Edge image detection method for power switch cabinet pressure plate

Abstract

This invention relates to the technical field of power systems, and provides a method for edge image detection of power panel cabinet pressure plates. The method includes: processing a captured image of the pressure plate in grayscale to obtain an image; performing convolutional smoothing on the image to generate a denoised and smoothed image; performing noise reduction, abrupt change calculation, and edge calculation on each pixel within a region of the grayscale image to obtain another image; assigning a grayscale value of 0 to a pixel located at the center of its neighborhood if the grayscale value of all pixels in its column is less than the average grayscale value of the neighborhood, thus generating a grayscale image; assigning values ​​to pixels in the grayscale image whose grayscale values ​​are not zero, resulting in grayscale values ​​greater than 0, and saving the coordinates of these pixels to an array; assigning a grayscale value of 125 to all pixels in the array whose coordinates are within the array, and setting the rest to 0, thus obtaining an edge map. Edge recognition is achieved entirely through convolutional calculations, eliminating the need for offline training with massive amounts of deep learning samples and complex network iterative computations.

Description

Technical Field

[0001] This invention relates to the technical field of power systems, and in particular to a method for edge image detection of power switchgear pressure plates. Background Technology

[0002] Power distribution cabinets contain various functional pressure plates, which are key components of power control systems. Different pressure plates have clearly defined functions: maintenance pressure plates ensure the safety of on-site maintenance personnel; local pressure plates enable remote control shielding; alarm pressure plates provide immediate on-site audible and visual alarms; and protection interlocking pressure plates manage the relay protection interlocking logic. A single cabinet can contain up to 30 pressure plates, often concentrated in confined areas. Currently, most pressure plates lack digital acquisition sensors, making it impossible to automatically upload their activation / deactivation status. Therefore, in power auxiliary monitoring system operation and maintenance scenarios, it is crucial to rely on video image recognition technology to determine the actual activation / deactivation position of the pressure plates online and promptly identify abnormal activation / deactivation faults.

[0003] Existing image detection solutions for pressure plate status mostly rely on machine learning. This involves first building an image recognition model, training it using offline sample datasets, and then using the completed model to capture real-time images for pressure plate status identification. This traditional approach has significant drawbacks: model training requires a massive amount of image samples, resulting in a large workload for data preparation, complex overall algorithm logic, low equipment processing speed, and high costs for project implementation and hardware / software modifications. Consequently, it is difficult to achieve efficient and low-cost intelligent online monitoring of pressure plates in cabinets. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for edge image detection of the pressure plate of a power distribution cabinet, comprising: S1. Use a camera to photograph the pressure plates of the substation cabinets, and perform grayscale processing on the photographed pressure plate images to obtain the original grayscale image. ; S2, for the original grayscale image A 5th-order smooth convolution kernel with a discrete approximation of a two-dimensional Gaussian is used. Perform convolutional smoothing to generate a noise-reduced and smoothed grayscale image. ; S3, extract grayscale image medium pixel Defined by pixels A square region centered at a point with sides of 9 pixels is The region, and on Each pixel within the region is truncated using upper and lower thresholds to remove noise, resulting in a denoised grayscale image. ;in Indicates the first Line 1 The number of pixels in a column; S4, for grayscale images use 3D longitudinal convolution kernel The grayscale image is obtained by calculating the mutation amount. ; S5, for grayscale images pass Single-column vertical edge convolution kernel Edge calculation is performed to obtain a grayscale image. ; S6, grayscale image In the middle If the grayscale value of the pixel at the center of the neighborhood is less than the average grayscale value of all pixels in the neighborhood, then the grayscale value of that pixel is set to 0, and a grayscale image is generated. ; S7, convert the grayscale image For pixels with a non-zero grayscale value, their grayscale value is assigned as 8 times the current grayscale value minus the sum of the grayscale values ​​of their 8 neighboring pixels. If the assigned grayscale value is greater than 0, the coordinates of that pixel are saved to an array. Among them, the 8 neighboring pixels refer to the 8 pixels centered on the given pixel. S8, the original grayscale image The midpoint coordinate in the array All pixels in the array are assigned a grayscale value of 125 and are not in the array. The grayscale value of all pixels is set to 0 to obtain the edge map. .

[0005] Furthermore, the convolution kernel The kernel template values ​​are: all edges are 1, all inner rings are 2, and the center position is 4. The sum of all elements is 36. The expression is: .

[0006] Furthermore, in step S3, for The specific steps for noise reduction processing of each pixel within the region are as follows: If the grayscale value of this pixel is greater than the grayscale upper limit threshold Then its grayscale value is set to The grayscale value of a pixel is less than the lower grayscale threshold. Then its grayscale value is set to ; In the formula, For the ZR region OK The grayscale value of each column pixel.

