Near Surface Geophysics, 2004, 49-57 © 2004 European Association of Geoscientists & Engineers 49 A comparison of segmentation techniques for target extraction in ground-penetrating radar data S. Shihab* and W.Al-Nuaimy Dept. of Electrical Engineering, University of Liverpool, Brownlow Hill, Liverpool L69 3BX, UK Received September 2003, revision accepted December 2003 ABSTRACT In a typical GPR survey, only a small fraction of the collected data actually represent useful data (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 con- sumes 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 sta- tistical features from A-scan segments while the second technique computes statistical fea- tures from B-scan regions. In the third technique, some regional properties of B-scan seg- ments 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 test- ed 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. INTRODUCTION The growing use of ground-penetrating radar (GPR) as a non-destructive subsurface detection tool during the recent years has highlighted the need to develop not only the sys- tem itself, but also the accompanying processes necessary to produce an accurate final interpretation. In order to present the GPR displays in a more suitable condition for interpretation, some preprocessing as well as post-processing operations are carried out.These operations are usually handled manually by human operators, and con- sequently they consume considerable amounts of time and effort. In addition, these operations are exposed to the human inconsistency and error factor. Thus, it became nec- essary to find an automatic mechanism to carry out all the processing needed for the desired operation. Ground-penetrating radar (GPR) radargrams consist of a collection of time-series returns, viewed stacked side-by- side as a depth profile of the subsurface, and often displayed as an intensity-modulated raster plot of echo-strength vs. traveltime. Each vertical line corresponds to the signal received by the receiving antenna at a particular point above the surface. Due to the nature of the antennae and the geometry of the data acquisition arrangement, the trans- mitted signal propagates along more than one path towards the receiving antenna. The received signal is thus the combi- nation of the reflections along these different paths. In order to identify the subsurface targets amidst the surrounding clutter, it is therefore necessary to locate and distinguish the genuine target reflections from spurious reflections. Typically such data is processed off-line by a combination of manual and automated processing stages. Separating out genuine targets from background clutter and accounting for various environmental, system and subsurface effects require operator skill, experience and, most significantly, time.The analysis and interpretation of the large volumes of data generated by practical GPR surveys are extremely challenging and often present an implementation bottle- neck, influencing the cost-effectiveness and applicability of the technique. GPR images are unlike conventional images in that they * sass@liv.ac.uk