Optimal Control, Safety Technology, and Electromagnetic Compatibility in the Driving System of Electric VehicleView this Special Issue
An Electrical Insulator Defects Detection Method Combined Human Receptive Field Model
Nondestructive inspection of electrical insulators subjected to the high electrical stress and environmental damage is fundamental for reliable operation of a transmission lines. The breakage and defect of the insulator have great influence on the safe of transmission lines, and insulator defect detection with difference types is a complex work. This paper proposed an insulator defect detection method inspired by human receptive field model, which meets the requirements for detecting defect insulator in a simple background. In this method, the defect detection combined human receptive field model of human visual system is constructed and applied on the different insulators, so as to achieve accurate detection of the insulator defected parts. Experimental results show that the method can accurately and robustly detect the defect (such as cracks and damage) of electrical insulator in case of noise affect.
High-voltage transmission lines and transmission towers in China are usually in a harsh environment where their components are often eroded by rain or damaged by unpredictable foreign objects. When an electrical insulator is in operation at the high-voltage transmission line systems, those devices are subjected to a strong electrical stress and also damaged by the severely environmental conditions . It is well known that the presence of voids and inclusions are introduced in the manufacturing process, or generation and propagation of defects, such as holes and cracks inside and outside of the electrical insulators, when those devices are in operation under high-voltage stress, a partial discharge begins as localized dielectric breakdown. Furthermore, partial discharge can also occur along the boundary between different insulating materials, ultimately leading to electrical breakdown and, eventually, explosion with stop of energy transportation [2–4]. Therefore, the defect of electrical insulator is an important reason causing electrical network accident, so it is a crucial work to detect the defect of insulator periodically and minimize interruption of energy accidents. The traditional way of monitoring and troubleshooting is to conduct manual inspection of the videos and images of the transmission lines photographed by the helicopter. But it is too time-consuming and labor-intensive to get efficient defect detection, usually easy to make wrong judgements because of the large amount of data. At present work, D. Zuo et al. used SVM classification to detect missing insulators . Qaddoumi et al. introduced artificial neural network-based near-field microwave to accurately test damage of outdoor insulators . Armando et al. detected the cracks and operation defects in the insulators and breaker by a portable low cost computed radiography systems . Chen et al. extract single insulator disk in infrared image of insulator string . Huang et al. proposed some back propagation neural network method for equipment fault diagnosis [9, 10]. However, the above methods are operated in the special imaging situation to achieve the insulator defect detection, it has great limitation to the natural captured insulator images in damage detection. With the developing of high-speed digital imaging and processing technique, intelligent identification technology began to be used in high-voltage safe operation and long-distance power transmission lines inspection [11–16]. Figure 1 is the illustration of electrical safe operation and transmission inspection, the electrical insulator (glass/ceramic material) image is captured by UAV (unmanned aerial vehicle), which is suitable for several kinds electrical safe inspection. In view of the above details, this paper proposed an electrical insulator defect detection method inspired by human brain receptive field (RF) model, aiming to achieve accurate detection including defect, crack even dirty of different damaged insulators, so as to improve safety and reliability of high-voltage transmission line. Experimental results show that this method can quickly and accurately detect whether the insulator is broken or damaged.
The paper is organized as follows. In Section 2, we detailed the biology basis of human brain RF and the proposed method combined RF model. Section 3 discussed important parameter selection and experimental performance evaluation. Finally, we give a brief summary and draw a conclusion in Section 4.
