You are on page 1of 5

2017 9th International Conference on Measuring Technology and Mechatronics Automation

Leakage Region Detection of Gas Insulated Equipment by Applying Infrared Image


Processing Technique
Dangdang Dai1, Xianpei Wang1, Ying Zhang2, Le Zhao1, Junwei Li2
1 School of Electric Information, Wuhan University, Wuhan, 430072, China
2 Electric Power Research Institute, Guizhou Power Grid Co., Ltd., Guiyang, 550002, China
xpwang@whu.edu.cn

Abstract- Gas insulation equipment are widely used in high With the development of on-line leakage detectors, more
voltage electrical field, which use SF6 gas as insulation medium and more new devices are applied to the power system. In
and have many advantages such as low maintenance and high [7], SF6 gas leak is measured by using ultraviolet radiation
reliability. But SF6 leakage caused by factors such as for generating ion. However, the leakage region fails to be
manufacturing, installation and material aging will severely given by this method. In [8], SF6 gas leak is measured
reduce the insulation performance of equipment and based on spectrum absorption theory, the system has fast
eventually lead to the occurrence of insulation fault. Therefore, responding speed and the concentration of the SF6 gas can
the detection of SF6 leakage is of great significance to the safe be acquired by calculating the absorbance. However, it is
operation of the equipment. The traditional off-line detection
also unable to locate the leakage region. In [9], infrared
methods have disadvantages of low degree of automation, low
imaging technique is used to detect SF6 gas leak. Firstly, an
sensitivity and low precision, while the existing on-line leak
detectors are unable to locate the leakage region accurately. infrared camera is used to collect field video data. Then, the
Therefore, this paper proposes a SF6 leak detection method by leakage region is located by human observing. Although
using infrared image processing technique. First of all, original this method has greatly improved the detection efficiency,
infrared image is improved by using histogram equalization artificial judging can still lead to high false rejection rate
and wiener adaptive filtering algorithm. Then the rapid and false detection rate.
location of leakage region is realized by using improved For the above reasons, a method based on infrared
Surendra algorithm. Theoretical simulation and field test show image processing technique is proposed to detect SF6
that this method can realize the fast and accurate location of leakage region in this paper, which can complete the
the SF6 leakage region. automatic, fast and accurate detection of the leakage area so
that the workload and the misjudgment rate of artificial
Keywords- Detection SF6 Leakage Region; Infrared Image judgment can be reduced. Firstly, histogram equalization
Processing; Improved Surendra Algorithm strategy is used to enhance the contrast of the infrared
image, enlarge the dynamic range of image grey value and
I. INTRODUCTION improve the image gray levels. Then, the image is de-noised
by adaptive wiener filtering algorithm, by which the
In recent years, the use of SF6 electrical equipment is
Gaussian noise is eliminated and the high frequency
becoming increasingly wider. SF6 circuit breakers have
information and boundary information of the image are
already displaced the traditional oil circuit breakers. SF6
preserved. Finally, the SF6 gas leak region is located by
transformers have also acquired fast popularization and
improved Surendra algorithm.
application [1, 2]. However, because of the difference of
material quality and production processes, SF6 gas may leak II. BASIC PRINCIPLE OF SF6 INFRARED
from some equipment with the increase of usage time and IMAGING LEAKAGE DETECTION TECHNOLOGY
the aging of the equipment, which will severely reduce the
insulation performance of the equipment and eventually SF6 gas and air have different infrared absorption
lead to the occurrence of insulation fault [3, 4]. The properties, which makes it easy to observe SF6 gas that leaks
detection of SF6 leakage is of great importance to the safe into the air through infrared imaging detection technique
operation of the equipment and the reliability assurance of [12, 13]. The principle block diagram is shown in Figure 1.
the power grid [5, 6].
The traditional off-line SF6 leakage detection methods
can be divided into two categories, which are qualitative SF6
detection and quantitative detection. Qualitative detection
methods include the famous vacuum chase leaks et al; Infrared
Infrared
Quantitative detection methods include local bandaging object video
camera
method and hang bottle method et al. These traditional image
detection methods have low intelligentialize degree and
need a huge manpower and material resources. In addition, Air
the detection results are not convincing because of their
great subjectivity [7].
Figure 1. Principle block diagram of infrared imaging detection

