Image processing-based quality detection method for surge arrester monitor

Abstract

The present application relates to the technical field of image processing, and more particularly to a lightning arrester monitor quality detection method based on image processing. The method comprises: collecting and processing a surface image to obtain a gray image; calculating edge uniformity of each image block in the gray image through a sliding window; dividing the image into a dial area and a background area according to the edge uniformity, and performing adaptive filtering on the surface image; converting the filtered color image to HSV space, combining color similarity and spatial continuity to identify the pointer position; and finally determining the quality of the monitor according to the deviation of the pointer position and the dial scale parameters through arc fitting. The present application can effectively suppress noise and protect key features such as the pointer under complex lighting conditions through adaptive filtering based on edge uniformity, significantly improving the accuracy and robustness of quality detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a lightning arrester monitor quality detection method based on image processing. BACKGROUND

[0002] The lightning arrester monitor is a key protection device of the power system, and its factory quality is directly related to the safety of the power grid. In production and assembly, ensuring that the pointer is accurately stopped at the zero scale point when not in use is a core quality control indicator. Deviation of the initial position of the pointer will cause reading errors and interfere with maintenance judgment during operation, and will also expose defects in the assembly process and mechanical structure stress problems.

[0003] The traditional lightning arrester monitor quality detection method is mainly manual visual detection. This method has obvious limitations in modern large-scale industrial production. Manual inspection requires high experience and concentration of the detection personnel, and long-time work can easily cause visual fatigue, resulting in missed detection or misjudgment.

[0004] In recent years, lightning arrester monitor quality detection methods based on image processing have gradually attracted attention. In related technologies, a Gaussian filter is often used to detect the surface image of the lightning arrester monitor to be processed. The process is usually as follows: using a preset uniform filter strength for denoising, and then using edge detection or template matching method to locate the pointer position. However, using a preset uniform filter strength for denoising has obvious limitations: the lighting conditions in the production workshop are not absolutely constant, and slight fluctuations in light and slight differences in reflectivity of the glass cover of different batches of products will produce noise or highlight interference in the image. The preset uniform filter strength cannot adapt to these dynamic changes. Overly strong filtering will blur the already small pointer edges, and overly weak filtering will not effectively suppress the interference, both of which will cause pointer positioning failure or precision decline, and thus affect the judgment of the quality of the lightning arrester monitor. SUMMARY

[0005] To solve the technical problem that the above-mentioned image processing method based on a preset uniform filter strength cannot adapt to dynamic changes in the production environment, resulting in inaccurate pointer positioning and affecting the judgment of the quality of the lightning arrester monitor, the present application provides a lightning arrester monitor quality detection method based on image processing, which comprises the following steps:

[0006] The surface image of the lightning arrester monitor is collected, and grayscale processing and edge detection are performed to obtain a grayscale image; a preset sliding window is set to move on the grayscale image to calculate the edge neatness of the image block corresponding to the sliding window during movement, and the edge neatness is used to represent the complexity and direction consistency of the edges in the image block; adaptive filtering processing is performed on the image block corresponding to the surface image to obtain a filtered image; the filtered image is divided into a dial area and a background area according to the edge neatness of all the image blocks; the image corresponding to the dial area is converted from an RGB color space to an HSV color space, and pointer pixels are screened out according to the color similarity and spatial continuity of the pixels; straight line detection is performed on the pointer pixels to identify the pointer position of the lightning arrester monitor; arc fitting is performed on adjacent edges of the pointer position to obtain a dial scale arc, and the quality of the lightning arrester monitor is determined according to the pointer position and the dial scale arc.

[0007] The preset sliding window is set to move on the grayscale image, and the edge neatness of the image block corresponding to the sliding window during movement is calculated. The edge neatness is used to represent the complexity and direction consistency of the edges in the image block. The adaptive filtering processing is performed on different image blocks based on the edge neatness. The accuracy and reliability of the automatic detection under complex lighting and background environment are improved.

