A multi-dimensional joint determination disconnector opening and closing state monitoring method

Abstract

The application discloses a kind of multi-dimension combined determination isolation switch opening and closing state monitoring method, belong to electric power system technical field, first by YOLO26 model accurate positioning ROI area effect excludes background sundry interference, and then with reference figure Similarity calculation is completed fast preliminary screening, and the obvious intermediate state is effectively identified;Subsequently, for suspected to be in place state, further adopt Shi-Tomasi corner detection and Euclidean distance calculation to carry out pixel-level fine verification.The layered strategy of the application effectively avoids the misjudgment of single algorithm caused by background interference or local occlusion, greatly improves the confidence of state recognition, and uses the relative Euclidean distance between corner points instead of absolute coordinates for judgment, which can reduce the measurement error caused by slight camera shaking to a certain extent, ensuring the monitoring stability under long-term operation and reducing the misjudgment of "false split and combination".

Description

Technical Field

[0001] This invention belongs to the field of power system technology, specifically relating to a multi-dimensional joint determination method for monitoring the opening and closing status of disconnecting switches. Background Technology

[0002] Disconnect switches are widely used and distributed electrical equipment in power systems, primarily used to create a clear disconnection point during maintenance to ensure personnel safety. Current monitoring methods for the open / closed status of disconnect switches, such as traditional image processing, are extremely sensitive to changes in lighting (e.g., backlighting, shadows) and weather conditions (rain, fog, snow), exhibiting poor anti-interference capabilities. Manual inspections are inefficient, lack real-time performance, and pose personal safety risks for close-range observation of high-voltage equipment. Therefore, developing a method for monitoring the open / closed status of disconnect switches that is highly resistant to environmental interference and can accurately identify minute displacements is an urgent need for current smart grid operation and maintenance. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a multi-dimensional joint determination method for monitoring the opening and closing status of disconnect switches. This method uses the YOLO26 model to accurately locate the ROI region, effectively eliminating background clutter interference. Then, similarity calculation with a benchmark image is performed for rapid initial screening, effectively identifying obvious intermediate states. Subsequently, for suspected in-position states, Shi-Tomasi corner detection and Euclidean distance calculation are used for pixel-level fine verification. This layered strategy effectively avoids misjudgments by a single algorithm due to background interference or partial occlusion, significantly improving the confidence level of state recognition. Furthermore, using relative Euclidean distance between corner points instead of absolute coordinates for determination can reduce measurement errors caused by slight camera shake to a certain extent, ensuring monitoring stability during long-term operation and reducing false "opening / closing" misjudgments.

[0004] The technical solution of the present invention is as follows: A multi-dimensional joint determination method for monitoring the opening and closing status of a disconnecting switch, comprising the following steps:

[0005] S1. Train the YOLO26 object detection model;

[0006] S2. Obtain the baseline image;

[0007] S3. Train a Siamese neural network model for similarity detection;

[0008] S4. Target localization of the image to be tested;

[0009] S5. Similarity-based preliminary screening and state prediction: The similarity calculation with the benchmark image is used to complete the rapid preliminary screening and effectively identify obvious intermediate states.

[0010] S6. Refined state determination based on Shi-Tomasi corner detection: For the suspected in-place state in step S5, Shi-Tomasi corner detection and Euclidean distance calculation are used for pixel-level fine verification.

[0011] Preferably, in step S1, the YOLO26 target detection model is used to detect the ROI area and contact area of ​​the disconnecting switch.

[0012] Preferably, the specific steps of step S2 are as follows: S21, firstly, acquire an image of the disconnector switch, and process the image using the YOLO26 target detection model; S22, accurately locate and crop out the ROI region and contact region containing the main body of the disconnector switch, and generate a baseline image of the disconnector switch open state and a baseline image of the disconnector switch closed state respectively, as reference templates for subsequent comparison.

[0013] Preferably, the specific steps of step S3 are as follows:

[0014] S31, YOLO26 Object Detection: Using the YOLO26 object detection model trained in step S1, inference is performed on the input image, and the bounding box coordinates of the isolation switch ROI are output. and the bounding box coordinates of the contact point position ;

[0015] S32, ROI Cropping: The ROI region of the disconnector switch is cropped according to the bounding box coordinates and uniformly scaled to a fixed resolution of 224×224. This eliminates the scale differences caused by different camera installation distances or focal length changes. Then, a similarity calculation model is trained based on the cropped 224×224 image.

