Beruflich Dokumente
Kultur Dokumente
org/
ABSTRACT
INTRODUCTION
Use of potential-field data to locate edges facilitates geologic interpretation because the edges on the potential-field often indicate
subsurface faults, contacts, and other tectonic features. However,
the horizontal boundaries of potential-field-anomaly sources cannot
be determined directly based on the potential-field data because of
the high frequency and low amplitude with respect to the source
boundaries. It is often necessary to enhance the horizontal edges
of the sources to assist structural interpretation, as well as environmental and engineering investigations. To date, several methods exist to emphasize geologic contacts and highlight source edges,
including methods based on horizontal and vertical derivatives
(e.g., Evjen, 1936; Nabighian, 1972; Cordell, 1979; Blakely and
Simpson, 1986; Mitova and Jarosalav, 1993; Hsu et al., 1996;
Fedi and Florio, 2001). However, these high-pass filters mainly detect the edges of high-amplitude sources but are not effective for
Manuscript received by the Editor 11 September 2014; revised manuscript received 10 June 2015; published online 13 October 2015.
1
Formerly China University of Geosciences, Institute of Geophysics and Information Technology, Beijing, China; presently China Aero Geophysical Survey
and Remote Sensing Center for Land and Resources, Beijing, China. E-mail: yysun2009@126.com.
2
Chinese Academy of Geological Sciences (CAGS), Institute of Geology, State Key Laboratory of Continental Tectonics and Dynamics, Beijing, China.
E-mail: yangwencai@cashq.ac.com; zeng.xz@163.com.
3
China University of Geosciences, Institute of Geophysics and Information Technology, Beijing, China. E-mail: zhangzhi.yong2008@163.com.
2015 Society of Exploration Geophysicists. All rights reserved.
G1
Sun et al.
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
G2
mpq
n X
m
X
Gf u ; f v f pu f qv f u f v ;
(1)
v1 u1
m20
N X
M
X
g2x xj ; yi ;
(2)
g2y xj ; yi ;
(3)
gx xj ; yi gy xj ; yi ;
(4)
i1 j1
m02
N X
M
X
i1 j1
m11
N X
M
X
i1 j1
where
gx x; y
gx; y;
x
gy x; y
gx; y;
y
(5)
where i 1;2; : : : ; N , j 1;2; : : : ; M , and gx; y is the potential field data within moving window w1. Here, m20 and m02 show
the variance of slopes of the small potential field surface within w1
in the x- and y-directions, respectively, whereas m11 denotes the
covariance of the slopes in x- and y-directions within w1 .
Note that the spectral moments change with the rotation of the
coordinates. Therefore, the statistically invariable quantities that are
independent of the system rotation are defined as (Longuet-Higgins,
1957; Huang, 1984, 1985; Yang et al., 1992; Li et al., 2000)
M2 m20 m02 ;
(6)
(7)
M Ma Mb .
(8)
p
n20 n02 n211
Ma sgnM2
;
n20 n02
2
(9)
where n20 , n02 , and n11 are the three elements of the spectral moments on the surface of M2 within the moving window (referred to
as w2 ) with size of M by N points, and represents the Laplacian
operator. The opposite sign of M2 is set to separate the ridge and
valley lines because it is the ridge lines that we are interested in.
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
In equation 9, the edge coefficient component Ma is proportional to the square root of 2 , showing the anisotropy of the surface
topography, which contributes to narrow the edges. Additionally, it
is also inversely proportional to M2 , enhancing the recognition ability to weak edges on the surface.
G3
Figure 1. Comparison of several edge-detection methods. (a) Synthetic gravity data set. A, B, and C are the source bodies whose outlines are
shown in solid lines. The depths of prisms A, B, and C are 5, 5, and 2 km, respectively; the density of prisms A, B, and C are 0.03, 0.5, and
0.02 gcm3 , respectively; (b) TDX; (c) profile curvature with zero contours overlaid; (d) LFA; (e) TAHG; (f) theta map; (g) NSTD; and
(h) edge coefficient computed using equation 8.
G4
Sun et al.
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
Mb sgnM 2
2
Epro
arctan
; (10)
2
Spro
Wx
W x xj ; yi
proxj ; yi q
Wx2 Wy2
Wy
q W y xj ; yi ;
Wx2 Wy2
(11)
where
Figure 2. Edge detection by existing methods. (a) Synthetic gravity anomalies, adding random noise with an amplitude of 0.1% of the anomalous maximum to the data in Figure 1a; A, B, and C are the source bodies whose outlines are shown in solid lines; (b) TDX; (c) profile
curvature; (d) LFA; (e) TAHG; (f) theta map; (g) NSTD; and (h) edge coefficient computed using equation 8.