2157-1481/17 $31.00 2017 IEEE 94


DOI 10.1109/ICMTMA.2017.29
The traditional way to recognize leakage region is
realized by human observing of the acquired infrared video
image, which may leads to high false rejection rate and false
detection rate. With the development of digital image
processing technology, it is considerable to locate the
leakage region by moving target detection. The detailed
procedure of the proposed method is shown in Figure 2,
which includes the following 5 steps:
(1) Read the original video data, extract time-series images
and convert each image to gray image.
(2) Image preprocessing to the each gray image, including
image cropping, image enhancement, image de-noising et al. Figure 3(a). An original infrared image
(3) Use moving target detection algorithm to detect
sequence image motion area.
(4) Edge detection to the result of step (3), extract the
contour of the edge and show it in the original image.
(5) Compare the test results with real spill area.
III. INFRARED IMAGE PREPROCESSING
Due to the fact that the shooting process can be easily
affected by the external environment, the background
acquired by infrared imaging leak detector is dark, which
makes the contrast between the spill area and background
very small and difficult to detect the leakage region directly. Figure 3(b). The cropped image
So preprocessing to the original image is needed so as to
enhance the contrast of the original image, restrain In order to enhance the contrast of the image after
background noise and improve the signal-to-noise ratio. cutting, histogram equalization is performed on the cropped
Taking a SF6 circuit breaker video taken by an infrared image [14, 15]. Histogram equalization processing is a kind
camera as example, one frame of the original infrared video of histogram correction method that based on cumulative
is shown in Figure 3(a). To improve the efficiency of the distribution function transformation, the discrete form of the
algorithm, first of all, the original image is cropped in order cumulative distribution function can be expressed as (1):
to remove the redundant data and retain only the required k n k
Sk T (rk )
j
image area. The image after cropping is shown in Figure pr (r j )
j 0 N j 0 (1)
3(b).
( 0 d rk d 1 k 0,1, 2,..., L  1)
Start Where L is the grey scale number, Pr (r j ) is the probability
that level grey value is j . n j is the number of the pixel that
Read infrared in level grey j . N is the total pixels of the image.
video image data Histogram equalization can expand the dynamic range of
the pixel gray value and improve the image gray level,
which will make the detail of the image clearer and enhance
Image preprocessing the contrast of the image.
As mentioned above, due to the difference of the
instrument specifications, the shooting jitter and other
environmental factors, noise will be introduced in the
Moving target detection imaging process. Gaussian noise is the main component,
which need to be removed [16]. The median filter has good
effect in filtering random noise and pulse interference,
which can retain the edge information of the image
Contour extracting effectively but fail in detail protection. The wavelet
threshold de-noising algorithm can extract the weak signal
in the image, but lose the high frequency information. In
order to remove Gaussian noise while preserving the detail
Evaluation algorithm information of the original image, adaptive wiener filtering
accuracy algorithm is adopted in this paper to de-noise the original
image.
Figure 2. The process of infrared video image processing

95
A N u M window is selected as the estimated Where, I i and Ii 1 are the current frame of image and last
neighborhood of the local mean and local variance of each frame of image, respectively. TH is the selected threshold,
pixel in the image. It is easy to obtain equation (2)(3). abs means absolute value operation.
1
P I ( x, y )
NM x , yK
(2) (4) Morphological processing to BWi and conducting
connectivity analysis, get the revised binary image FBWi ;
1 (5) Update background BGi based on FBWi ;
V2
NM
K I 2
( x, y )  P 2 (3)
x , y
BGi 1 ( x, y) if FBWi ( x, y) 1
BGi ( x, y) (6)
Where, P and V are the local mean and local variance of
2
qI i  (1  q) BGi 1 ( x, y) if FBWi ( x, y ) 0
the neighborhood, respectively. K is the N u M Where, BGi ( x, y) and FBWi ( x, y) are the value of the
neighborhood range of each pixel in image I ( x, y) . So the background image and the binary image in coordinate (x, y) ,
new pixel value can be expressed as equation (4): q is the speed of update, which is set 0.1.
V 2  v2 (6) Until the background of all pixels in one image is
I c( x, y) P  ( I ( x, y)  P ) (4)
2
V updated, set i=i+1, back to (3) until i=100. BG100 is the
2
Where v is the variance of the noise, which can be extracted background.
estimated and adjusted according to local variance of the (7) Get the differential image Di 1 ( x, y) ;
image I ( x, y) . Di 1 ( x, y) Ii 1 ( x, y)  BGi ( x, y) (7)
IV. LEAKAGE REGION DETECTION (8) Select an appropriate threshold, binarization processing
and morphological operation to Di 1 ( x, y) so as to
In a short period of time, the leakage region is fixed, SF6
gas exists as mist in the video. Due to the high internal get FDi 1 ( x, y) .
pressure, SF6 gas will diffuse irregularly. So it is (9) Use canny edge detection algorithm for contour
considerable to locate leakage region by moving target extraction.
detection. There are three commonly used moving target (10) Mark the leakage region in the original image.
detection algorithms, which are optical flow method [17], Start
inter-frame difference method [18, 19] and background
subtraction division method [20, 21]. Read infrared video
image data
Constraints must be added in the optical flow algorithm
when estimating the optical flow. As optical flow can be i=0,take the first frame of
image I0 as the original
affected by the object form and field noise, the detection of background BG0
moving target will be affected accordingly. Although the
inter-frame difference method has strong anti-interference Set an appropriate iteration
ability, it suffers difficulty in threshold selection, which will number Clet i=i+1
make it difficult to extract the whole moving region and
Get the current
result in empty phenomenon. Surendra algorithm can differential binary
effectively extract the whole moving region. However, it image BWi
fails to process the results after frame difference furtherly
and can mistake the foreground that has similar texture as Morphological processing
and connectivity analysis End
background. This will make the updated background Get the revised
inaccuracy, the detection of the motion area can be affected image FBWi
Mark the
accordingly. leakage region