[0016] Preferably, the specific steps of step S4 are as follows: S41, acquire the on-site image of the disconnector switch that needs to be monitored in real time; S42, use the YOLO26 target detection model trained in step S1 to perform detection; S43, extract the ROI region of the disconnector switch at the current moment to ensure the consistency of the image to be tested with the reference image in terms of spatial position and scale, and eliminate background interference.

[0017] Preferably, the specific steps of step S5 are as follows: S51, calculate the similarity between the ROI area of ​​the disconnecting switch to be tested and the two reference maps in step S2; S52, calculate the similarity between the ROI area of ​​the disconnecting switch to be tested and the reference map of the open state (Open-Sim) and the reference map of the closed state (Close-Sim) using the similarity model in step S3, and make a preliminary judgment by setting a similarity threshold; S53, if the similarity of both is lower than the threshold, the device is directly determined to be in the "intermediate state" of the operation process; if either similarity is higher than the threshold, the device enters the fine judgment stage, and the similarity threshold is determined by a large amount of data testing and statistics.

[0018] Preferably, the specific steps of step S6 are as follows: S61, for the test image that has passed the coarse screening, the contact area is extracted, and the Shi-Tomasi algorithm is used to extract the corner points with high stability on the corresponding reference image; S62, the Lucas-Kanade sparse optical flow algorithm is used to calculate the corresponding corner points on the test image, and the average value of the Euclidean distance between the corner points of the test image and the corner points of the corresponding reference image is calculated; S63, by comparing the average value of the Euclidean distance with the distance threshold, the actual opening and closing state of the disconnector switch in the test image is finally accurately determined. The distance threshold is also determined by a large amount of data testing and statistics.

[0019] Preferably, the average Euclidean distance between the corner point of the image to be tested and the corresponding corner point of the reference image is calculated by equation (1):

[0020] (1)

[0021] The set of valid corner points extracted from the benchmark image is as follows: The set of corresponding corner points of the image to be measured obtained by optical flow method is , For the reference figure, the first There are 1 valid coordinate point, with coordinates as follows: , For the first part of the graph to be tested There are 1 valid coordinate point, with coordinates as follows: , This represents the total number of valid corner points that were successfully matched.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] 1. Advantages at the functional level: (1) Strong anti-interference ability and high monitoring accuracy: This invention abandons the single-dimensional judgment logic and proposes a two-level judgment mechanism of "coarse screening + fine judgment". In actual testing, it effectively avoids false alarms caused by background clutter or viewing angle deviation; (2) Adaptive computing power allocation and fast response speed: Through the pre-set similarity threshold judgment logic, the system can automatically identify whether the device is in the "intermediate state" of the operation process. For the intermediate state, the result is directly output. Only for the suspected position state, the high-precision corner detection process is triggered. While ensuring high precision, the average computing load of the system is significantly reduced.

[0024] 2. Advantages at the structural level: (1) Structural design based on multi-level filtering: As shown in Figure 1, a similarity threshold determination branch structure is added to the overall process structure of this invention. The advantage of this structural design is that it constructs an efficient filtering barrier. When the similarity between the image to be tested and the reference image is lower than the threshold, the system directly determines it as an "intermediate state" and terminates the subsequent operation. This structure effectively avoids invalid high-time-consuming operations, making the system more adaptable to the computing power limitations of the equipment on the edge side of the substation; (2) Corner verification structure: This invention adopts a combination structure of "Shi-Tomasi corner detection + Euclidean distance calculation" in the fine judgment stage. The advantage of this structure is that it uses the Euclidean distance between corners instead of absolute coordinates for judgment, which reduces the measurement error caused by slight camera shaking to a certain extent. At the same time, compared with simple pixel difference comparison, this structure is more sensitive to the small displacement of the contact texture, which reduces the false judgment of "sham separation and combination" to a certain extent. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this invention to illustrate the operations performed by the system according to the embodiments of the present invention. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0026] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0027] Figure 2 This is a schematic diagram of the ROI area (green box) and contact area (blue box) of the disconnector switch in this invention;

[0028] Figure 3 This is a schematic diagram of the closed state of the disconnector switch in this invention;

[0029] Figure 4 This is a schematic diagram of the ROI of the disconnector switch in the closed state in this invention;

[0030] Figure 5 This is a schematic diagram of the contact corner detection of the disconnector switch in the closed state in this invention;

[0031] Figure 6 This is a schematic diagram of the disconnector switch in the open state in this invention;

[0032] Figure 7 This is a schematic diagram of the ROI of the disconnector switch in the open state in this invention;

[0033] Figure 8 This is a schematic diagram of the contact corner detection in the open state of the disconnector switch in this invention;

[0034] Figure 9 This is a schematic diagram of the corner points of the image to be tested in this invention. Detailed Implementation

[0035] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on the specific implementation methods, structures, features and effects of the present invention.

