Scribd
Upload
51 views4 pages

Application of Multiresolution Analysis in Removing Ground-Penetrating Radar Noise

This document discusses the application of multiresolution analysis (MRA) using wavelet transforms to remove noise from ground-penetrating radar (GPR) data. MRA allows GPR data to be decomposed into different frequency bands, enabling efficient filtering. A synthetic model and field example demonstrate how MRA can enhance GPR data quality by removing white noise and ringing noise. Key detail coefficients containing noise are identified and removed, preserving reflection signals. MRA provides an effective technique for improving the signal-to-noise ratio of GPR data.

Uploaded by

chat_watchara
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
51 views4 pages

Application of Multiresolution Analysis in Removing Ground-Penetrating Radar Noise

This document discusses the application of multiresolution analysis (MRA) using wavelet transforms to remove noise from ground-penetrating radar (GPR) data. MRA allows GPR data to be decomposed into different frequency bands, enabling efficient filtering. A synthetic model and field example demonstrate how MRA can enhance GPR data quality by removing white noise and ringing noise. Key detail coefficients containing noise are identified and removed, preserving reflection signals. MRA provides an effective technique for improving the signal-to-noise ratio of GPR data.

Uploaded by

chat_watchara
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Application of Multiresolution Analysis in Removing Ground-penetrating Radar Noise

Yih Jeng*, Chuan-Hung Lin, Yi-Wei Li, Chih-Sung Chen, and Hsin-Han Huang
Department of Earth Sciences, National Taiwan Normal University, Taipei, Taiwan, ROC
[email protected]

Summary
The wavelet method has already been applied extensively to image processing and potential-field data, but
rarely to GPR data. Multiresolution wavelet analysis (MRA), one of the most effective wavelet techniques
used in image processing, allows an image to be described in terms of a rough shape, plus details in a broad
range which may be useful for filtering processes. The 2D GPR section is similar to an image in all aspects
if each data point of the GPR section is considered to be an image pixel in general. By this method, the
MRA provides an efficient filtering basis. A synthetic model study followed by a field example is presented
to demonstrate the feasibility of this technique in the filtering process for enhancing the S/N ratio of the
GPR data. We paid special attention to the data with serious ringing or scattering noise sources, where this
method proved to be very useful.
Introduction
Due to the kinematic similarities between electromagnetic (radar) and seismic wave propagation, GPR data
are usually processed with a technique similar to the technique used in seismic data. Although a multi-
channel GPR system is currently available, the most common GPR survey still acquires data in single fold
for the purposes of prompt field operation and low cost. Without the stacking procedures used in seismic
CMP processing, the S/N ratio enhancement of GPR records can be achieved only in the subsequent data
processing. In this paper, we propose an MRA technique based essentially in wavelet analysis for noise
suppression. The MRA is a well established mathematical tool in wavelet analysis for image compression
(Stollnitz et al., 1995) and de-noising (Mohideen et al., 2008), and has been introduced to the geophysical
data processing area in recent years with limited published literature (Matos and Osorio, 2002; Nuzzo and
Quarta, 2004). Most of the related studies were applying the wavelet transform to suppress noise in a time-
frequency sense (Deighan and Watts, 1997; Leblanc et al., 1998; Miao and Cheadle, 1998).
The operation of MRA is to recursively calculate the average and difference of the pixel values for each row
(or column) pairwise until an overall average (approximation coefficient) is obtained, and the differencing
values for each step are details (detail coefficients). By this way, the MRA provides an efficient filtering
basis to suppress events of specific scales locally but leave the rest of the data unaffected in general. In this
study, we show the application of this technique to enhance the S/N ratio of GPR sections.
Theory and Methods
There are two ways we can perform the MRA, i.e., the standard and nonstandard decompositions (Stollnitz
et al., 1995). The standard decomposition is to first apply the one-dimensional wavelet transform to each

Frontiers + Innovation – 2009 CSPG CSEG CWLS Convention 416


row of pixel values. The process yields an average and details for each row. We then apply the one-
dimensional transform to each column of these transformed rows as if they were themselves an image. The
final pixels of the transformed image are all detail coefficients except for a single overall average coefficient
at the first row, first column.
The nonstandard decomposition is an interchange between operations on rows and columns. First, we
perform one step of horizontal pairwise averaging (low-pass filtering, noted by L) and differencing (high-
pass filtering, noted by H) in each row of the image pixels, followed by applying vertical pairwise averaging
and differencing to each column of the previous horizontally transformed pixels. The subsequent process is
to recursively perform the transform only on the quadrant containing averages in both directions (Fig. 1).
Both of these decompositions have their advantages, depending upon the application. For the purpose of 2D
multiresolution filtering, we prefer the nonstandard decomposition because at each level of decomposition,
it results in a low-resolution approximation (LL quadrant in Fig. 1a) and three types of different resolution
details, i.e., horizontal (LH quadrant in Fig. 1a), vertical (HL quadrant in Fig. 1a), and diagonal (HH
quadrant in Fig. 1a), which would be easier for us to identify the signal and noise components. Moreover,
the nonstandard decomposition requires fewer operations and has square supports for each basis function.
Fig. 1b illustrates the hierarchical structure of nonstandard 2D decomposition.