2. Defect Detection Method Combined RF Model
2.1. Biology Basis of the RF Model
A large number of biological experiments show that the primary visual pathway of the human brain can obtain the main information of the object and play a key role in the overall perception of the object [17, 18]. In Figure 2(a), visual information reaches the eye through the cornea and pupil, passes through the photoelectric conversion of photoreceptor cells in the retina, then passes into the lateral geniculate nucleus (LGN), and projects to V1 area, finally creating visual perception of the brain [19, 20]. LGN is an important step to V1 area of HVS, Figure 2(b) represents two types of LGN RF model, that is, center-on (“+”) and center-off (“-”) cells. In these cells, the central region is excitatory (marked by red in center-on and blue in center-off) and the surrounding is inhibitory (color opposite to excitatory area), respectively. From Figure 2(b), it can be seen that the center of the on-type is surrounded by off-type and the center off-type around is surrounded by on-type. Both of these cells are arranged and their total number is substantially equal and detect contrast changes. Some researchers have proved that the on-type and off-type channels are located between the LGN and the visual cortex is fully parallel separation [21, 22]. The process of LGN receptive field is defined as a 2D Gaussian function in the following:In this equation, and represent spatial variables respectively, represents the type of cell (that is sensitivity of excitory and inhibition area of RF), which is “+” or “-”, is a mapping function of , when is “+”, a center-on RF cell is defined as , else is defined as the negative of , that is , so a center-off RF cell is . is the standard deviation of the outer Gaussian function, which represents the size of excitatory and inhibitory area of RF, usually is set by , which is in accordance with electron-physiological findings of LGN cells in mammals.
(a) Basic processing pathway of human visual system
(b) Illustration of classical receptive field. (The left is center-on LGN cells and the right is center-off LGN cells.)
2.2. Construction of the Detection Method with RF Model
For the above details, the proposed electrical insulator defect detection method is inspired by human brain RF model, which is followed in . Figure 3 shows the relationship of above described LGN cell and the main information of an image (object). In Figure 3, center-on LGN RF cell is marked by green and center-off LGN RF cell is marked by red, V1 simple cell RF is marked by blue circle. The center of RF is positioned on an edge (marked by points) which gives rise to model LGN cell responses rendered in the left two columns of Figure 3. The eight small points represent the RF centers of eight subunits, four of center-on (in green), and four of center-off (in red) type. For the considered example, it results in the inclusion of eight subunits in the model. The number of subunits depends on the number of circles we considered and the specific input pattern presented at the time of configuration. In this insulator defect detection method, each subunit included is represented in parametric form by a tuple , where the parameters represent the polarity of subunit, is the scale of the involved model LGN cells, the radius , and the polar angle of the RF center of the subunit relative to the RF center of the model cell [21, 23], respectively.
We denote by 4-tuples of the configured subunits of model. According to the right three subparts in Figure 3, we explain the general configuration process of the proposed detection method in more detail below. A cell group computes the sum of the weighted responses of the model LGN cells it receives input from. These models LGN cells have the same polarity (on or off), RF size and neighboring RFs. In this way, a subunit detects contrast changes, similar to a model LGN cell. A subunit can be thought of as a dendrite branch of a simple cell which receives input from a pool of adjacent LGN cells. In Figure 3, input image passes through LGN cell and gets their responses, several LGN cells with same polarity form a cell group. Afterwards, a plurality of LGN cells of the same attribute forms a cell group, a certain number of same polarity cell groups are arranged colinearly and different attribute cell groups are arranged in parallel, the interaction forms a V1 simple cell with different preference directions, and each output is simple cell response. Finally, different simple cell responses were integrated by weighted summation to obtain the overall responses of V1 simple cells. Figure 3 is the construction process with a simple image from RUG dataset [21, 23], with only specific cells preferentially oriented at 0, 90, 180, and 270 degrees.
The concentric receptive field of a single LGN cell can be characterized as (1). For an LGN cell that located at coordinate , the response can be obtained by convolving the intensity distribution function of the image with the cell receptive field, that is represented by the following:Considering the different may lead to being negative value, but the self-firing rate is positive in biology, the nonnegative half-wave operation [22, 24] is adopted to obtain single LGN cell response as follows:All the feature characteristic LGN cells can be combined to a group, which is defined as follows similar to .In (4), represents the quality of cell group included in each cell group. For each specified , and represent the single cell relative to the center of cell group. For the LGN cell group located at , the response is described as follows:The parameters satisfied the following conditions:Define as the simple cell response of the direction , which is determined by all cell group response in the defined region. In Hubel-Wiesel’s work, only the response of all LGN cell group included in given region can excite simple cell. In order to show the biological nature, all of the cell group responses are multiplied to get geometry average, so as to represent the simple cell response, which is mathematical explained in the following:In the above equation, it is satisfied as and , the direction function is defined as , , all of these simple cell responses are combined to get the overall response of V1 area in the following:
3. Performance Evaluation of the Method
3.1. Key Parameter Selection
The appropriate selection of parameter is a key step in the proposed detection method, because the structure of the electrical insulator is relatively simple, in order to show the detection efficiency by different parameters on insulator defect detection. In order to verify the parameter selection difference, we choose the image as an example from RUG (university of Groningen) image dataset . Because the texture of the natural image (such as basket) is more complex and more detailed than a simple insulator image, as shown in Figure 1. Figure 4(a) is the original image, Figures 4(b)–4(d) are the basic contour of original image in various parameters condition, and Figures 4(e)–4(g) are the binary detection result by different parameters, which takes 8, 12, and 16, respectively.
