A GIS shell production line real-time monitoring method and system

Abstract

The present application relates to the technical field of image processing, and more particularly to a GIS shell production line real-time monitoring method and system. The method comprises the steps of: acquiring a surface image of the GIS shell and performing grayscale processing; and performing adaptive noise reduction processing on the grayscale image. The adaptive noise reduction processing comprises: for each target pixel point, determining its noise attribute based on its grayscale value and neighborhood relationship; identifying a key region in the image representing a stress concentration part of the GIS shell, and determining the position importance of the target pixel point according to the spatial proximity and edge feature consistency of the target pixel point and the key region; and combining the noise attribute and the position importance to adaptively adjust the filtering strength in the filtering processing. The present application obtains unique filtering strength for each target pixel point, efficiently removes noise and preserves details, and improves the accuracy and reliability of defect detection.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for real-time monitoring of a GIS shell production line. Background Technology

[0002] As a core component of power systems, GIS housings are commonly used in substations and power transmission and distribution networks. Their surface quality directly affects the insulation performance and operational safety of the equipment. During the casting process, uneven heating and stress concentration can easily lead to potential defects such as dents and deformations on the surface of GIS housings. Failure to detect these defects in a timely manner can result in serious safety hazards. Therefore, real-time defect detection of GIS housing surfaces on the production line is crucial, and high-quality surface images are a prerequisite for accurate detection.

[0003] In actual production, image acquisition is easily affected by electromagnetic interference, equipment vibration, and changes in lighting in the industrial environment, resulting in a large number of noisy pixels mixed in the image, which seriously interferes with the extraction and recognition of defect features.

[0004] Existing technologies often employ PM filtering algorithms for image denoising, but this algorithm has significant limitations: it uses a fixed time step for all pixels, failing to adapt to the varying characteristics of different regions within the image. In flat areas with less noise, or in critical areas with stress concentration such as around holes and openings, a fixed time step can easily lead to over-smoothing, resulting in the loss of edge details for minor defects. Conversely, in areas with dense noise, a fixed time step may be insufficient to completely remove noise. This processing method makes it difficult for the denoised image to simultaneously meet the dual requirements of noise suppression and detail preservation, directly impacting the accuracy of subsequent defect detection. Summary of the Invention

[0005] To address the technical problem that existing filtering algorithms use a fixed time step to denoise GIS shell surface images, which cannot adapt to the differences in noise and structural characteristics in different regions, resulting in the denoised image failing to balance noise suppression and detail preservation, thus affecting the accuracy of subsequent defect detection, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for real-time monitoring of a GIS shell production line, the method comprising the steps of:

[0007] A surface image of the GIS shell on the production line is acquired and converted to grayscale to obtain a grayscale image. Adaptive denoising processing is then performed on the grayscale image to obtain a denoised image. Defect detection is performed on the GIS shell based on the denoised image to output its real-time monitoring status on the production line. The adaptive denoising processing includes: for each target pixel in the grayscale image, determining the noise attribute of the target pixel based on its grayscale value and the grayscale relationship with its neighboring pixels; identifying at least one key region in the grayscale image representing stress concentration areas of the GIS shell to determine the positional importance of the target pixel relative to the key region; and adaptively adjusting the filtering intensity applied to the target pixel in subsequent filtering processes, combining the noise attribute and positional importance of the target pixel.

[0008] This invention acquires and grayscales images of GIS shell surfaces, providing a foundation for subsequent analysis. The core of this invention lies in adaptive denoising, which determines the noise attributes of target pixels and their positional importance relative to key stress concentration areas, thereby adaptively adjusting the filtering intensity. Finally, defect detection is performed based on the denoised image. This invention solves the problem of traditional fixed-parameter filtering failing to balance denoising and detail preservation. It can enhance filtering for high-noise pixels to remove interference while weakening filtering for pixels in key areas to protect potential defect information. This provides high-quality image support for real-time production line monitoring, significantly improving the accuracy and reliability of GIS shell surface deformation defect detection, and meeting the high-precision quality control requirements of power equipment production.

[0009] Preferably, the noise attribute of the target pixel is quantified by the noise probability of the target pixel; wherein the noise probability is proportional to both the grayscale extreme degree and the grayscale isolation degree of the target pixel.