[0008] Preferably, the calculation of the edge neatness of the image block includes: identifying the edge pixels obtained by edge detection in the image block; calculating the edge richness of the image block based on the density of the edge pixels in the image block and the slope difference between adjacent edge pixels; calculating the image entropy of the image block; and the edge neatness of the image block is negatively correlated with the product of the image entropy of the image block and the edge richness of the image block.

[0009] The edge richness representing the number and direction change of edge points is combined with the image entropy representing the texture complexity, and the negative correlation between the edge richness and the image entropy is established, which provides a very reliable and accurate basis for the subsequent adaptive filtering and region division.

[0010] Preferably, the division of the filtered image into the dial area and the background area includes: using a k-means clustering algorithm to cluster the filtered image into two clusters according to the edge neatness, and the two clusters correspond to the dial area and the background area, respectively.

[0011] Preferably, the adaptive filtering process for the image patch corresponding to the surface image includes: applying Gaussian filtering to the image patch corresponding to the surface image, wherein the adaptive standard deviation of the Gaussian filter for any image patch satisfies the following relationship:

[0012] ;

[0013] in, It is the first Adaptive standard deviation of each image patch; It is the preset standard deviation of the surface image; It is the first Edge regularity of each image patch; It is the first Any pixel obtained by edge detection from an image patch Edge strength; It is the maximum value of the edge intensity of the pixel obtained by edge detection of the surface image; It is a preset tiny value; It is the standard normalization function.

[0014] The standard deviation of the Gaussian filter in this invention is dynamically correlated with the edge neatness and edge strength of the image patch. This enables refined and intelligent control of the filter intensity, automatically enhancing the filter for background areas with low edge neatness and weakening the filter for dial areas containing important details, thus resolving the contradiction between noise reduction and fidelity preservation.

[0015] Preferably, the color similarity of the pixels satisfies the following relationship:

[0016] ;

[0017] in, The pixels of the dial area Color similarity, The pixels of the dial area Pixel saturation; The pixels of the dial area Brightness; It is the maximum value of saturation; It is the maximum value of brightness; It is a natural exponential function.

[0018] This method calculates color similarity using an exponential function that combines saturation and brightness within the HSV color space, which is more favorable for color perception. This more effectively eliminates the influence of lighting variations on color, and is particularly effective in distinguishing the black of a pointer from dark shadows in the background.

[0019] Preferably, the step of filtering pointer pixels based on the color similarity and spatial continuity of pixels includes: calculating the spatial continuity of the pixel based on the HSV distance and direction difference between the pixel and each pixel in its preset neighborhood; calculating the probability that the pixel is a pointer pixel based on the color similarity and spatial continuity of the pixel; and recording the pixel as a pointer pixel when the probability that the pixel is a pointer pixel is greater than a preset threshold.

[0020] Preferably, the probability that the pixel is a pointer pixel satisfies the following relationship:

[0021] ;

[0022] in, The pixels of the dial area The probability of being a pointer pixel; The pixels of the dial area Color similarity; The pixels of the dial area Any pixel within the preset neighborhood The coordinate values; The pixels of the dial area The preset neighborhood; It is a pixel. Its preset neighborhood pixels Spatial similarity; It is the pixel within the preset neighborhood. Relative to pixel The weight, It is the standard normalized function.

[0023] This scheme introduces spatial continuity and combines color similarity to calculate the probability that a pixel is a pointer pixel. That is, a pixel must not only have the correct color but also form a continuous structure with its surrounding pixels. This enhances the robustness of pointer recognition, effectively filtering out isolated noise points and artifacts, ensuring that the finally identified pointer has a true geometric shape, and achieving a recognition accuracy far superior to methods relying solely on color selection.

[0024] Preferably, the edge detection is performed on the grayscale image using the Canny edge detection algorithm to extract edge pixels and the edge intensity of the pixels.

[0025] Preferably, determining the quality of the surge arrester monitor based on the pointer position and the dial scale arc includes: identifying the pointer tip based on the pointer position; calculating the absolute difference in distance from the pointer tip to the midpoint of the dial scale arc to obtain a first deviation; and calculating the distance from the pointer tip to the center of the arc. The included angle of the axis and the endpoint of the arc relative to the center of the arc an absolute difference value of the included angle of the shafts, to obtain a second deviation amount; and evaluating the quality of the lightning arrester monitor based on a product of the first deviation amount and the second deviation amount.