For the above reasons, an improved Surendra algorithm Yes


FBWi(x,y)=1
No
proposed in [9] is adopted in this paper to realize rapid, Use canny edge
detection algorithm
complete and precise location of leakage region. The BGi=q*Ii+ for contour
BGi=BGi-1 extraction after
process of the algorithm is shown in Figure 4, which (1-q)BGi-1
morphological
includes the following 10 steps: operations
Get the current
(1) Take the first frame of image I 0 as the original background BGi
Select an
background BG0 , let i=0; Calculate the
appropriate
No differential
Yes threshold
(2) Set iteration number C=100 ; i=C image D of the
get the binary
background and
(3) Get the current differential binary image BWi ; the next frame
image FD of D

1 if abs( Ii  I i 1 ) t TH
BWi (5) Figure 4. Process for the detection of leakage region
0 if abs( Ii  I i 1 )  TH

96
V. TEST RESULTS AND DISCUSSION The red circles in Figure 7 show the real leakage region
As shown in Figure 5(a), the histogram equalization of detected by soap method. Another two detection results that
3(b) significantly improved the contrast of the original respectively obtained by the traditional Surendra algorithm
image. Meanwhile, by comparing Figure 5(b) with 5(c), it is and the improved Surendra algorithm are compared with
easy to find that the gray scale distribution area of the artificial off-line detection.
original image is significantly enlarged, which will decrease Figure 8 shows the background extraction results of the
the difficulty in leakage region recognition. same image by the traditional Surendra algorithm and the
improved Surendra algorithm. The background extracted by
the classical Surendra algorithm can be easily affected by
the interference in cross-region, so there are many outliers
in the obtained differential binary image, which results in
misjudgments. The improved algorithm obviously has better
detection effect. Although the iterative efficiency of
background update is slightly reduced, the error points in
the differential binary image also decreases, which
significantly improves the detection accuracy

Figure 5(a). After image enhancement

3000

2500

2000

1500
Figure 6(a). The differential binary image without filtering process
1000

500

0 50 100 150 200 250

Figure 5(b). The Gray histogram of original image

4000

3500

3000

2500

2000
Figure 6(b). The differential binary image after adaptive wiener filtering
1500
process
1000

500

0 50 100 150 200 250

Figure 5(c). Histogram equalization

Adaptive wiener filtering algorithm can effectively


remove the random noise and improved the detection
precision. As can be seen in Figure 6(a) and 6(b), there is a
large number of noise in 6(a), which is the differential
binary image that obtained by background subtraction
division without filtering. By contrast, noise in 6(b) is
significantly reduced as it is filtered by adaptive wiener Figure 7. The real leakage area of SF6
filtering algorithm.
The real leakage region can be obtained by artificial off-
line detection, which can be used as a verification check.