[0010] This invention quantifies the noise attributes of target pixels through noise probability, and the noise probability is positively correlated with the grayscale extreme degree and the grayscale isolation degree. It can accurately characterize the possibility of pixels being affected by noise, providing a scientific quantitative basis for subsequent filter intensity adjustment, and ensuring the objectivity and accuracy of noise identification.

[0011] Preferably, the noise probability of the target pixel satisfies the following relationship: ;in, It is the first The noise probability of each target pixel , They are the first The absolute difference between the gray value of a target pixel and the lower limit of the gray range, and the absolute difference between the gray value and the upper limit of the gray range; It is the first The sum of the absolute differences between the gray value of a target pixel and the gray values ​​of all pixels in its neighborhood. It is a preset micro value. It is the standard normalized function.

[0012] Preferably, identifying at least one key region in the grayscale image representing a stress concentration area of ​​the GIS shell includes: identifying the key region with a circular outline in the grayscale image by performing a Hough circle transform on the grayscale image.

[0013] This invention uses Hough circle transform to identify key areas of stress concentration in GIS shells. By utilizing the algorithm's ability to accurately detect circular contours, it can efficiently locate areas prone to deformation defects, such as holes and openings, providing an accurate benchmark for subsequent location importance assessment.

[0014] Preferably, determining the positional importance of the target pixel relative to the key region includes: determining the spatial proximity and edge feature consistency of the target pixel relative to the key region to determine the positional importance of the target pixel; wherein the positional importance of the target pixel is positively correlated with both spatial proximity and edge feature consistency.

[0015] This invention determines the positional importance of target pixels by spatial proximity and edge feature consistency, both of which are positively correlated with positional importance. This can comprehensively characterize the degree of association between target pixels and key regions, ensuring that the structural information of key regions and neighboring pixels is protected during denoising and avoiding loss of details.

[0016] Preferably, the spatial proximity is characterized by the Euclidean distance between the target pixel and the nearest key region; the edge feature consistency is characterized by the variance of the gradient magnitude between the target pixel and all edge pixels of the nearest key region.

[0017] Preferably, the subsequent filtering process is a PM filtering process, and the filtering intensity is achieved by adjusting the time step of the PM filtering process.

[0018] Preferably, the filtering intensity is achieved by adjusting the time step of the PM filtering process, including: generating a time step correction factor for the target pixel, wherein the time step correction factor is positively correlated with the noise attribute of the target pixel and negatively correlated with the positional importance of the target pixel; and calculating an adaptive time step for the target pixel based on the generated time step correction factor and a preset reference time step to complete the adjustment of the time step.

[0019] This invention generates a time step correction factor, establishes its correlation with noise attributes and positional importance, and then calculates an adaptive time step by combining it with a baseline time step. This achieves pixel-level dynamic optimization of the PM filtering time step, ensuring that each target pixel receives a filtering intensity that matches its characteristics, thereby improving the targeting and effectiveness of denoising.

[0020] Preferably, the adaptive time step satisfies the following relationship: ;in, It is the first Adaptive time step for each target pixel It is a preset fixed-size time step for each target pixel. It is the first Time step correction factor for each target pixel.

[0021] In a second aspect, the present invention provides a real-time monitoring system for a GIS shell production line. The real-time monitoring system for a GIS shell production line includes a memory and a processor. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the real-time monitoring method for a GIS shell production line according to the first aspect of the present invention is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the real-time monitoring method for a GIS shell production line according to the first aspect of the present invention, and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0023] The beneficial effects of this invention are as follows: The adaptive denoising method of this invention, for each target pixel in a grayscale image, no longer uses the fixed time step of traditional PM filtering. Instead, it dynamically adjusts the filtering intensity through a refined evaluation across two dimensions. This invention integrates two core indicators—noise attributes and location importance—to generate a unique time step correction factor tailored to each target pixel, thereby adjusting the time step of the PM filtering. This mechanism allows the invention to apply strong filtering to ordinary areas with dense noise, while performing moderate and gentle protective filtering in areas containing minor defects or located at the edges of critical structures. Therefore, this invention efficiently removes noise while preserving the edge contours of real defects and the geometric details of critical structures to the maximum extent, significantly improving the fidelity of the denoised image. This provides a high-quality data foundation for subsequent high-precision defect detection, ultimately improving the product quality monitoring level of the entire production line. Attached Figure Description

[0024] Figure 1 A flowchart of a real-time monitoring method for a GIS shell production line provided in an embodiment of the present invention;

[0025] Figure 2 This is a structural block diagram of a real-time monitoring system for a GIS shell production line provided in an embodiment of the present invention. Detailed Implementation

[0026] The first aspect of this invention provides a method for real-time monitoring of a GIS shell production line, such as... Figure 1 As shown, the method includes steps S100-S300:

[0027] Step S100: Obtain the surface image of the GIS shell on the production line and perform grayscale processing to obtain a grayscale image.

