variance and, consequently, miscalibration. We then apply a monotonic calibration strategy that maximizes AUC and demonstrate the gain in observed performance that results when a landmine detection algorithm is properly calibrated within each context.

A layer tracking approach to buried surface detection

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

A hybrid approach to the fusion of partially coinciding alarms from asynchronous or specialized detectors

The accurate detection of a diverse set of targets often requires the use of multiple sensor modalities and algorithms. Fusion approaches can be used to combine information from multiple sensors or detectors. But typical fusion approaches are not suitable when detectors do not operate on all of the same locations of interest, or when detectors are specialized to detect disjoint sets of target types. Run Packing is an algorithm we developed previously to optimally combine detectors when their output never coincides, which can be expected when the detectors are specialized to detect different target types. But when asynchronous detectors sometimes coincide, or specialized detectors sometimes detect the same target, Run Packing ignores this coincidence information and thus may be suboptimal in certain cases. In this paper, we show how multi-detector fusion involving partially coinciding alarms can be re-framed as an equivalent fusion problem that is optimally addressed by Run Packing. This amounts to a hierarchical or hybrid approach involving fusion methods to join coinciding alarms at the same location into a single unified alarm and then using Run Packing to optimally fuse the resulting set of non coinciding alarms. We report preliminary results of applying the method in a few typical landmine detection scenarios to demonstrate its potential utility.

Context-dependent detection via Alarm-Set Fusion and segmentation

A COntext-DEpendent (CODE) detection method is described. Detectors are run in separate image segments, and the detector outputs are aggregated to produce a system output. The process of aggregating outputs is called Alarm-Set Fusion (ASF). A Run Packing (RP) ASF technique is described for training an ASF algorithm. The RP ASF algorithm uses dynamic programming to optimize area under the Receiver Operating Characteristic (ROC) curve. Experimental results are presented using real hyperspectral images segmented using hyperspectral and LiDAR features. Context-dependent and -independent Hybrid Sub-pixel Detector (HSD) and Adaptive Cosine Estimator (ACE) detectors were compared. The CODE method yielded better performance. A comparison to an ACE-RX detector was conducted. Detection performance was similar but the CODE computational requirements are lower. Experiments also show that the ASF algorithm can learn non-obvious threshold combinations.