[0007] Furthermore, in step S4, the vertical convolution kernel... The expression is: .

[0008] Furthermore, in step S5, the convolution kernel... The expression is: .

[0009] The present invention has the following beneficial effects: (1) This invention is designed for the working conditions of concentrated and densely arranged pressure plates of power panel cabinets. It adopts a 9×9 large-size neighborhood threshold constraint and a 5×5 approximate Gaussian smoothing multi-level large-scale denoising convolution strategy. Compared with small-size filtering, it can fully cover the noise caused by reflection, dust and ambient light in the continuous area of ​​pressure plates. Combined with the inherent structure of the pressure plate with the longitudinal contour as the main feature, it abandons the universal edge algorithm in all directions and customizes a dedicated longitudinal convolution operator to extract gray-scale change features. It accurately fits the longitudinal shape rules of the pressure plate lever and the vertical edge of the cabinet. Even in the scenario where the effective pixel ratio of the pressure plate is small, it can still accurately lock and completely extract the edge contour of a single pressure plate, avoiding edge omission and edge breakage problems due to the small number of pixels of the pressure plate and the cluttered background.

[0010] (2) The present invention relies on convolutional computation to achieve edge recognition throughout the process. It does not require offline training of massive samples of deep learning or complex network iterative computation. The recognition logic is simple and controllable. The amount of data involved in the global computation is greatly reduced, and the algorithm has a faster computation speed. It does not require high-performance industrial control hardware or image acquisition and can reuse existing monitoring cameras and conventional edge processing equipment in substations. The hardware modification cost and project implementation cost are lower, and it is easy to deploy in batches in the existing auxiliary monitoring system of substations.

[0011] (3) The present invention first filters the effective edge pixel coordinates in the processing image, and then maps the effective points back to the original grayscale image to generate a marked edge image. The edge pixels are assigned fixed values ​​and the non-edge pixels are set to 0. The final generated edge image has a clear outline, which makes it easy to quickly determine the position and status of the pressure plate based on the edge outline, and provides a reliable image basis for intelligent online monitoring of the pressure plate of the cabinet. Attached Figure Description

[0012] Figure 1 This is a flowchart of the edge image detection process of the present invention. Detailed Implementation

[0013] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. However, these embodiments are not intended to limit the present invention. Any similar structures and similar variations of the present invention should be included in the protection scope of the present invention. The commas in the present invention all indicate the relationship between and. The English letters in the present invention are case-sensitive.

[0014] like Figure 1 As shown, the present invention provides an edge image detection method for the pressure plate of a power distribution cabinet, comprising: S1. Use a camera to photograph the pressure plates of the substation cabinets, and perform grayscale processing on the photographed pressure plate images to obtain the original grayscale image. Specifically, industrial cameras are used to capture real-world color images of the pressure plates in the display cabinet. These images are then converted to grayscale to remove redundant RGB color information, compress the image data dimensions, and eliminate color interference. While preserving the outline and grayscale features of the pressure plates, the computational load of subsequent convolution and filtering operations is reduced to generate the original grayscale image. ; S2, for the original grayscale image A 5th-order smooth convolution kernel with a discrete approximation of a two-dimensional Gaussian is used. Perform convolutional smoothing to generate a noise-reduced and smoothed grayscale image. ; convolution kernel The kernel template values ​​are: all edges are 1, all inner rings are 2, and the center position is 4. The sum of all elements is 36. The expression is: An engineering approximation Gaussian smooth convolution is used to filter out ambient light spots, sensor salt-and-pepper noise, and high-frequency random noise caused by cabinet reflections. While blurring the fine noise, the edge contour of the pressure plate is preserved to the maximum extent, avoiding excessive blurring of the pressure plate edge caused by conventional mean filtering, thus completing the initial noise reduction to obtain a smooth grayscale image.

[0015] S3, extract grayscale image medium pixel Defined by pixels A square region centered at a point with sides of 9 pixels is The region, and on Each pixel within the region is truncated using upper and lower thresholds to remove noise, resulting in a denoised grayscale image. ,in Indicates the first Line 1 The pixels in the column; by truncating ultra-high brightness noise and ultra-low dark spot noise, and retaining the effective grayscale range of the pressure plate body, a second fine noise reduction is completed. The specific steps are as follows: If the grayscale value of this pixel is greater than the grayscale upper limit threshold Then its grayscale value is set to The grayscale value of a pixel is less than the lower grayscale threshold. Then its grayscale value is set to ; In the formula, for The first in the region OK The grayscale value of each column pixel.

[0016] S4, for grayscale images use 3D longitudinal convolution kernel The grayscale image is obtained by calculating the mutation amount. By using vertical single-column weighted differential convolution, the gray-level abrupt change features in the vertical direction of the pressure plate are extracted, the vertical side edges of the pressure plate and the vertical contour of the release lever are enhanced, the horizontal texture interference is filtered out, and the vertical edge pixels are highlighted through gradient operation to generate a vertical feature enhancement map.

