S. Shihab

Neural network target identifier based on statistical features of GPR signals

Accurate and consistent manual interpretation of the vast quantities of GPR data collected during a typical survey constitute an implementation bottleneck that often limits the practicality and cost-effectiveness of this tool for rapid site investigation. Automatic unsupervised interpretation of GPR data is achieved by training a neural network to discriminate between signals originating from different types of targets and other spurious sources of reflections such as clutter. This is achieved by computing a number of statistical data descriptors for feature extraction. The neural classifier is capable of returning 3-dimensional image outlining regions of extended targets (such as reinforced concrete, disturbed soil or storage tanks) and pinpointing the location of localised targets such as mines and pipes. These reports are accompanied by a written log detailing the depths and geometry of these targets. This classifier was applied to a variety of GPR data sets gathered from a number of sites. The obtained results were in close agreement with those obtained by a trained operator manually, but in a fraction of the time. Different targets have been successfully discriminated, with a consistency greater than that of the operator. Although the system is implemented in software, the rate at which classifications are rendered lends the system Authors would like to thank the Engineering and Physical Sciences Research Council (EPSRC) for funding this work as a part of a larger project regarding automatic data-processing of ground penetrating radar. Authors would like also to express their gratitude to Zetica (UK) Ltd. for supporting this work financially, and providing sites data and related software. favourably to near real-time on-site processing and interpretation.

Automatic target detection in GPR data

Automatic detection and characterization of the signatures of solid reflecting targets in ground-penetrating radar data is achieved by a combination of signal and image processing stages. For the class of target under consideration, namely localized or extended linear reflecting targets such as landmines, pipes or cables, the reflections exhibit a broad hyperbolic anomaly in the region of the target. Detection and characterization of these distinctive signatures yields information about the location of the targets as well as the surrounding medium. Edge enhancement and edge processing techniques are developed to trace the envelope of the reflected wavefronts. By fitting hyperbolae to these detected edges, the location of the targets and the relative permittivity of the medium are estimated. This estimate enables the effective elimination of the background clutter that leads to spurious non-hyperbolic reflections. Thus automatic target detection and mapping is achieved without the heavy computational demands of techniques such as synthetic aperture radar processing, enabling on-site data interpretation.

Time-frequency characteristics of ground penetrating radar reflections from railway ballast and plant

As increasing amounts of government and industry money are invested into the upgrading and maintenance of the European railway infrastructure, there is a growing demand for rapid and reliable non-invasive techniques for the characterisation of railway trackbed (ballast). Ground-penetrating radar (GPR) has been proposed as such a geophysical tool, employing pulses of high frequency electromagnetic radiation to characterise and profile the electrical properties (permittivity and conductivity) of the subsurface. Despite the popularity of this site investigation technique, scientific techniques and results have yet to be published presenting a systematic means to quantify and evaluate the GPR returns, correlating these with the degree of ballast deterioration. This study identifies a number of features extracted from time-frequency representations of GPR returns that are capable of accurately characterising the degree of ballast deterioration in accordance with a number of different ballast deterioration models.

A comparison of segmentation techniques for target extraction in ground penetrating radar data

In a typical GPR survey, a great amount of data is collected, and only a small percentage of this amount represent useful data (i.e. target data) whereas the majority of the data is considered non-useful. The first of the post-processing stages, which relies heavily on a skilled operator, involves pointing out the areas that may contain targets and suppressing others. Consequently, this process consumes considerable amounts of time and effort, apart from the fact that the existence of the human factor at this critical stage invariably introduces inconsistency and error into the interpretation. In this paper, three techniques for automatic GPR data segmentation are discussed and compared. The techniques rely on the computation of certain features from which a neural network can then arrive at a decision to classify the data segments in question as targets or otherwise. The first technique is based on extracting three statistical features from A-scan segments while the second technique computes statistical features from B-scan regions to produce the discrimination bases for the neural classifier to distinguish targets from non-targets. In the third technique, some regional properties of B-scan segments accompanied by the statistical features are used to achieve discrimination not only between targets and non- targets, rather between hyperbolic-shaped and non-hyperbolic-shaped targets as well. All three techniques were on different types of GPR data collected from a variety of sites, and they proved to be very efficient to form a robust automatic technique for data reduction and segmentation.

