Method and system for visual monitoring and adaptive identification of opening and closing states of contact network switches

Through the combination of deep learning and linear detection algorithm, the problem of identification of contact network switches under different environmental conditions is solved, and the adaptive and accurate state recognition of contact network switches is realized. It has a wide range of application and can accurately identify the opening and closing state and inaccurate position. state improves the accuracy and adaptability of device status recognition.

CN114155427BActive Publication Date: 2024-11-15CHENG DOU JIAO DA GUANG MANG SHI YE YOU XIAN GONG SI +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111555826.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-17
Publication Date
2024-11-15
Estimated Expiration
2041-12-17

AI Technical Summary

Technical Problem

The prior art is difficult to accurately identify the split and combine state of the contact network switch under the influence of different weather, light, foreign objects and installation locations, resulting in low recognition accuracy and high difficulty in affecting the safety and maintenance of power supply equipment.

Method used

The object detection model and image segmentation technology of deep learning combined with a variety of linear detection algorithms can be used to obtain the original image of the contact network switch, locate and segment the switch, extract the edges, calculate the opening and closing angles, and use the closing limit point and static contacts. The guide slope determines the critical angle threshold to achieve adaptive state recognition.

Benefits of technology

It realizes accurate positioning and status recognition of contact network switches under different environmental conditions. It has a wide range of application, reduces the need for labeling information, and can accurately identify the opening and closing state and inadequate state, improving the accuracy and adaptability of equipment status recognition. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114155427B_ABST
    Figure CN114155427B_ABST
Patent Text Reader

Abstract

The present invention provides a method for adaptive opening and closing state recognition of a contact network switch visual monitoring, comprising the following steps: obtaining an original image containing a contact network switch; locating the contact network switch in the original image based on a deep learning target detection model, and obtaining a contact network switch foreground image based on a deep learning image segmentation technology; performing edge extraction of the contact network switch foreground image by a plurality of straight line detection algorithms, and performing straight line fitting on a plurality of detection results to obtain a single long straight line detected; calculating the contact network switch opening and closing angle based on the obtained long straight line; calling configuration information, obtaining the critical angle threshold corresponding to the opening and closing angle, and performing contact network switch state recognition, wherein the critical angle threshold is determined by the closing limit point marking screw and the guide slope on the upper part of the static contact. The present invention adopts the contact network switch positioning, segmentation and edge extraction technology to realize adaptive and accurate state recognition, has a wide range of applications, and requires less information to be labeled.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of state recognition of power transmission and transformation equipment, and in particular to a method and system for self-adaptive opening and closing state recognition of visual monitoring of overhead contact network switches. Background Art

[0002] The overhead contact switch is an important part of the high-speed railway traction power supply circuit, and is also a high-voltage electrical equipment that is used more frequently and widely in the power system. Through the remote operation of the disconnector, the overhead contact power supply arm can be quickly shut down, cross-zone power supply can be supplied, and the scope of the fault can be reduced. It is one of the indispensable key equipment for the maintenance, emergency repair and EMU rescue of high-speed railway traction power supply equipment. The overhead contact switch visualization is achieved by installing a high-definition camera at the overhead contact switch to open up the interaction between the data monitoring system and the video monitoring system, and the visualization and separation and closing status of the overhead contact switch can be realized through intelligent image recognition technology, and early warning and management can be realized. The automatic recognition of the state of the overhead contact disconnector through intelligent recognition technology and the active push of the operating switch equipment state can effectively reduce the labor intensity of the operators and improve the work quality, which is of great significance to the efficient operation of the power supply equipment.

[0003] Because the contact network switch is generally deployed outdoors, its mechanical moving parts are directly exposed to the air, and are greatly affected by wind, sun, rain and other environmental factors. They are prone to rust and other problems, which can easily lead to abnormal contact network switches, which has a serious impact on the safety of power supply and creates great difficulties for the operation and maintenance of equipment. Since the contact network switch is in an outdoor environment, the collected video images are often affected by the following factors, such as weather, lighting, uncertain background, foreground effects (clouds, buildings, lines, etc.), foreign objects, and installation points. There are many and complex influencing factors, so it is a technical challenge to achieve adaptive and accurate state recognition of the contact network switch. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for visually monitoring and adaptively identifying the opening and closing states of contact network switches, aiming to solve the problems pointed out in the background technology. The method can realize accurate positioning of contact network switches under the influence of different weather, light, foreign objects and installation positions.

[0005] The embodiment of the present invention is implemented by the following technical solution: a method for visually monitoring the adaptive opening and closing state recognition of a contact network switch comprises the following steps:

[0006] Acquire the original image containing the overhead contact switch;

[0007] Based on the deep learning target detection model, the overhead contact network switch is located in the original image, and based on the deep learning image segmentation technology, the overhead contact network switch foreground image is obtained;

[0008] The edge of the foreground image of the contact network switch is extracted by using a variety of straight line detection algorithms, and a straight line fitting is performed on a variety of detection results to obtain a single long straight line.