[0026] Preferably, the straight line detection is by Hough straight line detection.

[0027] The present application has the beneficial effects that: in view of the contradiction that fixed parameter filtering cannot consider both noise reduction and detail preservation in the existing image detection technology, a new strategy of image partition and adaptive filtering based on edge regularity is proposed. The present application quantifies the structured features of local regions of an image, applies filtering strength matching different regions such as a dial and a background to the regions, suppresses noise, and protects the outlines of key features such as a pointer of a monitor. The quality of image preprocessing is fundamentally improved, a solid foundation is laid for subsequent accurate recognition, and the accuracy and environmental adaptability of a detection system are significantly improved. The present application performs regional division after filtering operation, combines a series of fine processing procedures such as color analysis, spatial continuity discrimination, straight line detection, and circular arc fitting, and solves the problems of low recognition accuracy and poor robustness caused by defects in the image preprocessing link. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 A flowchart of a lightning arrester monitor quality detection method based on image processing provided for an embodiment of the present application is shown in the figure.

[0029] Figure 2 A schematic diagram of a dial scale circular arc provided for an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0030] An embodiment of the present application provides a lightning arrester monitor quality detection method based on image processing, as shown in the figure, which includes steps S100-S700. Figure 1

[0031] Step S100, surface images of a lightning arrester monitor are collected, and grayscale processing and edge detection are performed to obtain a grayscale image.

[0032] It should be noted that this step is a data acquisition and basic feature extraction link of the entire quality detection process. The image collected by an original industrial camera contains a large amount of background information and color information, and direct processing has a large amount of calculation and is easily disturbed by coupling between color channels. Therefore, the original image must be converted into a grayscale image, and edge detection is performed, to provide basic data input for subsequent edge feature-based analysis.

[0033] ​Specifically, first, the surface image of the lightning arrester monitor is collected by the industrial camera deployed at the fixed station of the production line. Subsequently, the collected RGB color surface image is processed to grayscale, which is converted into a single-channel grayscale image. Then, the obtained grayscale image is subjected to edge detection.

[0034] As a preferred embodiment, the Canny edge detection algorithm can be used for edge detection. This algorithm can effectively locate the edge while suppressing noise, and through non-maximum suppression and double threshold processing, finally output a binary edge map which clearly identifies the position of the edge pixel points, and the edge intensity information of each pixel point can be obtained synchronously.

[0035] At this point, the grayscale image containing the lightning arrester monitor dial, and the edge pixel point set and corresponding edge intensity information obtained by edge detection are obtained.

[0036] Step S200, a sliding window is preset to move on the grayscale image to calculate the edge regularity of the image block corresponding to the moving process of the sliding window.

[0037] It should be noted that this step is the key prerequisite for realizing the adaptive filtering process of the present application. In the monitor dial image, the edges of the effective feature regions such as the pointer and the scale have high regularity and direction consistency, while the edges of the background noise or reflection regions present random and chaotic patterns. In order to quantify this difference, this step proposes the edge regularity index, which is used to mathematically represent the complexity and direction consistency of the edges in the local region of the image, thereby providing a reliable basis for subsequent region division.

[0038] Specifically, first, a sliding window of a fixed size, for example pixels, is preset to slide on the grayscale image obtained in step S100 with a preset step size, for example 8 pixels. For each image block covered by the moving sliding window, the edge regularity of the image block is calculated. The edge regularity of the image block integrates the edge richness of the image block and the image entropy. Next, they are described respectively.

[0039] For the edge richness of the image block, it should be noted that the edge richness of the image block can quantify the density and direction change of the edges in the image block. The more dense and unordered the direction change of the edges in a region, the higher the richness.

[0040] According to the above logic, the edge richness of any image block in the grayscale image satisfies the relationship:

[0041] ;

[0042] wherein, is the edge richness of the image block, and is the edge richness of the image block. Edge richness of each image patch; It is the first The total number of edge pixels in each image block; It is the first The total number of pixels in each image block; , They are the first The first image patch The, the The slope of each edge pixel It is the absolute value symbol.