Automatic mapping of linear structures in 3-dimensional space from ground-penetrating radar data

Non-invasive geophysical techniques such as ground-penetrating radar allow rapid and low-cost investigation of the shallow subsurface for the detection of such features as utilities and plant. This paper presents a pattern recognition approach based on the 3-dimensional Hough transform for the detection of extended linear targets in ground-penetrating radar data. By transforming spatially extended patterns into spatially compact features in parameter space, a difficult global detection problem in data space becomes a more easily solved local peak detection problem in parameter space. Due to the sparseness and variability of the data, the accumulator peak detection stage is replaced by a novel algorithm called the adaptive non-accumulated Hough transform (ANHT) 3-dimensional clustering algorithm. This technique allows the combination of qualitative site information and ground truth in order to increase the accuracy of the final result. The user is presented with a 3-dimensional site survey report detailing the length, depth and orientations (azimuth and zenith) of any pipes, cables or the like. The ANHT performs substantially superior to the standard Hough transform implementation in computer memory requirement. Our experimental results on the artificial 3-dimensional linear objects indicate that method works quite well under various background conditions. The automatic mapping of linear structures in 3-dimensional space from ground-penetrating radar data is achieved by implementing the ANHT in the detection of linear targets in ground penetrating radar data.

A comparison of segmentation techniques for target extraction in ground-penetrating radar data

In a typical GPR survey, only a small fraction of the collected data actually represent usefuldata (i.e. target data), whereas the majority of the data is considered redundant. The first of the post-processing stages, which relies heavily on a skilled operator, involves indicating those areas that may contain targets and suppressing others. Consequently, this process consumes considerable amounts of time and effort, apart from the fact that the existence of the human factor at this critical stage invariably introduces inconsistency and error into the interpretation. In this paper, automatic detection and segmentation techniques for GPR data are discussed and compared. The techniques rely on the computation of certain features from which a neural network is then able to arrive at a decision whether to classify the data segments in question as targets or otherwise. The first technique is based on extracting statistical features from A-scan segments while the second technique computes statistical features from B-scan regions. In the third technique, some regional properties of B-scan segments are used to achieve discrimination not only between targets and non-targets, but also between hyperbolic-shaped and non-hyperbolic-shaped targets. All the techniques were tested on different types of GPR data collected from a variety of sites, and they proved to be very efficient in forming a robust automatic technique for data reduction and segmentation. In addition, these techniques are carried out in near real-time enabling on-site processing and interpretation of collected data.

Comparison between surface impulse ground penetrating radar signals and ultrasonic time-of-flight diffraction signals

Surface impulse ground-penetrating radar (GPR) and ultrasonic time-of-flight diffraction (TOFD) are recent innovations in the respective geophysical remote sensing and non-destructive testing industries. Both techniques have proved highly versatile and valuable applications. This paper provides a brief description of the time-of-flight diffraction (TOFD) signals and the surface impulse ground penetrating radar (GPR) signals. The similarities between the two techniques are highlighted and it is shown how processing techniques developed for GPR data processing may be adapted for use with TOFD data.

Multi-Channel Filtering Approach for Unsupervised Segmentation of Subsurface Radar Images

The volume of image data generated in ground-penetrating radar surveys can severely restrict the practicality of this site investigation technique. This is particularly true in situations where automatic analysis or interpretation is required, as segmentation and classification tasks that utilise multivariate data are critically affected by the volume and dimensionality of the data. A general-purpose unsupervised image segmentation system is presented here for the automatic detection of image regions exhibiting different visual texture properties. A bank of Gabor filters is used to achieve multi-channel filtering analogous to the processing of information in the visual cortex. A suboptimal feature-selection procedure is proposed to automatically select the set of texture features best suited for the particular application. The reduction in the size of the feature set both reduces the computation time and improves the accuracy of the final classification. Unsupervised neural networks are employed to segmen...

Unsupervised segmentaiton of subsurface radar images

The volume of image data generated in ground-penetrating radar surveys can severely restrict the practicality of this site investigation technique. This is particularly true in situations where automatic analysis or interpretation is required, as segmentation and classification tasks that utilise multivariate data are critically affected by the volume and dimensionality of the data. A general-purpose unsupervised image segmentation system is presented here for the automatic detection of image regions exhibiting different visual texture properties. A suboptimal feature selection procedure is proposed to automatically select the set of texture features best suited for the particular application. The reduction in the size of the feature set both reduces the computation time and improves the accuracy of the final classification.

Automatic 3D mapping of features using GPR

Although GPR is normally capable of detecting the responsefrom buried plant, accurate detection and mapping of extended geometrical features in 3-dimensional data is often a major problem faced by the radar operators and geophysicists. This paper presents a pattern recognition approach based on the 3-dimensional Hough Transform for the detection of extended linear targets. By transforming spatially extended patterns into spatially compact features in parameter space, a difficult global detection problem in data space becomes a more easily solved local peak detectionproblem in parameter space. This technique allows the combination of qualitative site information and ground truth in order to increase the accuracy of the final result. Improved freedom of movement and accuracy is achieved by logging the movement of the GPR unit using DGPS. The user is presented with a 3-dimensional site survey report detailing the length, depth and orientations (azimuth and zenith) of any pipes, cables or the like.