Alternating-Sign Windowed Energy Projection of SAR Focused GPR Data V. Kovalenko, A. Yarovoy, L.P. Ligthart Delft University of Technology, IRCTR, Mekelweg 4, 2628 CD, Delft, The Netherlands, +31/(0)15/278 7860 Abstract — The automated detection of plastic-cased an- tipersonnel landmines in 3-D high-resolution UWB radar images is considered. The 3-D images are projected in dif- ferent ways onto 2-D confidence maps, in which the auto- mated APM detection is made by marking local maxima in the maps followed by thresholding. Specifically, we consider the confidence maps of the following types: energy projec- tion (EP), windowed energy projection (WEP) and intro- duced here alternating sign windowed energy projection (ASWEP). The latter is shown to allow diminishing of the false alarms number in comparison for the other two tech- niques for most of the considered cases. The advantages of the new technique are described by means of receiver op- erator curves (ROC). The ROC curves are built for the datasets acquired over several minefield simulation sites under different environmental conditions. I. INTRODUCTION The use of the GPR for the anti-personal mine (APM) detection lately often means acquisition of the 3-D GPR datasets (C-Scans) followed by their specific processing, which often involves SAR-type focusing [1]. The final result of the processing can be either a list of 2-D detec- tion coordinates if GPR is used as a standalone technique or a 2-D confidence map of the investigated ground [2, 3]. The latter is the case when any kind of a fusion with other methods is supposed [4, 5]. Anyway a construction of a 2-D confidence map from a 3-D processed dataset is arising as a part of the problem. This construction often requires a projection of a 3-D distribution of some kind of feature onto a flat surface. We shall call this feature a confidence defining one. In the present paper we discuss three different projection techniques, each involving its own confidence defining feature: 1. The Energy Projection (EP) 2. The Windowed Energy Projection (WEP) 3. The Alternating Sign Windowed Energy Pro- jection (ASWEP) The EP is the most natural projection scheme where the confidence defining feature is just the energy scat- tered by the subsurface in the given location. Therefore the confidence maps constructed by this projection con- sist of pixels with clear mathematical and physical sense, which allows accurate characterization of the clutter and predictable performance of detectors. The disadvantage of the approach is that it often leads to a large number of false alarms present in the resulting map. To lower the amount of the FA in the resulting maps the windowed version of the energy projected was introduced [6]. The present paper introduces the ASWEP, which is a devel- opment of the idea of WEP further exploiting the a- priory knowledge of the distribution of the energy of mines in focused C-Scans. The paper is organized in the following way: we briefly consider the formulations of the EP and WEP and introduce ASWEP in the Section II where the data proc- essing scheme and APM Local Maxima based detector is also briefly considered. The performance of the three projection methods is comparatively considered with use of ROC curves in the Section III. The concluding re- marks are given in the Conclusion section IV. II. THE DATA PROCESSING AND PROJECTION SCHEMES A. The Projection Schemes The most natural projection scheme, the energy projec- tion (EP) is given by M z ij z ij C EP 1 , 2 (1) Where y x z C , ; is a focused C-Scan and z is a depth coordinate. The confidence maps, which are obtained according to (1), are in fact energetic images of the sur- face under the investigation. As it has been mentioned in the introduction these maps normally contain large amount of false alarms and threshold detectors based on them are therefore outperformed by other feature-driven ones [3]. The WEP [6] is the 3D 2D mapping, which utilizes the facts that the images of APM in the migrated C-Scans normally occupy a fraction of the total range of the depths. A typical SAR focused C-Scan acquired for the APM detection possesses a range depth of approximately 5 cm in the open air and to ~12 cm in soil. The vertical sizes of APM images fall between 3 and 6 cm. Thus the fraction of the depth slices in which valuable information is present varies depending on the mine type from 1/6 to 1/3 of the all present slices. The energy spread over rest of the slices contributes only to the clutter level. To get rid of these contributions while retaining the information on the APM images, the WEP was introduced with: L k L k z ij k ij z C EW 2 max (2) In (2) L is the width of the sliding window. The WEP given by (2) represents each A-Scan of the migrated C- Scan by maximum of the energy contained in a window of width L sliding along the A-Scan. In other words, (2) can be viewed as an L -norm of a convolution result between a focused A-Scan and a rectangular window. The choice of the window width is dictated by two con-