Designing an Illegal Mining Detection System based on DinSAR.

Full text

Page 1

DESIGNING AN ILLEGAL MINING DETECTION SYSTEM BASED ON DINSAR

Zhe Hu, Linlin Ge, Xiaojing Li and Chris Rizos

Cooperative Research Centre for Spatial Information & School of Surveying and Spatial Information
Systems,
the University of New South Wales, Sydney, NSW, 2052, Australia

ABSTRACT

Satellite Differential Radar Interferometry (DInSAR) has
demonstrated its ability for monitoring mine-induced
ground subsidence. However, it is still a challenging task to
routinely identify all mining activities from the large-scale
coverage interferogram, especially the illegal mines. In
response to this challenge an underground mining detection
system based on DInSAR is described. The system is tested
over a dense mining area in Asia. With such a system it is
hoped that the detection efficiency of illegal underground
mining using DInSAR can be improved.

Index Terms— Differential SAR Interferometry,
Underground Mining Detection, Deformation Gradient

1. INTRODUCTION

Ground subsidence is a serious threat to surface facilities,
infrastructure, buildings, real estate or even human lives.
This phenomenon is considerably worse in underground
mining areas as excavation of minerals, etc., reduces the
support of the ground surface above the mine area. Close
monitoring therefore must be maintained in order to predict
and/or minimise the impacts of land subsidence. The task
can be generally be undertaken by using ground surveying
methods, such as digital levels, total stations and global
positioning system (GPS) techniques. However, in the case
of illegal underground mining such techniques are
inadequate because the locations of the mining are (in
general) not known. In this case, satellite Differential Radar
Interferometry (DInSAR) could be used to monitor mining
across a large region.
The satellite DInSAR technique has demonstrated its
advantages and capability of monitoring land subsidence
caused by underground mining activities, as reported in [1]
and [2]. Compared to traditional surveying methods it is
able to simultaneously satisfy the goals of high accuracy
and large coverage. Meanwhile, recent developments in
relation to Synthetic Aperture Radar (SAR) satellites ensure
that there will be sufficient data for long term surveillance.
To obtain the monitoring result, a raw interferogram image
is first generated, in order to estimate the ground
deformation patterns. It is obtained by calculating the phase
(travel time) difference of radar echoes between the SAR
satellite’s two passes over the same area, separated by
multiples of the satellite repeat period. Then, the
topographic phase contribution to the raw interferogram is
removed, with the residual phase (differential interferogram)
mostly attributable to ground deformation. If the magnitude
of deformation is required, a few subsequent processes will
be implemented, such as phase unwrapping [3], obtaining
the spatial deformation signature between two SAR image
acquisitions from the interferogram.
Thanks to the large-scale coverage of SAR images, the
mining status and mine site locations over the whole region
can be easily extracted from DInSAR results. However, it is
challenging to locate all illegal mines across a vast region
using manual processes. To deal with this problem, an
Illegal Mining Detection System (IMDS) is proposed in this
paper. A demonstration of the IMDS will illustrate the
performance of this system.

2. SYSTEM DESIGN AND METHODOLOGY

2.1. Principle of Illegal Mine Detection

In order the locate mines from a DInSAR interferogram, the
unique attributes of the mining subsidence and its pattern in
the interferogram can be utilised. A typical pattern due to
mining subsidence is also shown in Figure 1.

Figure 1. Typical pattern of mining subsidence in an
DInSAR interferogram
A. The first obvious characteristic is that the deformation
due to mining is that the surface under which mining occurs
will sink (shown in Figure 1 as the colour of the subsidence
fringes changing from purple to red). The greatest

Page 2

subsidence mostly occurs in the surface centre of
underground mining activities, and the subsidence
magnitude decreases from centre to edge, finally forming a
spatial funnel in the area.
B.
Another characteristic that can be deduced from
attribute A is that since the deformation typically has a
funnel shape, the mining centre is surrounded by slopes. It
also means that the absolute values of the deformation
gradient near the centre should be larger than the area where
there is no deformation occurring. Meanwhile, the
directions of the gradient are approximately pointing outside
the pattern with reverse extensions to the subsidence centre.
The 2D gradient is expressed as a complex number, where
the magnitude represents the strength of the gradient and the
phase indicates its direction:
()
,f x y
x
∂
C.
Also, as shown in Figure 1, it can be concluded that
the pattern is typically a round or oval shape. This is
because the land surface can generally be considered as
elastic [4], which means that the deformation spreads stably
from the centre. Consequently the contours of deformation
fringes represent a closed circle in the DInSAR’s line-of-
sight (LOS) direction.
Based on these attributes it is possible to distinguish the
patterns of subsidence due to underground mining from
DInSAR results, and locate their positions automatically.
The processes are described in the following section.

2.2. Processing Flow of Automatic Detection

The detailed processing flow of IMDS is illustrated in
Figure 2, consisting of six modules.

