Signal Processing Aspects of Polarimetric Random Noise Radar Data for Shallow Subsurface Imaging Yi Xu, Paul D. Hoffmeyer, Ram M. Narayanan, and John 0. Curtis* Department of Electrical Engineering and Center for Electro-optics University of Nebraska, Lincoln, NE 68588-0511, USA T: 402.472.5141 F: 402.472.4732 EMail:
[email protected] *U. S. Army Corps of Engineers, Waterways Experiment Station Environmental Engineering Division 3909 Halls Ferry Road, Vicksburg, MS 39180-6199, USA T: 601.634.2855 F: 601.634.2732 EMail:
[email protected]
Abstract-A novel polarimetric random noise radar system has been developed by the University of Nebraska for shallow subsurface probing applications. The radar system has been fabricated and tested, and its initial performance has been found t o be quite satisfactory in detecting and locating buried objects. The system produces images of the co-polarized amplitude, cross-polarized amplitude. depolarization ratio, and polarization phase difference of the radar reflccted signal as the antennas are scanned over t,he surface below which the objects are buried. Various signal processing algorithms are being explored to enhance target detection and clutter suppression. Since the radar system provides polarization phase differences between the orthogonal receive channels, algorithms based on Stokes matrix processing are being explored to detect and identify specific targets. One of the main advantages of Stokes matrix processing is in detecting long and slender cylindrical targets, which are not clearly detected in the conventional images. Furthermore, the optimal use of thresholding and smoothing operations to reduce and eliminate clutter is being examined. Examples of the preprocessed and post-processed images are presented. INTRODUCTION The use of radar techniques to detect, locate and identify buried shallow subsurface objects is of considerable interest in recent years [l].The University of Nebraska has developed a polarimetric random noise radar system used mainly for detecting sliallowly buried minelike objects. This novel Ground Penetrating Radar (GPR) system was designed, built and tested over the last two years. Simulation studies [a] and performance tests [3] on the system confirm its ability to respond to phase differences in the received signal, despite the fact that the probing waveform in random noise. This GPR system uses a wide bandwidth random noise signal operating within the 1-2 GHz frequency range. A block diagram of the system is shown in Fig. 1. High spatial resolution in the depth (range) dimension is achieved due t,o the wide bandwidth of the transmit signal. Complete details on backscatter data ac-
quired by this system from various targets are provided in [4]. The radar system is operated and controlled by a PC and the data acquired is stored in the hard drive in the real time. From the raw data, the system produces four images corresponding to the co-polarized receive amplitude, cross-polarized received amplitude, depolarization ratio .and polarimetric phase difference between the orthogonally polarized received signals. The polarimetric random noise radar system was used t o gather data from a variety of buried targets from a specially designed sand box 3.5 m long and 1.5 m wide. Targets that were buried included metallic as well as non-metallic objects of different size and shapes that mimicked land mines as well as other objects. These objects were buried at different depths and with different relative orientations. This paper describes two examples out of the entire data set collected that demonstrate the usefulness of St.okes matrix processing to enhance target detection. From the raw pre-processed images, we describe how the Stokes matrix images are computed. Simple image processing techniques, such as smoothing and thresholding algorithms are studied in order t o enhance detection, and these results are also shown. I M A G E ANALYSIS U S I N G S T O K E S M A T R I X FORMULATION From the raw data collected by the radar system, we generate images based on the Stokes matrix formulation for facilitating the detection and recognition of targets using the polarimetric information on the buried target. The Stokes vector is a convenient method for representing the polarization state of an electromagnetic wave, and is denoted as [SI, given by
(1)
The individual elements of the [SIvector are defined as follows:
T h e partial support of this work by the U.S. Army Corps of Engineers. Waterways Experiment Station under Contract #DACA3993-K-0031 is gratefully acknowledged.