[0017] Vertical convolution kernel The expression is: .

[0018] S5, for grayscale images pass Single-column vertical edge convolution kernel Edge calculation is performed to obtain a grayscale image. ; convolution kernel The expression is: .

[0019] S6, grayscale image In the middle If the grayscale value of the pixel at the center of the neighborhood is less than the average grayscale value of all pixels in the neighborhood, then the grayscale value of that pixel is set to 0, and a grayscale image is generated. ;use Local neighborhood column grayscale statistics rules remove invalid isolated noise points and false edges: the center point of the entire column with low grayscale is determined to be a non-pressing plate valid edge and is set to 0 to remove it, filter out the residual fine false edges, and purify the real pressing plate edge pixels. S7, convert the grayscale image For pixels with a non-zero grayscale value, assign their grayscale value to 8 times the current grayscale value minus the sum of the grayscale values ​​of their 8 neighboring pixels. If the assigned grayscale value is greater than 0, save the coordinates of that pixel to an array. The 8 neighboring pixels refer to the 8 pixels centered at the given pixel. The remaining valid edge pixel coordinates after pseudo-edge removal are filtered and stored in an array. Complete edge point positioning and accurately lock the pixel position of the actual outline of the pressure plate; S8, the original grayscale image The midpoint coordinate in the array All pixels in the array are assigned a grayscale value of 125 and are not in the array. The grayscale value of all pixels is set to 0 to obtain the edge map. The filtered valid edge coordinates are mapped back to the original image. Edge points are assigned a grayscale value of 125, while non-edge points are set to 0, ultimately generating a binarized edge image of the pressure plate. The outline of the screen cabinet pressure plate's projection and retraction structure is clearly highlighted, and the edge of the pressure plate is accurately extracted.

[0020] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

Claims

1. A method for edge image detection of the pressure plate of a power distribution cabinet, characterized in that, include: S1. Use a camera to photograph the pressure plates of the substation cabinets, and perform grayscale processing on the photographed pressure plate images to obtain the original grayscale image. ; S2, for the original grayscale image A 5th-order smooth convolution kernel with a discrete approximation of a two-dimensional Gaussian is used. Perform convolutional smoothing to generate a noise-reduced and smoothed grayscale image. ; S3, extract grayscale image pixels in Defined by pixels A square region centered at a point with sides of 9 pixels is The region, and on Each pixel within the region is truncated using upper and lower thresholds to remove noise, resulting in a denoised grayscale image. ;in Indicates the first Line 1 The number of pixels in a column; S4, for grayscale images use 3D longitudinal convolution kernel The grayscale image is obtained by calculating the mutation amount. ; S5, for grayscale images pass Single-column vertical edge convolution kernel Edge calculation is performed to obtain a grayscale image. ; S6, grayscale image In the middle If the grayscale value of the pixel at the center of the neighborhood is less than the average grayscale value of all pixels in the neighborhood, then the grayscale value of that pixel is set to 0, and a grayscale image is generated. ; S7, convert the grayscale image For pixels with a non-zero grayscale value, their grayscale value is assigned as 8 times the current grayscale value minus the sum of the grayscale values ​​of their 8 neighboring pixels. If the assigned grayscale value is greater than 0, the coordinates of that pixel are saved to an array. Among them, the 8 neighboring pixels refer to the 8 pixels centered on the given pixel. S8, the original grayscale image The midpoint coordinate in the array All pixels in the array are assigned a grayscale value of 125 and are not in the array. The grayscale value of all pixels is set to 0 to obtain the edge map. .

2. The edge image detection method for the pressure plate of the power distribution cabinet according to claim 1, characterized in that, The convolution kernel The kernel template values ​​are: all edges are 1, all inner rings are 2, and the center position is 4. The sum of all elements is 36. The expression is: 。 3. The edge image detection method for the pressure plate of the power distribution cabinet according to claim 1, characterized in that, In step S3, for The specific steps for noise reduction processing of each pixel within the region by upper and lower threshold truncation are as follows: If the grayscale value of this pixel is greater than the grayscale upper limit threshold Then its grayscale value is set to The grayscale value of a pixel is less than the lower grayscale threshold. Then its grayscale value is set to ; In the formula, For the ZR region OK The grayscale value of each column pixel.

4. The edge image detection method for the pressure plate of the power distribution cabinet according to claim 1, characterized in that, Vertical convolution kernel in step S4 The expression is: 。 5. The edge image detection method for the pressure plate of the power distribution cabinet according to claim 1, characterized in that, Convolution kernel in step S5 The expression is: 。

Images