(
)(
∂
)
,,f x y
∂
f x y
y
i
∂
∇=+
(1)

Figure 2. Processing flow of the IMDS
The most important module of IMDS’s initial steps is
Phase Unwrapping, which is used to obtain the absolute
deformation values from the differential interferogram. It is
also a precondition to evaluate whether Attribute A is
satisfied or not. In order to obtain a high quality unwrapped
result, use of a mask is necessary [5]. It is used to block out
the areas that may induce errors in the result. As illustrated
in Figure 2, the Mask Generation module is therefore the
first to be implemented, right before the Phase Unwrapping
module.
The objective of Mask Generation identifies the areas
with dense residual points, which are too noisy and include
DEM (Digital Elevation Model) errors, and these will not be
unwrapped. In this paper, the determination of residual
point is done by examining if there is over π
change between nearby points. Then, a moving window is
adopted to check the density of the residual points. The
block with high density (over the predefined threshold) will
be treated as the masked area, finally building a full area
mask for the sequential Phase Unwrapping module. The
detailed procedure for Mask Generation is illustrated in
Figure 3.
±
phase

Figure 3. Detailed processes of the Mask Generation
module
After obtaining the absolute phase values of the
differential interferogram from Phase Unwrapping, the
gradient of deformation will be computed by the Gradient
Calculation module. Since the magnitude of gradient on the
edge of deformation area are the same and larger than those
in the non-deformating region, according to Attribute B,
contours are then generated by the Contours Generation
module. The gradient values on the contours can be
obtained from deformation modes for specific mining areas.
The reason why contours are computed on the gradient
map rather than implemented on the unwrapped
interferogram directly is that, although the topographic
phase has been removed from the DInSAR results, the
absolute phase values still have significant differences in
non-deformating areas. Take the example of phase caused
by atmospheric effects [7]. It will result in the same
apparent deformation values over long distances being
different in DInSAR results. However, since the gradient is
a relative value that represents the degree of deformation, it
is impacted little by the residual phase, hence it is more
suitable for generating contours.

Figure 4. Detailed processes of the Gradient Calculation
and Contours Generation modules
Referring to Attribute A, only subsidence patterns could
be caused by underground mining, but uplifting areas can
also have large magnitude gradients. Hence the contours
encircle rising patterns should be disregarded. Small
contours surrounded by large ones will also be removed.

Page 3

Then, the regions enclosed by the remaining contours are
candidates for underground mining activity locations. The
operations of the Gradient Calculation module and the
Contours Generation module are illustrated in Figure 4.
Correlation operations are implemented between mining
candidates enclosed by selective contours and two
references. The first one is a shape reference represented as
an oval, which is based on Attribute C. The size and the
curvature of the shape reference are determined by the
target candidate. Another one, referred to as the gradient
reference, is produced based on the selected contour. The
gradient magnitudes are determined by the gradient of the
contours. Referring to Attribute B, directions of all the
gradients are perpendicular to the contour line, pointing
outwards. The correlation is:
{
( )
EXE X
−
⎡
⎣
( )
E i is the mean value operator. When shape
correlation is implemented, X denotes the points on the
contour line and Y denotes the shape reference. If under
gradient correlation, X is the gradient of the unwrapped
region that is enclosed by the contours, and Y represents
the gradient reference. The final correlation coefficient used
for mining detection is:

mshape
Cora Cor
= ⋅
where a and b are the weighting parameters of shape
correlation and gradient correlation respectively. In the
following case study they are both set to 0.5. An example of
these two references for a typical contour is given in Figure
5.

()
( )
E X
( )
}
{}
( )
{}
22
,
EXY E Y
Cor X Y
EY E Y
−⋅−
⎡
⎣
⎤ ⎡
⎦ ⎣
⋅
⎤
⎦
=
−
⎤
⎦
⎡
⎣
⎤
⎦
(2)
where
grad
b Cor
+ ⋅
(3)

Figure 5. Example of references for a typical contour
After calculating the correlation coefficients for each
mining candidates, the ones with high correlation values are
then selected, confirming as mining locations. Finally, all of
these detected mines are then compared with the mining
plan to identify illegal mines.

3. CASE STUDY

To demonstrate the performance of the IMDS a case study
of underground coal mines (including illegal ones) in Asia
is examined.
The DInSAR interferogram for the mining region, is
shown in Figure 6, whose coverage area is approximately
70×70 km2. The interferogram is generated by using the
two-pass DInSAR method [6] with two ALOS (Advanced
Land Observing Satellite) PALSAR (Phased Array type L-
band SAR) images. The master image was acquired on 15
December 2007 and the slave one was acquired on 30
January 2008. Their track and frame numbers are 453 and
72 respectively. As can be seen from Figure 6 (left), there
are several areas with noise fringes, so a mask is generated
based on the density of residual points to ensure a high
quality unwrapped result. The mask and unwrapped
interferogram are shown in Figure 6.