97
Morphology processing and
Differential binary image Leakage region recognition
contour extraction

Improved
algorithm

traditional
algorithm

Figure 8. The result of leakage area detection

[8] JIANG Baolin, SUN Jiquan, XU Liang, et al. Apparatus Measuring


VI. CONCLUSION SF6 Gas Leak Utilized Ultraviolet Radiation for Generating Ion[J].
Modern Instruments, 2005(04): 56-58.
To solve the problem that the existing on-line SF6 [9] XU Yuanzhe, HU Zhihui, LIU Xian, et al. Research on the
leakage detection methods are unable to effectively Technology of Infrared Imaging Optical SF6 Breaker Leakage
recognize the leakage region, a method based on infrared Detection[J]. Jouranal of Electric Power,2009, 24(1): 12-15.
video image processing technique is proposed in this paper. [10] HU Weitao,SUI Shaochen,HAN Jianbo. Application of Gas Find IR-
LW in Condition-based Maintenance of SF6 Electrical Equipment[J].
Through comprehensive theoretical analysis, histogram High Voltage Apparatus,2010, 46(10):90-92.
equalization and adaptive wiener filtering algorithm are [11] YUAN Jiwei,SHI Zhongke. An Algorithm of Automatic
used to improve the original infrared images. An improved Background Ex traction of Target Based on Fast Motion[J].
Surendra algorithm is also selected to optimize the Application Research of Computers, 2004, 21(8):128-129.
[12] LI Jie. Research of SF6 Leakage Detection Based on Infrared
acquisition of image background. The proposed method Video[D]. Xi-an University of Science and Technology,2014.
significantly improve the detection precision of the moving [13] LIU Jing, ZHANG Mingxiang, LIU Jun. Application of SF6 Leak
target and realize the rapid and accurate positioning of SF6 Infrared Detection Technology in Anhui Grid[J]. ANHUI
leakage region, which greatly improve the efficiency of ELECTRIC POWER, 2012, 29(2):23-24.
leakage detection and repair of the gas insulated equipment. [14] MA Lin, YU Ning. Analysis of Digital Image Enhancement
Algorithm[J]. Technology & Economy in Areas of Communications,
ACKNOWLEDGEMENTS 2011, 13(1):122-125.
[15] LI Nan, LIU Yongqiang, HAN Guohua. Application of histogram
The authors are grateful to the support from the Guizhou equalization in digital image processing[J]. Gan-su Science and
Power Grid Co., Ltd. Technology, 2010, 26(18):59-60.
[16] TANG Lin, LIU Lin, SU Junhong. Modeling and Simulation
REFERENCE Research of Infrared Image Noise[J]. Infrared Technology, 2014,
36(7):542-548.
[1] YU Lixin, QI Xiaoyan, LI Chao, et al. Investigation and Research on [17] Iketani A, Kuno Y, Shimada N, et al. Real-time Surveillance System
the Application of SF6 at Typical 500 kV Substations[J]. Northeast Detecting Persons in Complex Scenes: Image Analysis and
Electric Power Technology, 2015, 36(6): 7-9. Processing, 1999. Proceedings. International Conference on,
[2] LIU Weihong. Analysis of Structure and Technique for High 1999[C]. IEEE.
Voltage SF6 Potential Transformer[J]. High Voltage Apparatus, [18] Cheng Y H, Wang J. A Motion Image Detection Method Based on
2005, 41(2): 146-147. the Inter-Frame Difference Method, Beijing, China, 2014[C]. Trans
[3] ZHANG Ying, YU Pengcheng, LI Junwei, et al. Discussion of Tech Publications Ltd, 2014.
Leakage Analysis and Management Measures to SF6 Electrical [19] Zheng H, An W J, Li Z. A Motion Vehicle Detection Method Based
Equipment Leakage in Guizhou Power Grid[J]. High Voltage on Self-adaptive Background Subtraction with Cumulative Inter-
Apparatus, 2015, 51(9):200-204. frame Difference[M]//2013:890-894.
[4] WANG Wei, FENG Xinyan, NIU Lin, et al. SF6 Gas Leakage [20] JI Qinghua, YU Suping. Object Detection Algorithm Based on
Detection Using Infrared Imaging Method for GIS[J]. High Voltage Surendra Background Subtraction and Four-Frame Difference[J].
Apparatus, 2012, 48(4): 84-87. Computer Applications and Software, 2014, 31(12):242-244.
[5] Barnett R P, Nelson J H. SF6 Gas Monitoring and Leakage Detection [21] QU Jingjing, XIN Yunhong. Combined Continuous Frame
in Gas Insulated Switchgear[C]// International Conference on High Difference with Background Difference Method for Moving Object
Voltage Engineering and Application. 2016: 676 - 679. Detection[J]. Acta Photonica Sinica, 2014,43(7):219-226.
[6] LI Sha. New Design of SF6 Leak Detector[J]. Machine Building & [22] TANG Lulu, ZHANG Qican, HU Song. An Improved Algorithm for
Automation, 2013, 42(2): 60-61. Canny Edge Detection with Adaptive Threshold[J]. Opto-Electronic
[7] YUAN Shiqi, DAI Zhou, CHEN Fang. Comparison of SF6 Leakage Engineering, 2011, 38(5):127-132.
Detection Methods for High-VoltageElectrical Equipments[J].
Southern Power System Technology, 2013, 7(2): 54-58.

98

You might also like