(a) (b)

Figure 1: (a) Nonstandard decomposition of 2D discrete wavelet transform. (b) Filter bank of nonstandard 2D decomposition
where S represents the original image, An the approximation coefficient of nth level. Hn , Vn , and Dn are the horizontal,
vertical, and diagonal detail coefficients of nth level, respectively.

Examples
Synthetic model study
In order to carry out a systematic investigation of using MRA in the GPR data filtering, a controlled
synthetic model study is performed to provide an empirical filter bank for use in reconstructing the signal.
Fig. 2a is a noise free 2D synthetic GPR model section with four horizontal reflection events. To test the
effects of the 2D MRA filtering, we then added white Gaussian noise (S/N = 1) to the model (Fig. 2b). The
signal events of the model are nearly invisible once the noise was added. The reconstructed image after 2D
MRA filtering is shown in Fig. 2c, in which the components H1, D1, V1, H2, D2, V2, H3, V3, D3, D4, and V4
containing mostly white noise were zeroed off to obtain a global filtering.

Frontiers + Innovation – 2009 CSPG CSEG CWLS Convention 417


(a) (b)

(c)

Figure 2: (a) 2D GPR reflection model without noise. (b) Noise added model. (c) MRA processed result.

Field example—Removal of ringing noise


Applying the technique developed in the model study, we take a real GPR section recorded at a relic site at
the Taichung Municipal Park in central Taiwan. The data show a strong local ringing section at distance
between 24.3 m and 25.2 m (Fig. 3a), which may be caused by a metallic lid of an abandoned cistern near
the earth surface. The first stage MRA results of this section indicate that the detail components of H1, V1,
D1, H2, V2, D2 and D3 contain strong ringing noise, and should be removed. H3 was not removed in the
beginning because it contained both reflection signal and ringing noise. To retrieve the signal in component
H3, we treated it as an original image and applied the MRA procedure again to that single component. The
result of decomposing H3 indicates that the components of H23, D23, V23, H33 contain mostly ringing noise
(H23 denotes the second level horizontal detail coefficient of H3, and so forth). The components beyond
level 3 of the original image decomposing were trivial; therefore no further action was taken. Fig. 3c is the
filtered results of removing original detail coefficients H1, V1, D1, H2, V2, D2, D3, H23, D23, V23, and H33.
Conclusions
We have demonstrated the MRA technique in primary and secondary decomposition as well as how to use
them for suppressing random noise and removing ringing noise with less distortion of GPR signals. Before a

Frontiers + Innovation – 2009 CSPG CSEG CWLS Convention 418


more robust algorithm is developed, acquiring reasonable knowledge of the noise pattern to the data would
help the investigator to resolve the noise components for signal enhancement.

(a) (b)

(c)

Figure 3: (a) Original field section with local ringing noise. (b) MRA filtered result by zeroing off components H1, V1, D1, H2, V2,
D2, and D3. (c) Further suppression of ringing noise by removing sublevel components of H23, D23, and V23 in (b).

Acknowledgements
This research was supported financially in part by the NSC of Taiwan, ROC Grant NSC 95-2116-M-003-002.

References
Deighan, A. J., and Watts, D. R., 1997, Ground-roll suppression using the wavelet transform: Geophysics, 62, 1896–1903.
Leblanc, G. E., Morris, W. A., and Robinson, B., 1998, Wavelet analysis approach to de-noising of magnetic data: SEG Expanded Abstract,
554-557.
Matos, M. D. and Osorio, P. M., 2002, Wavelet transform filtering in the 1D and 2D for ground roll suppression, SEG Expanded Abstract,
2245-2248.
Miao, X., and Cheadle, S. P., 1998, Noise attenuation with wavelet transforms: SEG Expanded Abstract, 1072-1075.
Mohideen, S. K., Perumal, S. A., and Sathik, M. M, 2008, Image de-noising using discrete wavelet transform: International Journal of Computer
Science and Network Security, 8, January, 213-216.
Nuzzo, L., and Quarta, T., 2004, Improvement in GPR coherent noise attenuation using τ - p and wavelet transforms: Geophysics 69, 789–802.
Stollnitz, E. J., DeRose, T. D., and Salesin, D. H., 1995, Wavelets for Computer Graphics: A primer, Part 1: IEEE Computer Graphics and
Applications, 15, May, 76-84.

Frontiers + Innovation – 2009 CSPG CSEG CWLS Convention 419

You might also like