[0009] Calculate the opening and closing angle of the overhead line switch based on the obtained long straight line;

[0010] The configuration information is called to obtain the critical angle threshold corresponding to the opening and closing angle to identify the contact network switch state. The critical angle threshold is determined by the closing limit point marking screw and the guide slope on the upper part of the static contact.

[0011] Furthermore, the deep learning target detection model adopts the YOLOv4 model.

[0012] Furthermore, the deep learning image segmentation technology adopts Mask R-CNN or Unet.

[0013] Furthermore, the plurality of line detection algorithms at least include hough line detection, LSD line detection and EDLines line detection.

[0014] Furthermore, the calculation of the contact network switch opening and closing angle based on the obtained long straight line also includes:

[0015] Obtain the second longest straight line: identify the closing limit point through the positioning network model, segment the limit position based on the segmentation algorithm, and then calculate the limit point position through the centroid method. The line connecting the limit point position and one end point of the long straight line is used as the second longest straight line;

[0016] The opening and closing angles of the overhead contact network switch are calculated using the long straight line and the second long straight line.

[0017] The present invention also provides a contact network switch visual monitoring adaptive opening and closing state recognition system, which is applied to the above-mentioned method, including:

[0018] An image acquisition unit, used for acquiring an original image including the overhead contact network switch;

[0019] A positioning and segmentation unit is used to locate the contact network switch in the original image based on a deep learning target detection model, and obtain a contact network switch foreground image based on a deep learning image segmentation technology;

[0020] An edge extraction unit is used to extract the edge of the foreground image of the overhead line switch by using a variety of straight line detection algorithms, and to perform straight line fitting on a variety of detection results to obtain a single long straight line;

[0021] An angle calculation unit, used for calculating the opening and closing angle of the overhead line switch based on the obtained long straight line;

[0022] The switch state identification unit is used to call the configuration information, obtain the critical angle threshold corresponding to the opening and closing angle, and identify the contact network switch state. The critical angle threshold is determined by the closing limit point marking screw and the guide slope on the upper part of the static contact.

[0023] Furthermore, the deep learning target detection model adopts the YOLOv4 model.

[0024] Furthermore, the deep learning image segmentation technology adopts Mask R-CNN or Unet.

[0025] Furthermore, the plurality of line detection algorithms at least include hough line detection, LSD line detection and EDLines line detection.

[0026] Furthermore, the angle calculation unit includes a second long straight line acquisition module, which is used to identify the closing limit point through the positioning network model, implement the segmentation of the limit position based on the segmentation algorithm, and then calculate the limit point position through the centroid method, and use the line connecting the limit point position and one end point of the long straight line as the second long straight line.

[0027] The technical solution of the embodiment of the present invention has at least the following advantages and beneficial effects: the present invention adopts the contact network switch positioning, segmentation and edge extraction technology to realize adaptive and precise state recognition; it can realize continuous state recognition of the contact network switch under the influence of different weather, light, foreign objects, and installation positions, with a wide range of applications and less information required to be marked; combined with the closing limit point to calculate the critical angle threshold judgment standard, it can adapt to the accurate recognition of the switch opening and closing state and the out-of-position state. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A schematic flow chart of a method for identifying an open / closed state provided in Example 1 of the present invention;

[0029] Figure 2 This is a recognition flow chart provided in Example 1 of the present invention;

[0030] Figure 3 A schematic diagram of an open state of a contact network switch provided in Embodiment 1 of the present invention;

[0031] Figure 4 A schematic diagram of a contact network switch in a closed state provided in Embodiment 1 of the present invention;

[0032] Figure 5 A schematic diagram of a switch-on interface provided in Embodiment 1 of the present invention;

[0033] Figure 6 A schematic diagram of an original image provided in Example 1 of the present invention;

[0034] Figure 7 A schematic diagram of the positioning effect provided by Embodiment 1 of the present invention;

[0035] Figure 8 A schematic diagram of the segmentation effect provided in Example 1 of the present invention;

[0036] Fig. 9 This is a schematic diagram of the edge extraction effect provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0038] Example 1

[0039] The applicant has found through research that existing technologies such as CN109712118A propose the use of Mask R-CNN segmentation and classification to realize the state recognition of the power station disconnector. This method simply divides the disconnector state into three states: open, closed, and intermediate. This is inconsistent with the actual state being a range rather than three definite states. In addition, it does not take into account the impact of camera installation on the algorithm, resulting in adaptive deviations in the algorithm and the inability to give an accurate disconnector state.