0-7803-3068-4/96$5.000 1996 IEEE
2030
S,
=
21HllVICOSBd
(4)
5’3
=
21HllVIsinBd
(5)
In the above equations, Bd is the polarimetric phase angle, i.e, the difference between the phase angle of the horizontally received signal and the vertically received signal. Also, IHI and IVI are the electric field amplitudes of the horizontally and vertically polarized received signals, whose squared values represent the co-polarized reflected power and cross-polarized reflected power respectively (assuming the transmit polarization is horizontal). We recognize So as the total reflected power (sum of the co-polarized and cross-polarized reflected power). S1 is recognized as the difference between the co-polarized and cross-polarized reflected power. Sa is proportional to the cosine of the polarimetric phase angle, while 5’3 is proportional to the sine of the polarimetric phase angle B d . Both S2 and S3 are weighted by the absolute electric field amplitudes of the reflected co-polarized and cross-polarized signals, as can be seen from their definitions. It is also to be noted that
s,”= s; + s; + s;
(6)
The use of S2 and S3 is very helpful in detecting targets, since these parameters move in opposite directions, and thereby provides additional information about the reflected signal. When 5’2 is high, S3 is low, and vice versa. r h u s , no matter what the polarimetric phase angle, the target image is bound to show up in either Sa or S3, or sometimes in both. The above mentioned Stokes matrix images were generated, and combined with simple image processing operations to improve target detectability and clutter rejection. The smoothing filter is used for reduction of radar clutter and noise. It was found from the original raw data that high-frequency tonal variations were prevalent in regions without targets, and these grainy variations were attributed to the fact that the soil volume was inliomogeneous, and contained voids and rocks. The smoothing Dperation, when performed, results in low pass filtering and eliminates the high-frequency noise components. The thresholding operation is applied on the global scale to the entire smoothed image. It enhances image intensities sbove the mean intensity of the entire image, thereby enhancing target detectability, while simultaneously eliminating clutter, identified as low intensity areas, by setting these to zero digital number. As will be shown, these post-processing operations are successful in reducing clutter and enhancing target detectability. We emphasize here that currently, smoothing and thresholding operations have been investigated only for So and SI. In the following figures, the four pre-processed images show the co-polarized received power (top), the cross-
to two objects, one a round metal plate 23 cm in diameter and 2 cm thick, and the other i t wooden plate of the same shape and dimensions, is shown in Fig. 2. The objects are buried in dry sand a t 23 crrt depth each, with a lateral separation of 32 cm. The Stokes matrix processed images are shown in Fig. 3. Both objects, especially the wooden plate (right object) is detectable in the SIimage. We also show images of polarization sensitive objects to demonstrate the capability of the system t o utilize polarimetric features of the target in Fig. 4 and Fig. 5. In these figures, images were obtained for combinations of target orientation parallel t o (Fig. 4) and perpendicular to (Fig. 5) the scan direction. The transmit polarization was parallel to the longitudinal axis of the object, which was a metal pipe 6 cm in diameter and 85 cm long. When the transmit polarization was perpendicular t o the object axis, detection was not possible; hence, these images are not shown. From the processed images, we observe that a long slender object can be detected, no matter what its orientation is with respect to the scan direction, as long as the transmit polarization is parallel to the object orientation. This indicates that a dual-polarized transmitter, i.e, one that simultaneously or switchable transmits vertical and horizontal polarized signals can easily detect such an object. CONCLUSIONS From the experimental resullts obtained using the polarimetric random noise radar system, we find that it is capable of detecting shallowly buried objects. Application of the Stokes matrix image processing techniques provides a method for extracting the target polarimetric reflected properties for enhancing target detection and identification. Although the algorithm for image processing is relatively simple, it provides good results for enhancing target detectability. The experiments described in this paper were primarily performed with targets buried in dry sand, which is a low loss medium. Future study is needed to evaluate the radar system and its signal processing algorithms under more adverse conditions, such as wet clayey soil.
REFERENCES [l] D.J.Daniels, D.J.Gunton, and H.F.Scott, ”Introduction to subsurface radar,” IEE Proceedings Part F , vol. 135, pp. 278-320, August 1988. [a] R.M.Narayanan, Y.Xu, and D.W.Rhoades, ”Simulation of a polarimetric ramdom noise/spread spectrum radar for subsurface probing applications,” Proc. IGARSS’94, Pasadena, CA, pp. 2494-2498, August 1994. [3] R.M.Narayanan, Y.Xu, P.D.Hoffmeyer, and
Dolarized reccivcd power (second from top), the depolar-
J.O.Curtis, ”Design and performance of a po-
.zation ratio (second from bottom), and the polarimetiic phase angle bottom). The four post-processed images show SO(top left), SI(bottom left), S2 (top right), and 5’3 :bottom right). The pre-processed image corresponding
larimetric random noise radar for detection of shallow buried targets,” Proc. SPIE, Conf. on Detection Technologies for Mines and Minelike Targets, Orlando, FL, pp. 20-30, April 1995. 2031
Y.Xu, R.M.Narayanan, and [41 P.D.Hoffmeyer, J. 0.Curtis , "Backscatter characteristics of buried targets measured using an ultrawideband polarimetric random noise radar," Proc. IGARSS'96, Lincoln, NE, May 1996. Tronsnit Anteono
Receive
kntenno
ft2
Figure 3: Post-processed image of t,wo buried objects.
C-POL Phase
Phase L
Figure 1: Block diagram of tl!i> polarimetric random noise radar svstcm.
Figure 4: Post-processed image of slender pipe parallel to scan direction.
Figure 5 : Post-processed image of slender pipe perpendicular to scan dircction. Figure 2: Pre-processed image of two buried objects. 2032
