Multiple-Instance Hidden Markov Models With Applications to Landmine Detection

Yüksel, Seniha Esen; Bolton, Jeremy; Gader, Paul D.

doi:10.1109/tgrs.2015.2447576

Cited by 29 publications

(15 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As a result, all alarms within a certain spatial distance were clustered and assigned to the same fold, to avoid training and testing over the same physical area. To properly handle the issues associated with proper cross-validation on this type of data set, researchers at the University of Florida developed software which is used here and has been used in many previous studies [1], [6], [7], [11], [29], [34].…”

Section: Cross-validation and Performance Metricsmentioning

confidence: 99%

“…To compare the detection performance of each trained classifier, receiver operating characteristic (ROC) curves are used. ROC curves are a common metric for comparing machine learning algorithms, and they are likewise popular in the BTD algorithm research literature [1], [6], [7], [11], [29], [34]. ROC curves plot the relationship between the false detection rate (xaxis) and true detection rate (y-axis) of a detection algorithm, as the sensitivity of the algorithm is varied.…”

Section: Cross-validation and Performance Metricsmentioning

confidence: 99%

See 1 more Smart Citation

On Choosing Training and Testing Data for Supervised Algorithms in Ground-Penetrating Radar Data for Buried Threat Detection

Reichman

Collins

Malof

2018

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Abstract-Ground penetrating radar (GPR) is one of the most popular and successful sensing modalities that has been investigated for landmine and subsurface threat detection. Many of the detection algorithms applied to this task are supervised and therefore require labeled examples of target and non-target data for training. Training data most often consists of 2-dimensional images (or patches) of GPR data, from which features are extracted, and provided to the classifier during training and testing. Identifying desirable training and testing locations to extract patches, which we term "keypoints", is well established in the literature. In contrast however, a large variety of strategies have been proposed regarding keypoint utilization (e.g., how many of the identified keypoints should be used at targets, or non-target, locations). Given the variety keypoint utilization strategies that are available, it is very unclear (i) which strategies are best, or (ii) whether the choice of strategy has a large impact on classifier performance. We address these questions by presenting a taxonomy of existing utilization strategies, and then evaluating their effectiveness on a large dataset using many different classifiers and features. We analyze the results and propose a new strategy, called PatchSelect, which outperforms other strategies across all experiments. Index Terms-training, ground penetrating radar, landmine detection I. INTRODUCTIONA popular approach for detecting buried threats, such as landmines and other explosive hazards, is the use of remote sensing technologies. One of the most successful modalities for remote sensing of buried threats is the ground penetrating radar (GPR) [1]- [5]. The typical GPR consists of an array of antennas that are directed toward the ground. An individual GPR antenna operates by emitting a radar signal towards the ground and then measuring the energy that is reflected back. The result of this sensing process is a time-series of energy measurements for the given antenna, referred to as an A-scan [6].In the context of buried threat detection (BTD), GPR sensors collect A-scans at regular spatial intervals as they move across the surface of the ground (e.g., on the front of a vehicle as it drives forward). The resulting A-scans, each collected at a different spatial location, can then be concatenated to form images of the subsurface, termed B-scans [1], [2], [7]. B-scans have one spatial axis, and one temporal axis. The signals returned from buried threats typically exhibit characteristic hyperbolic patterns in the B-scans, which can be leveraged for detection [6], [8]-[10]. Figure 1 shows several examples of Bscans collected over buried threats.Although it is possible to manually identify buried threats in GPR data, a great deal of published research has focused on automating this process with computer algorithms that provide a confidence of buried threat presence at each spatial location [4] A typical processing pipeline for supervised detection algorithms begins with a "prescreening"...

show abstract

Section: Cross-validation and Performance Metricsmentioning

confidence: 99%

Section: Cross-validation and Performance Metricsmentioning

confidence: 99%

On Choosing Training and Testing Data for Supervised Algorithms in Ground-Penetrating Radar Data for Buried Threat Detection

Reichman

Collins

Malof

2018

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

show abstract

“…In these systems, the multiple windows are either tested independently of each other and then their partial confidence values are combined, 14 or are combined into a bag representation and a multiple instance learning algorithm is used. 17,19,21,22 In our proposed approach, at each location, we treat the features extracted from the 10 windows as unlabeled data and solve for their labels simultaneously using (5). To obtain the final confidence value, we sort the 10 partial confidence values and average the largest three.…”