[0043] In this formula, The density of edge pixels was quantized. This quantifies the cumulative change in slope between adjacent edge pixels, reflecting the discreteness of edge orientation in an image patch. When an image patch contains a large number of randomly oriented edges, its edge richness increases significantly.

[0044] Regarding image entropy for image patches, it's important to note that image entropy measures the randomness of grayscale distribution within an image patch. A region with complex texture typically has high image entropy. The calculation of image entropy is a current technique and will not be elaborated upon here.

[0045] Two metrics for calculating the edge regularity of an image patch were obtained, and then the edge regularity of the image patch can be calculated. The richer the edges and the higher the image entropy of a region, the more disordered and complex its internal structure tends to be, and therefore the lower its regularity. In this embodiment, the edge regularity of an image patch is negatively correlated with the product of the image entropy and the edge richness of the image patch.

[0046] Preferably, the edge regularity of any image patch Satisfying the relation:

[0047] ;

[0048] in, It is the first Edge richness of each image patch; It is the first Image entropy of image blocks; It is a natural exponential function.

[0049] In this formula, when both the edge richness and image entropy of an image patch are high, their product is large, and the exponent is a large negative number, making the edge regularity approach 0. Conversely, for regions containing regular, unidirectional scale lines, both their edge richness and image entropy are low, and their edge regularity approaches 1.

[0050] At this point, the edge regularity corresponding to each image block in the surface image has been obtained.

[0051] Step S300: Perform adaptive filtering on the image blocks corresponding to the surface image to obtain the filtered image.

[0052] It should be noted that this step is one of the core technical means of the present invention to solve the problems of the prior art. This step does not directly process the grayscale image, but rather processes the surface image of the acquired surge arrester monitor. Since the image blocks here correspond one-to-one with the image blocks in step S200, the edge regularity of the image blocks obtained in step S200 can be directly applied to the filtering processing of the corresponding image blocks.

[0053] Gaussian filtering is a linear smoothing filtering method based on the weight distribution of a Gaussian function. It achieves denoising by constructing a Gaussian kernel and convolving it with the image. Its advantages lie in its ability to effectively suppress high-frequency interference such as Gaussian noise, and because the weights have a centrally symmetrical bell-shaped distribution, it preserves image details and overall structure well during denoising.

[0054] However, traditional global uniform filtering uses a preset fixed filtering strength, which cannot simultaneously achieve noise reduction and feature preservation. This step utilizes edge regularity to apply Gaussian filtering with different strengths to different image blocks: for image blocks with low edge regularity, i.e., more noise and complex and messy edges, a stronger filtering strength is used to achieve strong smoothing of noisy areas; for image blocks with high edge regularity containing effective features, clear edges, and consistent orientation, a weaker filtering strength is used to achieve fine protection of effective feature areas, thus effectively solving the limitations of traditional filtering methods.

[0055] Specifically, this step uses Gaussian filtering to adaptively filter each image patch corresponding to the surface image. The key difference lies in the fact that the standard deviation of the Gaussian filter is no longer a fixed value, but is dynamically adjusted according to the characteristics of the image patch itself.

[0056] The adaptive standard deviation of the Gaussian filter for any image patch satisfies the following relationship:

[0057] ;

[0058] in, It is the first Adaptive standard deviation of each image patch; It is the preset standard deviation of the surface image; It is the first Edge regularity of each image patch; It is the first Any pixel obtained by edge detection from an image patch Edge strength; is the maximum value of the edge strength of the pixel points obtained by edge detection on the surface image, is a preset tiny value for preventing the denominator from being 0, which can be set as 0.01 or according to requirements; is a standard normalization function.

[0059] In the formula, for an image block, if the edge regularity is high and approaches 1, or the inside contains very strong edges, approaches at this time, this term will be large, resulting in a very small final adaptive standard deviation. The smaller adaptive standard deviation means a weaker filtering effect, thereby maximizing the retention of the edge details considered as important features. Conversely, for the background or noise area with low edge regularity and weak edge strength, the adaptive standard deviation will approach the preset standard deviation, achieving strong denoising.