Figure 6. Differential interferogram (left), the mask for
phase unwrapping (middle) and the unwrapped
result (right) of selected site
After obtaining the unwrapped result, gradients are
computed from it. Sequentially, contours can be obtained to
aid selection of mining location candidates. In order to
represent all the following processes more clearly, results
are zoomed-in to the area indicated by the rectangle in
Figure 6. The gradients of the zoomed-in area calculated
from the unwrapped interferogram are shown in Figure 7.

Figure 7. Magnitude (left) and phase (middle) of
unwrapped result, and generated contours (right) in
zoomed-in area
Contours are drawn using the gradient magnitude to
determine the mining location candidates. Positions and
shapes of final selected contours of the zoomed-in area are
shown in Figure 7 (right image). The corresponding regions
in the gradient map are mining location candidates and the
correlation with shape and gradient references will be
carried out. All of these contours are stored in a vector file
for sequential processing. From the information in these
vectors, there are 28 mining location candidates in the
zoomed-in region.
The correlations are computed based on Equation (3),
and values that are larger than 0.5 are considered as
indicating underground mining activity. For the zoomed-in
area, 17 out of 28 of the mining location candidates have
been identified as mining sites (Figure 8). Finally, bu

Page 4

comparing these results with an official exploitation plan,
the location of illegal mines can be identified.

Figure 8. Detection result of underground mining

4. CONCLUDING REMARKS

In this paper the DInSAR remote sensing technique has
been used to detect underground mining-induced subsidence
across a large-scale area. Based on the obtained differential
interferogram, an IMDS system has been proposed that
increases the detection efficiency of underground mining
activity. From a demonstration of its application across a
region in Asia, it can be seen that most mines have been
identified from the initial interferogram, and then the results
are used to detect the locations of illegal mines in the region.
In future work improvements can be implemented by
addressing the following problems. First of all, the detection
accuracy relies on the quality of the unwrapped
interferogram. Hence precise phase unwrapping algorithms
are required. Secondly, the threshold of the gradient
contours is currently based on processing experience and
prediction of mining subsidence. If a deformation model
could be developed for the mining area it could be used to
define the appropriate threshold value. Moreover,
correlations are implemented according to two aspects, the
shape and the gradient patterns. This may be too simplistic.
Ideally, a spatial 3D reference should be adopted, which
will fit the characteristics of mining subsidence better.

ACKNOWLEDGEMENT

This research work has been supported by the Cooperative
Research Centre for Spatial Information (CRC-SI), whose
activities are funded by the Australian Commonwealth’s
Cooperative Research Centres Programme. The Australian
Research Council and the Australian Coal Association
Research Program have been funding radar related studies
by the team at the University of New South Wales (UNSW)
during the last few years.
The authors wish to thank the Earth Remote Sensing
Data Analysis Centre (ERSDAC) for providing ALOS
PALSAR data. METI and JAXA retain the ownership of the
ALOS PALSAR original data. The PALSAR Level-1.1
products were produced and provided to the CRC-
SI/UNSW by ERSDAC, Japan.

REFERENCES

[1]. L. Ge, H.C. Chang, C. Rizos, and M. Omura, “Mine
Subsidence Monitoring: a Comparison Among Envisat, ERS, and
JERS-1,” in 2004 ENVISAT Symposium, Salzburg, Austria,
Session 3P04-09, 6-10 September 2004.

[2]. U. Wegmuller, C. Werner, T. Strozzi, and A. Wiesmann,
“Monitoring Mining Induced Surface Deformation,” 2004 IEEE
International Geoscience and Remote Sensing Symposium
(IGARSS04), Anchorage, Alaska, Vol. 3, pp. 1933-1935, 20-24
September 2004.

[3]. C.W. Chen, and H.A. Zebker, “Phase Unwrapping for Large
SAR Interferograms: Statistical Segmentation and Generalized
Network Models,” IEEE Transactions on Geoscience and Remote
Sensing, Vol. 40, pp. 1709-1719, August 2002.

[4]. X. Papademetris, E. Turan Onat, A.J. Sinusas, D.P. Dione,
R.T. Constable, and J.S. Dunca, “The Active Elastic Model,”
Proceedings of the 17th International Conference on Information
Processing in Medical Imaging, pp. 36-49, 2001.

[5]. J. Strand, and T. Taxt, “Two-dimensional Phase Unwrapping
Using Robust Derivative Estimation and Adaptive Integration,”
IEEE Transactions on Image Processing, Vol. 11, Issue 10, pp:
1192-1200, October 2002.

[6]. L. Ge, K. Zhang, A.H.M Ng, Y. Dong, H.C. Chang, and C.
Rizos, “Preliminary Results of Satellite Radar Differential
Interferometry for the Co-seismic Deformation of the 12 May
2008 Ms8.0 Wenchuan Earthquake,” Geographic Information
Sciences, Vol. 14, No. 1, pp. 12-19, June 2008.

[7]. L. Pipia, X. Fabregas, A. Aguasca, and C. Lopez-Martinez,
“Atmospheric Artifact Compensation in Ground-Based DInSAR
Applications,” IEEE Geoscience and Remote Sensing Letters, Vol
5, Issue 1, pp. 88-92, January 2008.