[0028] It should be noted that this step is fundamental to the monitoring method. Its core purpose is to capture raw images that comprehensively and accurately reflect the current surface condition of the GIS shell through non-contact optical imaging. High-quality raw images are the fundamental guarantee for all subsequent refined analysis and decision-making; their acquisition quality directly determines the final accuracy and reliability of defect detection.

[0029] Specifically, image acquisition devices are fixedly deployed above or to the side of the conveyor belt on the GIS shell production line to acquire images of the GIS shell surface. These devices can include, for example, high-definition cameras or industrial cameras. Subsequently, to reduce the computational dimensionality of subsequent processing, remove color information that contributes little to defect morphology identification, and highlight brightness changes caused by surface defects such as dents and scratches, the acquired RGB color images are converted to grayscale. For subsequent analysis, each pixel in the image is analyzed as a target pixel.

[0030] Thus, a grayscale image capable of characterizing the geometric and textural features of the GIS shell surface was obtained.

[0031] Step S200: Perform adaptive denoising processing on the grayscale image to obtain a denoised image.

[0032] It should be noted that this step is the core of this invention in resolving the technical contradiction of balancing noise reduction and detail preservation in existing technologies. Grayscale images acquired in industrial settings inevitably contain random noise introduced by factors such as electromagnetic interference and sensor thermal effects. Traditional filtering algorithms, such as Gaussian filtering or PM filtering with fixed parameters, often perform indiscriminate smoothing on the image when filtering out noise. This can easily lead to blurring of defect edges and even complete loss of feature information for minute defects.

[0033] Based on this, this step proposes an adaptive denoising scheme. The core idea is to evaluate the noise attributes and positional importance of each target pixel in the image, and dynamically adjust the filtering intensity accordingly. This effectively removes noise while preserving key structural details to the maximum extent. Specifically, the adaptive denoising process in this step includes steps S210-S230:

[0034] Step S210: For each target pixel in the grayscale image, determine the noise attribute of the target pixel based on the grayscale value of the target pixel and the grayscale relationship between the target pixel and its neighboring pixels.

[0035] It should be noted that noise points in an image typically have two main characteristics: one is that their grayscale values ​​exhibit extremes, that is, they tend to be pure black or pure white; the other is that their grayscale values ​​differ significantly from those of the surrounding pixels, exhibiting isolation.

[0036] Based on the above logic, this invention constructs a noise attribute quantification model, namely, noise probability. This is used to comprehensively evaluate the probability that a target pixel is a noise point, and the noise probability satisfies the following relationship:

[0037] ;

[0038] in, It is the first The noise probability of each target pixel , They are the first The absolute difference between the gray value of a target pixel and the lower limit of the gray range, and the absolute difference between the gray value and the upper limit of the gray range; It is the first The sum of the absolute differences between the gray value of a target pixel and the gray values ​​of all pixels in its neighborhood. It is a preset microvalue used to prevent It can be set to 0, or 0.001, or as needed; It is a standard normalization function used to quantize calculation results to... For the interval, specific methods such as minimum-maximum normalization and Z-score standardization can be used, which are existing technologies and will not be elaborated on here.