[0036] A multi-dimensional joint determination method for monitoring the opening and closing status of disconnect switches includes the following main parts: YOLO26 target detection model training, acquisition of a baseline image, training of a similar Siamese neural network model, target localization in the image to be tested, coarse screening and state prediction based on similarity, and refined state determination based on Shi-Tomasi corner detection. The overall process is as follows: Figure 1 As shown.

[0037] The specific steps of each process in this invention are as follows:

[0038] (1) Training the YOLO26 object detection model

[0039] The target detection model is used to detect the ROI region and contact region of the disconnecting switch. The model detection results are as follows: Figure 2 As shown.

[0040] (2) Obtain the baseline map

[0041] First, acquire an image of the disconnector switch, such as... Figure 3 and Figure 6 As shown, the YOLO26 object detection model is used to process the image, accurately locating and cropping the ROI region containing the main body of the disconnector switch and the contact region, generating "disconnector switch open state baseline image" and "disconnector switch closed state baseline image" respectively. Figure 4 and Figure 7 As shown, this serves as a reference template for subsequent comparisons.

[0042] (3) Training a Siamese neural network model for similarity detection

[0043] Because similarity calculation is extremely sensitive to location, it is essential to ensure that the image to be tested is strictly aligned spatially with the reference image.

[0044] YOLO26 Object Detection: Using the YOLO26 object detection model trained in (1), inference is performed on the input image, and the bounding box coordinates of the isolation switch ROI are output. and the bounding box coordinates of the contact point position .

[0045] ROI cropping: The ROI region of the disconnector is cropped according to the bounding box coordinates and uniformly scaled to a fixed resolution of 224×224. This step eliminates the scale difference caused by different camera installation distances or focal length changes. Then, a similarity calculation model is trained based on the cropped 224×224 image.

[0046] (4) Target localization of the image to be tested

[0047] To acquire real-time images of the disconnecting switches, the YOLO26 target detection model is used to detect and extract the ROI region of the disconnecting switches at the current moment, ensuring the consistency of the image under test with the reference image in terms of spatial location and scale, and eliminating background interference.

[0048] (5) Similarity-based coarse screening and state prediction

[0049] The similarity of the ROI region of the disconnecting switch to be tested with the two reference maps mentioned above is calculated. The similarity model of (3) is used to calculate the similarity between the open state reference map (Open-Sim) and the closed state reference map (Close-Sim). A preliminary judgment is made by setting a similarity threshold: if the similarity of both is lower than the threshold, the device is directly judged to be in the "intermediate state" of the operation process; if either similarity is higher than the threshold, the device enters the fine judgment stage. The similarity threshold is determined by a large amount of data testing and statistics.

[0050] (6) Refined state determination based on Shi-Tomasi corner detection

[0051] For the test image that passes the coarse screening, the contact area is extracted, and the Shi-Tomasi algorithm is used to extract the corner points with high stability on the corresponding reference image, such as... Figure 5 and Figure 8 As shown, the Lucas-Kanade sparse optical flow algorithm is used to calculate the corresponding corner points on the image under test, such as... Figure 9As shown, the average Euclidean distance between the corner point of the image under test and the corresponding corner point of the reference image is calculated according to Formula 1. The actual opening and closing state of the disconnector in the image under test is finally accurately determined by comparing the average Euclidean distance with the distance threshold. The distance threshold is also determined by a large amount of data testing and statistics.

[0052] Formula 1

[0053] The set of valid corner points extracted from the benchmark image is as follows: The set of corresponding corner points of the image to be measured obtained by optical flow method is , For the reference figure, the first There are 1 valid coordinate point, with coordinates as follows: , For the first part of the graph to be tested There are 1 valid coordinate point, with coordinates as follows: , This represents the total number of valid corner points that were successfully matched.

[0054] The optimal use case of this invention is for determining the open / closed status of disconnect switches at fixed locations during intelligent substation inspections.