(a) Original image
(h) 0 degrees
(i) 101 degrees
(j) 214 degrees
(k) 270 degrees
(l) 326 degrees
The configured detection model cell response to a natural image is illustrated in Figure 4. When n=4, there are a large number of missed detections of the model and it is difficult to recognize the basic appearance of the object from the output information. The basic appearance of the object can already be basically resolved when n=8, but the included edge information is still incomplete. The output of the model contains relatively complete edges when n=12, which can clearly identify the appearance of the object. The last one in Figure 4(g) is n=16, the model will get more complete local contour information. As the above analysis, the parameter is selected as n=16 in our experiment. When the orientation selection is n=16, some simple cell with different orientations are shown in Figures 4(h)–4(l). This illustration proved that a single simple cell just sensitive to the specific orientation, different orientation cell groups can better describe various redundant detailed features.
3.2. Experimental Results
In this section, the evaluation performance of the proposed model in the electrical insulator defect detection task is shown. Although several types of insulators are used in China, this paper only showed two main types of insulators: one type is the white ceramic insulator and the other is the dark red insulator.
For the reason of lacking public electrical insulator dataset, all of the electrical insulators used in our experiment are from our team’s captured dataset. The dataset includes 300 images, there are 34 images are good electrical insulator images, and others are defect electrical insulator images. Figure 5 shows some ceramic electrical insulators in our dataset. The first row includes good ceramic electrical insulators and some broken/damaged ones are shown in the second row.
There are three experiments are adopted in this paper, which include simulated defect detection, real defect detection, and robust feature detection. All of these experimental results can sufficiently show the efficiency of the proposed detection method.
(A) Simulated Defect Detection. Figure 6 shows defect detection result of electrical insulator with simulated crack. Figures 6(a1) and 6(b1) are real original ceramic electrical insulators. Figures 6(a2) and 6(b2) give the edge detection of original ceramic electrical insulators without crack. In Figures 6(a3) and 6(b3), there is a simulated crack located on the surface of real ceramic electrical insulator respectively. Figures 6(a4) and 6(b4) clearly show the crack and achieve accurate defect detection.
(B) Real Defect Detection. In Figure 7, four real damaged ceramic electrical insulators are used in our experiment. Figures 7(a1)–7(d1) are real original damaged insulator images from our dataset, we can see that there are obviously damaged on the surface of electrical insulators. Figures 7(a2)–7(d2) is the defect detection result of these damaged electrical insulators. In order to further provide the efficiency of the proposed method, the results of Figures 7(a2)–7(d2) are reversed operation to show accurate detection result, which is shown in Figures 7(a3)–7(d3) and marked in red rectangle. In addition, except for the subjective experimental result, objective evaluation result such as BRISQUE metric  is also provided here. Objective experimental evaluation value reflects that the insulator defect detection is more acceptable, it is easy to achieve further fault judgement.
3.3. Robust Detection
Experiments in Section 3.2 are electrical insulator captured in a good natural condition, the image is clear and clean. However, the electrical insulator has been exposed in outdoor electricity transmission line for a long time in normal situation, so the captured insulator image including noise. In order to verify the robust of our method, the experiment is added in this section. Figures 8(a1)–8(c1) represent the original clean circle image, the noisy circle image additive with standard deviation is 0.01, and the noisy circle image additive with standard deviation is 0.05. The damaged part of the insulator is marked by red rectangle. The type of noise is Gaussian noise. In Figures 8(d1)–8(f1), we can see the detection result of fore-referred circle images. Figures 8(a2) and 8(a3) are noisy version of the same electrical ceramic insulator (the standard deviation is 0.01 and 0.05, respectively). Experimental results shown in Figures 8(b2) and 8(b3) are the detection results; the damaged part of the noisy defect electrical insulator is accurately detected and it is very clear. For the purpose of further show the robust detection result, Figures 8(c2) and 8(c3) are operated reversely in order to show more clear damaged part of the electrical insulator.