[0039] In this formula, The larger of the two values ​​is used to measure the extreme degree of grayscale value of the target pixel. The larger this value is, the more extreme the grayscale value of the target pixel. The more extreme the grayscale value of the target pixel, the more it indicates that the... The greater the likelihood that a target pixel is affected by noise, the higher the corresponding noise probability. The upper limit of the grayscale range is 255, representing pure white; the lower limit of the grayscale range is 0, representing pure black. Used to measure the degree of isolation of target pixels or local grayscale differences. The larger it is, the more likely it is to be the first The more isolated the grayscale value of the target pixel, the more it indicates that the first pixel is more isolated. The more extreme the grayscale value of a target pixel, the greater its credibility, indicating that it is more likely to be affected by noise, and the greater the corresponding noise probability. The neighborhood during acquisition can be set to the first... An 8-neighborhood window can be used for each target pixel. Of course, depending on the requirements for noise scale and computational complexity, a 4-neighborhood or larger neighborhood window can also be chosen, for example, centered on the target pixel. All other pixels within the window.

[0040] As an alternative implementation, frequency domain analysis can be used to determine the noise attributes of target pixels. Specifically, discrete wavelet transform or discrete cosine transform is performed on each target pixel and its neighborhood, and the energy or coefficients of its high-frequency subbands are analyzed. Typically, noise pixels exhibit abnormally high energy in the high-frequency subbands. This energy value can be used as an indicator to quantify its noise attributes. Wavelet transform is existing technology and will not be elaborated upon here.

[0041] At this point, the noise properties of each target pixel in the grayscale image have been obtained.

[0042] Step S220: Identify at least one key region in the grayscale image that represents the stress concentration area of ​​the GIS shell, in order to determine two indicators of the positional importance of the target pixel relative to the key region.

[0043] It should be noted that although step S210 obtains the noise probability based on the grayscale value analysis of each target pixel, this indicator alone cannot completely determine the time step of the target pixel during filtering. This is because the calculation of this indicator ignores the possibility of potential deformation defects in the image region where the target pixel is located, and this factor is crucial to the selection of filtering parameters.

[0044] Based on scenario research, multiple holes or openings on the GIS casing are often stress concentration points. When the GIS casing is subjected to pressure, deformation may occur around the holes due to excessive local stress. Therefore, if the target pixels affected by noise are located in such areas, determining the filtering time step solely based on their noise probability, while meeting denoising requirements, will destroy information about easily deformable areas in the image, resulting in insufficiently clear image support for subsequent detection of potential deformation defects. For these reasons, in denoising processing, it is essential to prioritize ensuring the integrity of the edge contours and structural information of these critical areas.

[0045] Specifically, this step first utilizes mature image recognition algorithms such as the Hough Circle Transform to automatically detect and locate all circular or near-circular contours in the grayscale image that conform to a preset radius range, marking these regions as key areas. The Hough Circle Transform is existing technology and will not be elaborated upon here. Subsequently, to determine the positional importance of each target pixel, this embodiment defines two metrics:

[0046] Spatial distance index , defined as the first The Euclidean distance between a target pixel and the center point of its nearest critical region. This metric value The smaller the value, the higher the value. The closer a target pixel is to the key region in space, that is, the higher its spatial proximity, the higher its positional importance.

[0047] Regional credibility index , defined as the variance of the gradient magnitudes of the target pixel and all edge pixels in its nearest key region. This metric value The smaller the value, the higher the consistency of the edge features of the key region, and the higher the positional importance of its neighboring pixels. In other words, the key region is a true structure rather than a pseudo-feature.

[0048] As an alternative implementation, spatial distance metrics can also be determined using other distance measurement methods, such as Manhattan distance and Chebyshev distance.

[0049] As an alternative implementation, the region confidence index can also be determined using other feature description and comparison methods, such as local binary patterns and histograms of oriented gradients (HOGs). For local binary patterns, the LBP feature histogram of the target pixel's neighborhood and the typical LBP feature histogram of the key region's edge can be calculated separately. Then, an inconsistency score is obtained by calculating the similarity between the two histograms, which serves as the region confidence index. For histograms of oriented gradients (HOGs), the region confidence index can be quantified by comparing the differences between the HOG features of the target pixel's neighborhood and the typical HOG features of the key region's edge.

[0050] This completes the identification of key areas on the GIS shell surface and obtains two indicators to quantify the importance of the location of each target pixel. and .

[0051] Step S230: Combining the noise attributes and positional importance of the target pixel, adaptively adjust the filtering intensity applied to the target pixel in subsequent filtering processes.

[0052] It should be noted that traditional filtering methods apply a uniform filtering intensity to the entire image, failing to balance noise suppression and detail preservation. This step aims to fuse the two core indicators—noise attributes and positional importance—obtained in the preceding steps to generate an adaptive filtering intensity for each target pixel. In this embodiment, the subsequent filtering processing is preferably Perona-Malik (PM) filtering, whose filtering intensity can be adjusted by changing the time step.