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-dimensional joint determination method for monitoring the opening and closing status of a disconnecting switch, characterized in that, Includes the following steps: S1. Train the YOLO26 object detection model; S2. Obtain the baseline image; S3. Train a Siamese neural network model for similarity detection; S4. Target localization of the image to be tested; S5. Similarity-based preliminary screening and state prediction: The similarity calculation with the benchmark image is used to complete the rapid preliminary screening and effectively identify obvious intermediate states. S6. Refined state determination based on Shi-Tomasi corner detection: For the suspected in-place state in step S5, Shi-Tomasi corner detection and Euclidean distance calculation are used for pixel-level fine verification.

2. The multi-dimensional joint determination method for monitoring the opening and closing status of a disconnecting switch as described in claim 1, characterized in that, In step S1, the YOLO26 target detection model is used to detect the ROI area and contact area of ​​the disconnecting switch.

3. The multi-dimensional joint determination method for monitoring the opening and closing status of a disconnecting switch as described in claim 2, characterized in that, The specific steps of step S2 are as follows: S21. First, acquire an image of the disconnector switch and process the image using the YOLO26 target detection model; S22. Accurately locate and crop out the ROI region and contact region containing the main body of the disconnector switch, and generate a baseline image of the disconnector switch open state and a baseline image of the disconnector switch closed state, respectively, as reference templates for subsequent comparison.

4. The multi-dimensional joint determination method for monitoring the opening and closing status of a disconnecting switch as described in claim 1, characterized in that, The specific steps of step S3 are as follows: S31, YOLO26 Object Detection: Using the YOLO26 object detection model trained in step S1, inference is performed on the input image, and the bounding box coordinates of the isolation switch ROI are output. and the bounding box coordinates of the contact point position ; S32, ROI Cropping: The ROI region of the disconnector switch is cropped according to the bounding box coordinates and uniformly scaled to a fixed resolution of 224×224. This eliminates the scale differences caused by different camera installation distances or focal length changes. Then, a similarity calculation model is trained based on the cropped 224×224 image.

5. The multi-dimensional joint determination method for monitoring the opening and closing status of a disconnecting switch as described in claim 1, characterized in that, The specific steps of step S4 are as follows: S41, acquire the on-site image of the disconnector switch that needs to be monitored in real time; S42, use the YOLO26 target detection model trained in step S1 to perform detection; S43, extract the ROI region of the disconnector switch at the current moment to ensure the consistency of the image under test with the reference image in terms of spatial position and scale, and eliminate background interference.

6. The multi-dimensional joint determination method for monitoring the opening and closing status of a disconnecting switch as described in claim 3, characterized in that, The specific steps of step S5 are as follows: S51, calculate the similarity between the ROI area of ​​the disconnecting switch to be tested and the two reference maps in step S2; S52, calculate the similarity between the ROI area of ​​the disconnecting switch to be tested and the two reference maps in step S2 using the similarity model in step S3, and make a preliminary judgment by setting a similarity threshold; S53, if the similarity of both is lower than the threshold, the device is directly determined to be in the "intermediate state" of the operation process; if either similarity is higher than the threshold, the device enters the fine judgment stage, and the similarity threshold is determined by a large amount of data testing and statistics.

7. The multi-dimensional joint determination method for monitoring the opening and closing status of a disconnecting switch as described in claim 1, characterized in that, The specific steps of step S6 are as follows: S61. For the test image that has passed the coarse screening, the contact area is extracted, and the Shi-Tomasi algorithm is used to extract the corner points with high stability on the corresponding reference image. S62. The Lucas-Kanade sparse optical flow algorithm is used to calculate the corresponding corner points on the test image, and the average value of the Euclidean distance between the corner points of the test image and the corner points of the corresponding reference image is calculated. S63. By comparing the average value of the Euclidean distance with the distance threshold, the actual opening and closing state of the disconnector switch in the test image is finally accurately determined. The distance threshold is also determined by a large amount of data testing and statistics.

8. The multi-dimensional joint determination method for monitoring the opening and closing status of a disconnector switch as described in claim 7, characterized in that, The average Euclidean distance between the corner points of the image under test and the corresponding corner points of the reference image is calculated by equation (1): （1） The set of valid corner points extracted from the benchmark image is as follows: The set of corresponding corner points of the image to be measured obtained by optical flow method is , For the reference figure, the first There are 1 valid coordinate point, with coordinates as follows: , For the first part of the graph to be tested There are 1 valid coordinate point, with coordinates as follows: , This represents the total number of valid corner points that were successfully matched.

Images