The application of image processing and machine learning method is a popular development trend in future electricity power transmission line inspections. Inspired by human brain visual pathway LGN and V1 simple cell RF characteristic, this paper proposed an electrical insulator defect detection method combined computational model in area V1 of visual cortex, different feature orientation selectivity is achieved by combining operation of a collection of LGN cells with center-surround RF. Demonstration of different experimental results shown that the proposed method can achieve accurate electrical insulator defect detection, even complete robust detection of noisy insulators. Because the background of the insulator image acquired by aerial photography is complicated and there is more than one insulator type, defects are difficult to detect, so that our further work will focus on optimizing this method, identifying and locating the insulator defect by segmentation, and extending it to more wide application. The aim is greatly improving the efficiency of electricity power transmission line and easy to find defect or fault in the system, so as to provide completely and timely guarantee for electricity grid dispatching and electrical equipment maintenance.
Because the original datasets are captured by our research group, the datasets generated during and /or analysed during the current study are available from the corresponding author on reasonable request.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
This work is supported by National Natural Science Foundation of China (61663008), Natural Science Foundation of Hubei Province (2015CFC781, 2014CFB612), and Ph.D. Technology Program of Hubei University for Nationalities (MY2014B018).
W. Angelika, “Vision diagnostics of power transmission lines: approach to recognition of insulators,” in Improvements to Segmentation Method of Stained Iymphoma Tissue Sections Images, 440, 431 pages, Springer, 2016.View at: Google Scholar
T. Alshaketheep, K. Murakami, and N. Sekimura, “Effect of insulator environment on thermal ageing estimations for ethylene propylene rubber insulated nuclear power plant cables,” Journal of Nuclear Science and Technology, vol. 53, no. 9, pp. 1366–1370, 2016.View at: Publisher Site | Google Scholar
H. Zou, X. Lei, and S. Ma, “Motion feature extraction of random video sequence based on visual cortex V1 model,” Journal of Computer Applications, vol. 36, no. 6, pp. 1677–1681, 2016.View at: Google Scholar
D. Huang, C. Chen, G. Sun, and L. Zhao, “Recognition and diagnosis method of objective entropy weight for power transformer fault,” Dianli Xitong Zidonghua/Automation of Electric Power Systems, vol. 41, no. 12, pp. 206–211, 2017.View at: Publisher Site | Google Scholar
D. Zuo, H. Hu, R. Qian, and Z. Liu, “An insulator defect detection algorithm based on computer vision,” in Proceedings of the 2017 IEEE International Conference on Information and Automation (ICIA), pp. 361–365, China, July 2017.View at: Publisher Site | Google Scholar
N. N. Qaddoumi, A. H. El-Hag, and Y. Saker, “Outdoor insulators testing using artificial neural network-based near-field microwave technique,” IEEE Transactions on Instrumentation and Measurement, vol. 63, no. 2, pp. 260–266, 2014.View at: Publisher Site | Google Scholar
H. Armando, Shinohara, and M. Danilo, “Defects detection in electrical insulators and breaker for high voltage by low cost computed radiography systems,” in Proceedings of the International Conference of symposium on digital industrial radiology and computed tomography, vol. 6, pp. 25–27, France, 2007.View at: Google Scholar
F. Chen and J. Yao, “Extraction method of single insulator disk in infrared image of insulator string,” Power System Technology, vol. 34, no. 5, pp. 220–224, 2010.View at: Google Scholar