[0060] At this point, the image after adaptive filtering is obtained, denoted as a filtered image.

[0061] Step S400, according to the edge regularity of all image blocks, the filtered image is divided into a dial area and a background area.

[0062] It should be noted that after adaptive filtering, although the image quality is improved, in order to improve the pertinence and efficiency of subsequent processing, for example, searching for a pointer only in the dial area, the image needs to be regionally divided.

[0063] Specifically, this step uses the edge regularity of all image blocks calculated in step S200 as a feature, and uses a k-means clustering algorithm to cluster all image blocks into two clusters. The k-means clustering algorithm is a classic unsupervised learning algorithm, which has the advantages of simple principle, fast convergence speed and high computational efficiency, and can quickly divide data into different clusters according to similarity, and is very suitable for processing data such as image block edge regularity which has obvious clustering characteristics.

[0064] From the image features, the dial area contains structured elements such as pointers and scales, so the edge morphology is relatively complex and the direction is diverse, and therefore the edge regularity value is generally low; and the background area is usually single or regular texture, and the edge features are simple and uniform, so the edge regularity value is generally high. This feature difference makes the two clusters obtained by clustering correspond to the dial area and the background area. Based on this, the number of cluster centers is set to 2, and after the algorithm converges, the cluster with a lower cluster center value is marked as the dial area, and the cluster with a higher cluster center value is marked as the background area.

[0065] At this point, the filtered image is logically divided into a dial area and a background area.

[0066] Step S500, the dial area corresponding image is converted from RGB color space to HSV color space, and the pointer pixel points are screened out according to the color similarity and spatial continuity of the pixel points.

[0067] It should be noted that this step aims to preliminarily screen out the pixel points constituting the pointer from the dial area. The pointer, as a physical entity, has not only similar color characteristics, but also spatial characteristics. By combining these two dimensions, a more robust screening model can be constructed to effectively exclude color-similar but isolated noise points.

[0068] Specifically, first, the image part corresponding to the dial area is converted from the RGB color space to the HSV color space. The screening is performed by calculating the probability of each pixel point becoming a pointer pixel point.

[0069] Regarding the color feature, the pointer is usually black, with very low saturation (S) and brightness (V). By calculating the color similarity of each pixel point, the higher the similarity, the more likely it is black. The pixel points that meet the color characteristics of the pointer can be preliminarily screened out.

[0070] According to the above logic, the color similarity of the pixel point satisfies the relationship:

[0071] ;

[0072] wherein, is the color similarity of the pixel point of the dial area, is the pixel saturation of the pixel point of the dial area; is the brightness of the pixel point of the dial area; is the maximum value of the saturation; is the maximum value of the brightness; is the natural exponential function.

[0073] This formula gets a higher color similarity when the saturation is low and the brightness is also low.

[0074] Regarding the spatial feature, relying solely on color screening may mistakenly consider black scales or characters as pointers. Therefore, it is necessary to further judge in combination with the spatial feature. The spatial feature is calculated by calculating the spatial continuity of each pixel point, i.e., calculating the similarity of the pixel point and other pixel points in its neighborhood in color and edge direction.

[0075] According to the above logic, the spatial similarity of the pixel point is calculated, which satisfies the relationship:

[0076] ;

[0077] wherein, is the pixel point of the dial area and any pixel point in its preset neighborhood ; is the pixel point of the dial area and the pixel point in its preset neighborhood ; is the gradient direction angle of the pixel point of the dial area ; is the gradient direction angle of the pixel point in the preset neighborhood of the pixel point of the dial area ; ; is the cosine function; is the absolute value symbol; is the standard normalization function.