Section: Selection Of Unlabeled Datamentioning

confidence: 99%

“…Examples of feature-based discriminative classifiers that have been used in this application include K-Nearest Neighbor (KNN), 14 Hidden Markov Models, 15 Support Vector Machines, 16 Random Forest 13 and Multiple-Instance Learning. [17][18][19] Most of these techniques rely on constructing a predictive model to characterize data, and as a consequence are prone to over-fitting the training data.…”

Section: Introductionmentioning

confidence: 99%

A label propagation approach for detecting buried objects in handheld GPR data

Reid

Frigui

2016

Detection and Sensing of Mines, Explosive Objects, and Obscured Targets XXI

View full text Add to dashboard Cite

Detection of buried landmines and other explosive objects using ground penetrating radar (GPR) has been investigated for almost two decades and several classifiers have been developed. Most of these methods are based on the supervised learning paradigm where labeled target and clutter signatures are needed to train a classifier to discriminate between the two classes. Typically, large and diverse labeled training samples are needed to improve the performance of the classifier by overcoming noise and adding robustness and generalization to unseen examples. Unfortunately, even though unlabeled GPR data may be abundant, labeled data are often available in small quantities as the labeling process is tedious and can be ambiguous for most of the data. In this paper, we propose an algorithm for detecting landmines and buried objects that uses unlabeled data to help labeled data in the classification process. Our algorithm is graph-based and propagates the nodes labels to neighboring nodes according to their proximity in the feature space. For labeled data, we use a set of prototypes that are extracted from a small set of labeled training samples. For unlabeled data, we use a collection of signatures that are extracted from the vicinity of the alarm being tested. This choice is based on the assumption that many spatially close signatures are expected to have similar features and thus, unlabeled samples can create dense regions that link different regions of the labeled samples and propagate their labels to test samples. In other words, unlabeled samples are explored to create a context for each test alarm. To validate the proposed label propagation based classifier, we use it to detect buried explosive objects in GPR data collected by an experimental hand held demonstrator. We show that our approach is robust and computationally efficient to be used for both target discrimination and prescreening.

show abstract

“…The spectral information is a combination of the reflection and/or emission of sunlight across wavelength by objects on the ground, and contains the unique spectral characteristics of different materials [2], [3]. The wealth of spectral information in hyperspectral imagery enables the possibility to conduct sub-pixel analysis in application areas including target detection [4], [5], precision agriculture [6], [7], biomedical applications [8], [9] and others [10], [11].…”

Section: Introductionmentioning

confidence: 99%

Multiple instance hybrid estimator for hyperspectral target characterization and sub-pixel target detection

Jiao

Chen

McGarvey

et al. 2018

ISPRS Journal of Photogrammetry and Remote Sensing

View full text Add to dashboard Cite

The Multiple Instance Hybrid Estimator for discriminative target characterization from imprecisely labeled hyperspectral data is presented. In many hyperspectral target detection problems, acquiring accurately labeled training data is difficult. Furthermore, each pixel containing target is likely to be a mixture of both target and non-target signatures (i.e., sub-pixel targets), making extracting a pure prototype signature for the target class from the data extremely difficult. The proposed approach addresses these problems by introducing a data mixing model and optimizing the response of the hybrid sub-pixel detector within a multiple instance learning framework. The proposed approach iterates between estimating a set of discriminative target and non-target signatures and solving a sparse unmixing problem. After learning target signatures, a signature based detector can then be applied on test data. Both simulated and real hyperspectral target detection experiments show the proposed algorithm is effective at learning discriminative target signatures and achieves superior performance over state-of-the-art comparison algorithms.

show abstract

Multiple-Instance Hidden Markov Models With Applications to Landmine Detection

Cited by 29 publications

References 45 publications

On Choosing Training and Testing Data for Supervised Algorithms in Ground-Penetrating Radar Data for Buried Threat Detection

On Choosing Training and Testing Data for Supervised Algorithms in Ground-Penetrating Radar Data for Buried Threat Detection

A label propagation approach for detecting buried objects in handheld GPR data

Multiple instance hybrid estimator for hyperspectral target characterization and sub-pixel target detection

Contact Info

Product

Resources

About