KEYWORDS: General packet radio service, Land mines, Target detection, Ground penetrating radar, Image compression, Analytical research, Detection and tracking algorithms, Explosives, Data modeling, Smoothing
Handheld ground penetrating radar (GPR) devices, such as the AN/PSS-14, produce image data for a detection sequence. Sequences contain sweeps of left to right and right to left swings of the device. By smoothing the image scan and examining local minima, we can determine the sweep ranges and turn around points contained within the data. Different filters are used to determine the interval between sweeps and approximate the exact turn around point for each sweep. Images are then annotated with the start and end of sweep locations. Results presented are both qualitative, based on comparison to labeling by humans, and quantitative, based on robot- collected data. Dynamic Time Warping (DTW) helps us align overlapping regions of a left to right sweep with its corresponding right to left sweep.
KEYWORDS: Land mines, General packet radio service, Ground penetrating radar, Analytical research, Antennas, Feature extraction, Target detection, Roads, Detection and tracking algorithms, Algorithm development
Multiple Instance Learning is a recently researched learning paradigm in machine intelligence which operates under conditions of uncertainty with the cost of increased computational burden. This increase in computational burden can be avoided by embedding these so-called multiple instances using a kernel function or other embedding function. In the following, a family of fast multiple instance relevance vector machines are used to learn and classify landmine signatures in ground penetrating radar data. Results indicate a significant reduction in computational complexity without a loss in classification accuracy in operating conditions.
Multiple instance learning is a recently researched learning paradigm in machine intelligence which operates under conditions of uncertainty. A Multiple Instance Hidden Markov Model (MI-HMM) is investigated with applications to landmine detection using ground penetrating radar data. Without introducing any additional parameters, the MI-HMM provides an elegant and simple way to learn the parameters of an HMM in a multiple instance framework. The efficacy of the model is shown on a real landmine dataset. Experiments on the landmine dataset show that MI-HMM learning is effective.
When processing ground penetrating radar (GPR) data for the detection of subsurface objects it is common to
align the data based on the location of the air-ground interface in order to eliminate the effects of antenna motion.
This practice assumes that the ground is mostly flat and that variations in the measured ground locations are
primarily due to antenna motion. In practice this assumption is often false so ground alignment will cause
true ground contours to be flattened, potentially distorting signatures from subsurface objects. In this paper
we investigate extracting edge histogram descriptor (EHD) features from GPR data with varying degrees of
alignment: unaligned, fully aligned and aligned only in the downtrack direction, where the effects of antenna
motion are most prevalent. One problem with not performing ground alignment is that features generated from
the ground surface or subsurface layers that follow the contour of the ground may cause false alarms. To address
this problem we also consider employing background subtraction prior to feature extraction on aligned data,
independent of the alignment method used for feature extraction. We compare the detection performance of
algorithms using each of these feature extraction approaches.
KEYWORDS: Detection and tracking algorithms, Land mines, Antennas, General packet radio service, Ground penetrating radar, Interfaces, Algorithm development, Reflection, Sensors, Data corrections
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.
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.
In remotely sensed hyperspectral imagery, many samples are collected on a given flight and many variable factors contribute to the distribution of samples. Measurements made at different flight times over the same swath may result in different spectral responses due to various environmental conditions and sensor calibration. Many classification methods attempt to classify a sample using labeled datasets or a priori information about the samples.
We present a possibilistic context-based approach for class estimation within a random set model. This approach includes novel formulations for model parameters with an intuitive base in probability and measure theory. This approach implicitly retains contextually correlated information in the data and uses it to estimate class labels in the presence of unknown factors-hidden contexts. This new method is applied to AHI (hyperspectral) imagery for the purposes of landmine detection. The results are compared to conventional methods and analyzed.
Sensor fusion has become a vital research area for mine detection because of the countermine community's conclusion that no single sensor is capable of detecting mines at the necessary detection and false alarm rates over a wide variety of operating conditions. The U. S. Army Night Vision and Electronic Sensors Directorate (NVESD) evaluates sensors and algorithms for use in a multi-sensor multi-platform airborne detection modality. A large dataset of hyperspectral and radar imagery exists from the four major data collections performed at U. S. Army temperate and arid testing facilities in Autumn 2002, Spring 2003, Summer 2004, and Summer 2005. There are a number of algorithm developers working on single-sensor algorithms in order to optimize feature and classifier selection for that sensor type. However, a given sensor/algorithm system has an absolute limitation based on the physical phenomena that system is capable of sensing.
Therefore, we perform decision-level fusion of the outputs from single-channel algorithms and we choose to combine systems whose information is complementary across operating conditions. That way, the final fused system will be robust to a variety of conditions, which is a critical property of a countermine detection system. In this paper, we present the analysis of fusion algorithms on data from a sensor suite consisting of high frequency radar imagery combined with hyperspectral long-wave infrared sensor imagery. The main type of fusion being considered is Choquet integral fusion. We evaluate performance achieved using the Choquet integral method for sensor fusion versus Boolean and soft "and," "or," mean, or majority voting.
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.