[0078] In the formula, reflects the normalized value of the HSV value distance of two pixel points, the closer the distance, the higher the color feature score, and the higher the spatial similarity. The normalized value of the Euclidean distance is the ratio of the Euclidean distance to the maximum possible Euclidean distance in the HSV color space, and the maximum possible Euclidean distance in the HSV color space is usually 386. The absolute difference of the gradient direction angle of two pixel points is calculated, and the cosine function is used to map the absolute difference to , the smaller the difference, the closer the cosine value to 1, the higher the edge direction score, and the higher the spatial similarity. is the spatial similarity of the pixel point and the pixel point in its preset neighborhood , which considers both the HSV color distance and the gradient direction difference, and represents the spatial continuity.

[0079] It should be noted that the preset neighborhood of the pixel point is a neighborhood window with a size of or centered on the pixel point , and the size of the neighborhood window can also be set according to requirements. The neighborhood window set in this step is also required for calculating the probability of each pixel point being a pointer pixel point in the following.

[0080] Finally, the probability of each pixel point being a pointer pixel point is calculated by combining the color similarity and the spatial continuity, and the probability satisfies the relationship:

[0081] ;

[0082] wherein, is a pixel point in the dial area is the probability that the pixel point is a pointer pixel point; is a pixel point in the dial area is the color similarity of the pixel point is a pixel point in the dial area is any pixel point in the preset neighborhood of the pixel point is the coordinate value of the pixel point is a pixel point in the dial area is the preset neighborhood of the pixel point is a pixel point in the preset neighborhood is the weight of the pixel point with respect to the pixel point Generally, a two-dimensional Gaussian kernel is used for calculation, so that the pixel points closer to the pixel point have larger weights. The two-dimensional Gaussian kernel calculation is a prior art and will not be described here; is a standard normalization function, is the spatial similarity of the pixel point with the pixel points in its preset neighborhood .

[0083] In the formula, the spatial continuity of the pixel point is calculated. When there are a large number of pixels similar in color and direction in the neighborhood of a pixel, the value of the item will be high, and vice versa.

[0084] Finally, a probability threshold needs to be set, and the pixel points whose probability of being a pointer pixel point is greater than the threshold are recorded as pointer pixel points. The threshold can be set to 0.8, or it can be set according to requirements.

[0085] Step S600, straight line detection is performed on the pointer pixel points to identify the pointer position of the lightning arrester monitor.

[0086] It should be noted that the candidate pixel point set obtained in step S500 may still have a small amount of breakpoints or noise, and a global geometric model needs to be used to fit a unique and accurate pointer straight line.

[0087] Specifically, all the pointer pixel points obtained in step S500 are processed using Hough line detection. Hough transformation is a robust algorithm for finding specific shapes such as straight lines and circles in an image. It detects collinear point sets by mapping points in the image space to the parameter space for voting. The peak point with the highest cumulative value in the parameter space corresponds to the most prominent straight line in the image, and this straight line is identified as the pointer of the lightning arrester monitor.

[0088] At this point, the accurate geometric information of the lightning arrester monitor pointer position, such as the starting point, end point, and angle of the straight line, is obtained.

[0089] Step S700, the adjacent edges of the pointer position are fitted with a circular arc to obtain a dial scale circular arc, and the quality of the lightning arrester monitor is determined according to the pointer position and the dial scale circular arc.

[0090] It should be noted that after accurately identifying the pointer position, the last step is to compare it with the quality standard, i.e., the zero scale point, and give a quantitative and traceable quality evaluation result.

[0091] Specifically, in combination with Figure 2 Explain the quality evaluation process, which includes: first, according to the identified pointer position, locate the adjacent dial scale line edges near the tip of the pointer. Fit these edge points with a circular arc to obtain a dial scale circular arc that can accurately represent the dial scale reference. Then, calculate the absolute difference between the distance from the pointer tip to the midpoint of the dial scale circular arc to obtain the first deviation; calculate the absolute difference between the angle of the pointer tip relative to the center of the circular arc and the angle of the circular arc endpoint relative to the center of the circular arc to obtain the second deviation; finally, based on the product of the first deviation and the second deviation, evaluate the quality of the lightning arrester monitor. It should be noted that when calculating the product, both the first deviation and the second deviation need to be normalized to eliminate the influence of the dimension. The smaller the product, the closer the pointer position to the standard zero point, and the better the quality of the monitor.