D.-R. Huang, C.-S. Chen, G.-X. Sun, L. Zhao, and B. Mi, “Linear discriminant analysis and back propagation neural network cooperative diagnosis method for multiple faults of complex equipment bearings,” Acta Armamentarii, vol. 38, no. 8, pp. 1649–1657, 2017.View at: Publisher Site | Google Scholar
H. Darong, T. Jianping, and Z. Ling, “A fault diagnosis method of power systems based on gray system theory,” Mathematical Problems in Engineering, vol. 2015, Article ID 971257, 11 pages, 2015.View at: Publisher Site | Google Scholar
X. Mei, T. Lu, X. Wu, and B. Zhang, “Insulator surface dirt image detection technology based on improved watershed algorithm,” in Proceedings of the 2012 Asia-Pacific Power and Energy Engineering Conference, APPEEC 2012, March 2012.View at: Google Scholar
M. Oberweger, A. Wendel, and H. Bischof, “Visual recognition and fault detection for power line insulators,” in Proceedings of the 19th Computer Vision Winter Workshop, Krtiny, vol. 2, pp. 3–5, Czech Republic, 2014.View at: Google Scholar
S. Liao and J. An, “A robust insulator detection algorithm based on local features and spatial orders for aerial images,” IEEE Geoscience and Remote Sensing Letters, vol. 12, no. 5, pp. 963–967, 2015.View at: Publisher Site | Google Scholar
L. Junfeng, L. Min, and W. Qinruo, “A novel insulator detection method for aerial images,” in Proceedings of the the 9th International Conference, vol. 2, pp. 18–21, Sydney, Australia, Feburary 2017.View at: Publisher Site | Google Scholar
J. Zhao, X. Liu, and J. Sun, “Detecting insulators in the image of overhead transmission line,” in Proceedings of the of International Conference on Intelligent Computing, pp. 442–450, 2012.View at: Google Scholar
H. Darong, K. Lanyan, C. Xiaoyan, Z. Ling, and M. Bo, “Fault diagnosis for the motor drive system of urban transit based on improved Hidden Markov Model,” Microelectronics Reliability, vol. 82, pp. 179–189, 2018.View at: Publisher Site | Google Scholar
M. Bartolucci and A. T. Smith, “Attentional modulation in visual cortex is modified during perceptual learning,” Neuropsychologia, vol. 49, no. 14, pp. 3898–3907, 2011.View at: Publisher Site | Google Scholar
S. Grossberg, E. Mingolla, and W. D. Ross, “Visual brain and visual perception: how does the cortex do perceptual grouping?” Trends in Neurosciences, vol. 20, no. 3, pp. 106–111, 1997.View at: Publisher Site | Google Scholar
J. Carl and B. Bassi, “Parallel processing in the human visual system,” in New Methods of Sensory Visual Testing, 59, 53 pages, Springer, New York, NY, USA, 1989.View at: Google Scholar
C. Grigorescu, N. Petkov, and M. A. Westenberg, “Contour detection based on nonclassical receptive field inhibition,” IEEE Transactions on Image Processing, vol. 12, no. 7, pp. 729–739, 2003.View at: Publisher Site | Google Scholar
G. Azzopardi and N. Petkov, “A CORF computational model of a simple cell that relies on LGN input outperforms the Gabor function model,” Biological Cybernetics, vol. 106, no. 3, pp. 177–189, 2012.View at: Publisher Site | Google Scholar
C. Grigorescu, N. Petkov, and M. A. Westenberg, “Contour and boundary detection improved by surround suppression of texture edges,” Image and Vision Computing, vol. 22, no. 8, pp. 609–622, 2004.View at: Publisher Site | Google Scholar
Q. Sang, B. Cai, and H. Chen, “Contour detection improved by contex-adaptive surround suppression,” PLoS ONE, vol. 12, no. 7, Article ID e0181792, 2017.View at: Google Scholar
D. G. Albrecht, R. L. De Valois, and L. G. Thorell, “Visual cortical neurons: are bars or gratings the optimal stimuli?” Science, vol. 207, no. 4426, pp. 88–90, 1980.View at: Publisher Site | Google Scholar
C. Lv, Y. Xu, and S. Li, “An edge detection model by brain-inspired,” Computer Engineering and Application, vol. 53, no. 24, pp. 142–146, 2017.View at: Google Scholar
T. Sun, X. Zhu, J. Pan, J. Wen, and F. Meng, “No-reference image quality assessment in spatial domain,” Genetic and Evolutionary Computing, vol. 21, no. 12, pp. 4695–4708, 2012.View at: Google Scholar