[0040] CN112906576A proposes to use NCC (normalized cross correlation) algorithm to realize the recognition of the state of disconnector equipment. This algorithm does not and can only provide the recognition of a specific state, but is not suitable for continuous state recognition. Moreover, the algorithm does not consider the influence of factors such as camera input, foreign objects, and background on the algorithm.

[0041] Based on this, an embodiment of the present invention provides a method for visually monitoring an overhead contact network switch and adaptively identifying the opening and closing states, which can effectively solve the above-mentioned problems. The method includes the following steps:

[0042] Get the original image containing the contact network switch, the original image reference Figure 6 shown.

[0043] Furthermore, a deep learning-based target detection model, such as the YOLOv4 model, is used to locate the contact network switch in the original image. The positioning effect is shown in Figure 2. Figure 7 As shown; based on deep learning image segmentation technology, such as Mask R-CNN or Unet, the foreground image of the contact network switch is obtained. The segmentation effect is referenced Figure 8shown.

[0044] Furthermore, the edge extraction of the foreground image of the contact network switch is performed through a variety of line detection algorithms, such as hough line detection, LSD line detection and EDLines line detection. The edge extraction effect is referenced Figure 8 As shown, a straight line is fitted to multiple detection results to obtain a long straight line of a single detection.

[0045] Calculating the contact network switch opening and closing angle based on the obtained long straight line, in one embodiment of the present invention, the calculating the contact network switch opening and closing angle based on the obtained long straight line further includes:

[0046] The second long straight line is obtained by using the rotation center and the switch limit point, or by the following method: identifying the closing limit point by positioning the network model, segmenting the limit position based on the segmentation algorithm, and then calculating the limit point position by the centroid method, and taking the line connecting the limit point position and one end point of the long straight line as the second long straight line; calculating the contact network switch opening and closing angle by the angle between the long straight line and the second long straight line.

[0047] Further, the configuration information is called to obtain the critical angle threshold corresponding to the opening and closing angle, and the contact network switch state is identified. The critical angle threshold is determined by the closing limit point mark screw and the guide slope on the upper part of the static contact. The specific structure is referenced Figure 5 As shown, Figure 5 In the static contact, there is a guiding slope on the upper part. From the upper edge of the main switch blade to below the slope (that is, when there is 100% contact area), to the time when the pressure blade shoulder axis touches the closing limit screw, the area of ​​about 10mm belongs to the closing area, which can meet the needs of conducting the rated current. Therefore, this area can be regarded as the closing position, and this indicator is limited to the judgment basis when the switch is in working state.

[0048] It should be noted that, in the embodiment of the present invention, the configuration information also includes the opening and closing angle limit position, the opening and closing angle reliable position, etc., so as to overcome the influence of the installation point on the state recognition; in addition, in order to calculate the accurate contact network switch state, the annotation information includes: the opening state angle and the pixel resolution. The critical point position of the non-position is calculated by the pixel resolution and the closing effective area (10mm), and then the critical angle threshold can be calculated, thereby the current contact network switch state can be identified, such as: the opening state (reference Figure 3 ), closed state (reference Figure 4 ) and out-of-place status for accurate identification.

[0049] The embodiment of the present invention also provides a contact network switch visual monitoring adaptive opening and closing state recognition system, which is applied to the above-mentioned method, including:

[0050] An image acquisition unit, used for acquiring an original image including the overhead contact network switch;

[0051] A positioning and segmentation unit is used to locate the contact network switch in the original image based on a deep learning target detection model, and obtain a contact network switch foreground image based on a deep learning image segmentation technology;

[0052] An edge extraction unit is used to extract the edge of the foreground image of the overhead line switch by using a variety of straight line detection algorithms, and to perform straight line fitting on a variety of detection results to obtain a single long straight line;

[0053] An angle calculation unit, used for calculating the opening and closing angle of the overhead line switch based on the obtained long straight line;

[0054] The switch state identification unit is used to call the configuration information, obtain the critical angle threshold corresponding to the opening and closing angle, and identify the contact network switch state. The critical angle threshold is determined by the closing limit point marking screw and the guide slope on the upper part of the static contact.

[0055] Furthermore, the deep learning target detection model adopts the YOLOv4 model.

[0056] Furthermore, the deep learning image segmentation technology adopts Mask R-CNN or Unet.

[0057] Furthermore, the plurality of line detection algorithms at least include hough line detection, LSD line detection and EDLines line detection.

[0058] Furthermore, the angle calculation unit includes a second long straight line acquisition module, which is used to identify the closing limit point through the positioning network model, implement the segmentation of the limit position based on the segmentation algorithm, and then calculate the limit point position through the centroid method, and use the line connecting the limit point position and one end point of the long straight line as the second long straight line.