1 X
proxj ; yi ;
N M i1 j1
v
u
N M
u 1 X
X
proxj ; yi Epro2 .
Spro t
N M i1 j1
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
Epro
(12)
1 X
Wx
W x ; y ;
N M i1 j1 x j i
Wy
1 X
W x ; y
N M i1 j1 y j i
(13)
and
Wx; y
p
M2 x; y;
Wx; y;
x
W y x; y Wx; y.
y
W x x; y
G5
(14)
Figure 3. Edge detection using the spectral-moment method with different window sizes. (a) Edge coefficient of the noise-corrupted synthetic
gravity anomalies in Figure 2a using a radius of three and three points for w1 and w2 , respectively. (b and c) Edge coefficient of the noisecorrupted synthetic gravity anomalies using a w2 radius of three points and a w1 of eight and 14 points, respectively. (d) Edge detection using a
w1 and w2 radius of eight and eight points, respectively.
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
G6
Sun et al.
of the y-axis) of source body C when projected onto the plane surface. Similar to the results in Figure 1b, detection by LFA (in Figure 1d) is also dominated by the response from larger amplitude
anomalies, although the outlines of smaller amplitude anomalies
(source C) can be delineated slightly better. TAHG (in Figure 1e)
is an effective method to balance the amplitudes of large- and smallamplitude anomalies. The edges of bodies A and B are successfully
delineated by the theta map method (Figure 1f), but it is less effective for body C, which has the lowest amplitude. In addition, the
edges recognized by this method appear very broad, which is
not conducive to locating source boundaries. Figure 1g shows
the NSTD using a window of 3 3 points. The amplitudes of
the different anomalies are balanced well, and it clearly delineates
the edges of the three source bodies, whereas the delineated outlines
are a little discontinuous. Figure lh shows the results of the spectralmoment method using the edge coefficient component M. As
mentioned above, two moving window sizes should be selected
to calculate the image of M; one (w1 ) of which is for computing
the image of M2 , whereas the other one (w2 ) is for calculating the
image of M. In this case, the radius of both Gaussian moving windows (w1 and w2 ) is three points. It is clear that the edge in this
figure appears the narrowest among all the above methods. In addition, the amplitudes of detected edges from different sources are
not only significantly balanced but are also continuous.
To test the relative sensitivity of the edge-detection methods discussed above to noise, a small amount of uniformly distributed random noise with amplitude of 0.1% of the maximum data magnitude
was added to the data in Figure 1a. Figure 2a shows the noise-corrupted anomalies, and the outlines of the sources are shown in solid
lines. The TDX method appears to be affected least by the noise.
Although the profile curvature (Figure 2c), LFA (Figure 2d), TAHG
(Figure 2e), and theta map (Figure 2f) methods can still locate the
source edges, the detected edges, especially the edges of smaller
amplitude anomalies, are blurred because of the added noise. Figure 2g is the NSTD of the data in Figure 2a using a larger window
size with 7 7 points. The edges of the smaller amplitude anomaly
(e.g., body C) are faint. Figure 2h is the image of the edge coefficient M of the data, which is calculated using w1 8 and w2 3.
Although several spurious edges, present in all the images, are
inevitably detected, it provides a better result than the other methods. The reason why the spectral-moment method is less sensitive to
noise than are the others is that the M2 of each grid point is computed using its surrounding data contained within w1 .
We also show how the Gaussian window size affects the resulting
images (Figure 3). We chose a w2 radius of three for w1 radius sizes
of three, eight, and 14 to calculate the edge coefficient from data in
Figure 2a, and the results are shown in Figure 3a3c, respectively.
When the radius of w1 equaled three (Figure 3a), the noise severely
Figure 4. Geologic sketch map of China. The areas covered by Figures 5a and 7a are shown. The pink color indicates platforms or massifs and
the blue color indicates orogenic or tectonic belts.
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
G7
Figure 5. Test cases of edge-detection methods based on the satellite gravity anomaly from a small section of northern China. The location is
shown in Figure 4. (a) Satellite gravity anomaly over a section of northern China; (b) TDX; (c) profile curvature; (d) LFA; (e) TAHG; (f) theta
map; (g) NSTD; and (h) edge coefficient computed using equation 8; the black ellipses indicate examples that edges detected by the spectralmoment method are clearer than by the other methods.
Figure 6. Map of the edge coefficient with the
simplified geologic map of the study area superimposed. Solid lines indicate the fault and dotted
lines indicate the buried faults. Dashed ellipses
indicate examples that edges are detected by
the geophysical method but not identified in
the geophysical map.