[0092] The above are preferred embodiments of the present application, and are not limited to the protection scope of the present application, therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for quality inspection of surge arrester monitors based on image processing, characterized in that, Including the following steps: The surface image of the surge arrester monitor is acquired, and grayscale processing and edge detection are performed to obtain a grayscale image; A preset sliding window is moved across the grayscale image to calculate the edge regularity of the corresponding image patch during the sliding window's movement. Edge regularity characterizes the complexity and directional consistency of edges within the image patch, including: identifying edge pixels obtained through edge detection within the image patch; calculating the edge richness of the image patch based on the density of edge pixels within the image patch and the slope difference between adjacent edge pixels; and calculating the image entropy of the image patch. The edge regularity of the image patch is negatively correlated with the product of the image entropy and the edge richness of the image patch. Adaptive filtering is applied to the image patches corresponding to the surface image to obtain the filtered image, including: applying Gaussian filtering to the image patches corresponding to the surface image, where the adaptive standard deviation of the Gaussian filter for any image patch satisfies the following relationship: ； in, It is the first Adaptive standard deviation of each image patch; It is the preset standard deviation of the surface image; It is the first Edge regularity of each image block; It is the first Any pixel obtained by edge detection from an image patch Edge strength; It is the maximum value of the edge intensity of the pixel obtained by edge detection of the surface image; It is a preset tiny value; It is a standard normalized function; Based on the edge regularity of all image blocks, the filtered image is divided into a dial area and a background area; The image corresponding to the dial area is converted from the RGB color space to the HSV color space, and the pointer pixels are selected based on the color similarity and spatial continuity of the pixels, including: The spatial continuity of a pixel is calculated based on the HSV distance and orientation difference between the pixel and each pixel in its preset neighborhood. Based on color similarity and spatial continuity of pixels, the probability of a pixel being a pointer pixel is calculated, satisfying the following relationship: ； in, The pixels of the dial area The probability of being a pointer pixel; The pixels of the dial area Color similarity; The pixels of the dial area Any pixel within the preset neighborhood The coordinate values; The pixels of the dial area The preset neighborhood; It is a pixel. Its preset neighborhood pixels Spatial similarity; It is the pixel within the preset neighborhood. Relative to pixel The weights; When the probability of a pixel being a pointer pixel is greater than a preset threshold, the pixel is recorded as a pointer pixel. Perform line detection on pointer pixels to identify the pointer position of the surge arrester monitor; Arc fitting is performed on the adjacent edges of the pointer position to obtain the dial scale arc. Based on the pointer position and the dial scale arc, the quality of the surge arrester monitor is determined, including: identifying the pointer tip based on the pointer position; calculating the absolute difference in distance from the pointer tip to the midpoint of the dial scale arc to obtain the first deviation; and calculating the distance from the pointer tip to the center of the arc. The included angle of the axis and the endpoint of the arc relative to the center of the arc The absolute difference in the included angle of the shafts is used to obtain the second deviation; the quality of the surge arrester monitor is evaluated based on the product of the first deviation and the second deviation.

2. The image processing-based surge arrester monitoring quality detection method according to claim 1, characterized in that, The step of dividing the filtered image into a dial area and a background area includes: The k-means clustering algorithm is used to cluster the filtered image into two clusters based on the edge regularity, corresponding to the dial area and the background area respectively.

3. The image processing-based surge arrester monitoring quality detection method according to claim 1, characterized in that, The color similarity of the pixels satisfies the following relationship: ; in, The pixels of the dial area Color similarity, The pixels of the dial area Pixel saturation; The pixels of the dial area Brightness; It is the maximum value of saturation; It is the maximum value of brightness; It is a natural exponential function.

4. The image processing-based surge arrester monitoring quality detection method according to claim 1, characterized in that, The edge detection is performed on the grayscale image using the Canny edge detection algorithm to extract edge pixels and their edge strength.

5. The image processing-based surge arrester monitoring quality detection method according to claim 1, characterized in that, The line detection is performed using Hough line detection.

Images