[0059] In summary, the present invention adopts contact network switch positioning, segmentation and edge extraction technology to achieve adaptive and precise state recognition; it can realize continuous state recognition of contact network switches under the influence of different weather, light, foreign objects, and installation positions, has a wide range of applications, and requires less information to be labeled; combined with the closing limit point to calculate the critical angle threshold judgment standard, it can adapt to the accurate recognition of the switch opening and closing state and the out-of-position state.

[0060] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Adaptive opening and closing state recognition method for visual monitoring of contact network switches, It is characterized in that The steps include: Acquire the original image containing the overhead contact switch; Based on the deep learning target detection model, the overhead contact network switch is located in the original image, and based on the deep learning image segmentation technology, the overhead contact network switch foreground image is obtained; The edge of the foreground image of the contact network switch is extracted by using a variety of straight line detection algorithms, and a straight line fitting is performed on a variety of detection results to obtain a single long straight line. Calculate the opening and closing angle of the overhead line switch based on the obtained long straight line; Call the configuration information to obtain the critical angle threshold corresponding to the opening and closing angle, and identify the contact network switch state. The critical angle threshold is determined by the closing limit point mark screw, the guide slope on the upper part of the static contact, and the critical point position of not in place. The critical point position of not in place is obtained according to the effective closing area. The effective closing area is the area from the upper edge of the main switch knife to below the guide slope, to the pressure knife shoulder shaft touching the closing limit point mark screw.

2. The method for self-adaptive opening and closing state recognition of catenary switch visual monitoring according to claim 1, It is characterized in that The deep learning target detection model adopts the YOLOv4 model.

3. The method for self-adaptive opening and closing state recognition of catenary switch visual monitoring according to claim 1, It is characterized in that The deep learning image segmentation technology adopts Mask R-CNN or Unet.

4. The method for self-adaptive opening and closing state recognition of catenary switch visual monitoring according to claim 1, It is characterized in that The multiple line detection algorithms at least include hough line detection, LSD line detection and EDLines line detection.

5. The method for self-adaptive opening and closing state recognition of catenary switch visual monitoring according to claim 1, It is characterized in that The calculation of the contact network switch opening and closing angle based on the obtained long straight line also includes: Obtain the second longest straight line: identify the closing limit point through the positioning network model, segment the limit position based on the segmentation algorithm, and then calculate the limit point position through the centroid method. The line connecting the limit point position and one end point of the long straight line is used as the second longest straight line; The opening and closing angles of the overhead contact network switch are calculated using the long straight line and the second long straight line.

6. A system for visually monitoring and adaptively identifying the opening and closing states of overhead contact switches, applied to the method according to any one of claims 1 to 5, It is characterized in that include: An image acquisition unit, used for acquiring an original image including the overhead contact network switch; A positioning and segmentation unit is used to locate the contact network switch in the original image based on a deep learning target detection model, and obtain a contact network switch foreground image based on a deep learning image segmentation technology; An edge extraction unit is used to extract the edge of the foreground image of the overhead line switch by using a variety of straight line detection algorithms, and to perform straight line fitting on a variety of detection results to obtain a single long straight line; An angle calculation unit, used for calculating the opening and closing angle of the overhead line switch based on the obtained long straight line; The switch state identification unit is used to call the configuration information, obtain the critical angle threshold corresponding to the opening and closing angle, and identify the contact network switch state. The critical angle threshold is determined by the closing limit point mark screw, the guide slope on the upper part of the static contact, and the critical point position of the non-position. The critical point position of the non-position is obtained based on the effective closing area. The effective closing area is the area from the upper edge of the main switch knife to below the guide slope, to the pressure knife shoulder shaft touching the closing limit point mark screw.

7. The contact network switch visual monitoring and adaptive opening and closing state recognition system according to claim 6, It is characterized in that The deep learning target detection model adopts the YOLOv4 model.

8. The contact network switch visual monitoring and adaptive opening and closing state recognition system according to claim 6, It is characterized in that The deep learning image segmentation technology adopts Mask R-CNN or Unet.

9. The contact network switch visual monitoring and adaptive opening and closing state recognition system according to claim 6, It is characterized in that The multiple line detection algorithms at least include hough line detection, LSD line detection and EDLines line detection.

10. The contact network switch visual monitoring and adaptive opening and closing state recognition system according to claim 6, It is characterized in that The angle calculation unit includes a second long straight line acquisition module, which is used to identify the closing limit point through the positioning network model, implement the segmentation of the limit position based on the segmentation algorithm, and then calculate the limit point position through the centroid method, and use the line connecting the limit point position and one end point of the long straight line as the second long straight line.