G8
Sun et al.
affected the detected boundaries, in particular, of body A. Comparing the map of Figure 3b and 3c, one can see that the larger sizes of
w1 can be less sensitive to the noise than are the smaller ones, but
the edges of the small-scale anomaly are also smeared out (Fig-
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
G9
(Figure 8a), the magnetic anomalies in this region are mainly caused
by serpentinites and eclogites. In this case, we have also compared
the detected results by the spectral-moment method with the simplified geologic structures. Figure 8b shows the map of edge coefficient with the simplified geologic map (Yang et al., 2005; Xu
et al., 2009) of this study area superimposed. The edges in the
dashed ellipse in Figure 7h correspond to the largest Mica-Eclogite
of the simplified geologic map. In addition, these small features enhanced by the spectral-moment method should be the edges of the
small magnetic objects, such as serpentinites or ultrabasic intrusion.
DISCUSSION
The most impressive advantages of this spectral-moment method
are its ability to balance the edges of large- and small-amplitude
anomalies and narrow the detected edges. In fact, the parameter
Sun et al.
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
G10
(in equation 10) can measure the ridge strength of the surface in
the moving window (Li et al., 2000). The edge coefficient has better
balancing ability because it is computed based on the ridge strength
that is independent of the magnitude of the edges. Furthermore, calculating parameter on the surface M2 is equivalent to extracting
the ridges of the edges detected by the parameter M2, which contributes to narrowing the extracted edges.
Another advantage of the spectral-moment method is that when
comparing with the other edge-enhancement methods mentioned in
this paper, it shows that it is less sensitive to the noise than are the
others when balancing the edges of different-amplitude anomalies.
Although the proposed method is not aimed at suppressing noise,
the moving window computation can make it is less sensitive to the
noise when enlarging the moving window. However, although the
results depend on the sizes of the moving windows, there is a lack of
standards for the selection of window sizes. One has to carry out
some tests before selecting appropriate window sizes.
Because the spectral-moment method picks out all edges, it is
usually difficult to interpret on its own. Interpretation efforts should
combine as many processed images as are deemed appropriate.
CONCLUSION
We have presented a new edge-detection method for the enhancement of potential anomalies in which the edges are identified by the
edge coefficient. The spectral-moment method was demonstrated
using synthetic and real data. For synthetic data, the method shows
better capabilities of balancing the edges of different amplitude
anomalies and narrowing the detected edges. In addition, larger
windows (w1 ) are less sensitive to noise than are small ones; however, the edges, smaller than the window size of the small-scale
anomaly, are also smeared out. Furthermore, the size of the moving
window w2 mainly controls the width of the detected edges. Using
field gravity and magnetic data reduced to the pole as examples, the
edge coefficient map provides more details, and the lineaments generated by the proposed method are consistent with geologic
structures.
ACKNOWLEDGMENTS
The authors gratefully acknowledge the financial support from
the National Science Foundation of China under grant number
41574111 and the Chinese Geological Survey with the project number 12120113099000. The authors thank assistant editor V. Socco,
the associate editor, and three reviewers for their constructive comments that greatly improved the original manuscript. We also thank
Y. Wang for providing the simplified geologic map.
REFERENCES
Blakely, R. J., and R. W. Simpson, 1986, Approximating edges of source
bodies from magnetic or gravity anomalies: Geophysics, 51, 14941498,
doi: 10.1190/1.1442197.
Cooper, G. R. J., 2009, Balancing images of potential-field data: Geophysics, 74, no. 3, L17L20, doi: 10.1190/1.3096615.
Cooper, G. R. J., and D. R. Cowan, 2006, Enhancing potential field data
using filters based on the local phase: Computers and Geosciences,
32, 15851591.
Cooper, G. R. J., and D. R. Cowan, 2008, Edge enhancement of potentialfield data using normalized statistics: Geophysics, 73, no. 3, H1H4, doi:
10.1190/1.2837309.
Cooper, G. R. J., and D. R. Cowan, 2009, Terracing potential field data:
Geophysical Prospecting, 57, 10671071, doi: 10.1111/j.1365-2478
.2009.00791.x.
Downloaded 04/08/16 to 41.73.242.244. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
Zhang, H. L., R. Dhananjay, Y. R. Marangoni, and X. Y. Hu, 2014, NAVEdge: Edge detection of potential-field sources using normalized
anisotropy variance: Geophysics, 79, no. 3, J43J53, doi: 10.1190/
geo2013-0218.1.
G11