[0053] Specifically, this step involves setting a time step of a fixed size. Multiplied by a dynamically calculated time step correction factor This allows for adaptive adjustment of the filter strength.

[0054] Regarding the time step correction factor, it should be noted that its core construction lies in combining two indicators, the calculated noise probability and the positional importance in the grayscale image, and realizing the differentiated control of the time step through quantization. Its value directly determines the time step and filtering intensity of the target pixel in the PM filtering process.

[0055] Specifically, from a noise attribute perspective, since the primary purpose of PM filtering is noise reduction, a higher noise probability for a target pixel indicates a greater likelihood of the pixel being affected by noise. This necessitates stronger filtering, corresponding to a larger time step. Therefore, the time step correction factor is related to the noise probability of the target pixel. Proportional.

[0056] From the perspective of positional importance, since filtering may lose details, the time step correction factor is negatively correlated with positional importance, meaning it is... and They are positively correlated. Therefore, the higher the location importance, the weaker the filtering should be, corresponding to a smaller time step. These two dimensions together constitute the constraint logic of the time step correction factor, ensuring that it can effectively suppress noise while protecting critical structural information.

[0057] Therefore, the time step correction factor and , , All three are positively correlated. Based on the above logic, the time step correction factor satisfies the following relationship:

[0058] ;

[0059] in, It is the first Time step correction factor for each target pixel It is the first The noise probability of each target pixel It is the first Proximity of each target pixel It is the first Consistency of edge features of each target pixel The adjustment coefficient for the time step correction factor corresponding to each target pixel can be set to 0.5, which will ultimately adjust the value range of the time step correction factor corresponding to each target pixel to the range... This allows for bidirectional optimization, increasing or decreasing the time step size obtained in subsequent calculations, based on a preset fixed time step size.

[0060] Finally, by applying the calculated time step correction factor to the base time step, the adaptive time step for each target pixel can be obtained, which satisfies the following relationship:

[0061] ;

[0062] in, It is the first Adaptive time step for each target pixel It is a preset fixed-size time step for each target pixel. It is the first Time step correction factor for each target pixel.

[0063] The selection of ... The upper limit of the value is usually 1. or Specifically, this depends on the discretization scheme used. Therefore, this embodiment will... Setting it to 0.15 represents a good trade-off between convergence speed and accuracy while ensuring stability. Implementers can also set it according to their needs.

[0064] At this point, an adaptive time step for each target pixel has been obtained, and PM filtering can be performed based on this to obtain the final denoised image. PM filtering is an existing technology and will not be elaborated on here.

[0065] Step S300: Perform defect detection on the GIS shell based on the denoised image to output its real-time monitoring status on the production line.

[0066] It should be noted that this step is the decision output stage of the monitoring method. After the adaptive denoising processing in the aforementioned steps, random noise in the image is effectively suppressed, while the features of real defects on the GIS shell surface, such as microcracks and pores, are well preserved or even relatively enhanced.

[0067] Specifically, mature defect detection algorithms can be applied to denoised images. For example, methods such as threshold segmentation, morphological operations, and connected component analysis can be used to extract and identify potential defect areas. Once a defect is detected, the system can output corresponding monitoring status, such as alarm signals or information such as defect coordinates and dimensions, thereby achieving real-time and automated monitoring of the production quality of GIS shells.

[0068] The second aspect of this embodiment provides a real-time monitoring system for a GIS shell production line, such as... Figure 2 As shown, the real-time monitoring system for a GIS shell production line includes a memory and a processor. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the first aspect of the present invention, a real-time monitoring method for a GIS shell production line, is implemented.

[0069] The real-time monitoring system for the GIS shell production line also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their setup and functions are known in the art and will not be described in detail here.

[0070] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (DRAM), high-bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0071] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A GIS housing production line real-time monitoring method, characterized in that, The method comprises the steps of: acquiring a surface image of a GIS shell on a production line and performing grayscale processing to obtain a grayscale image; performing adaptive denoising processing on the grayscale image to obtain a denoised image; detecting defects in the GIS shell based on the denoised image to output a real-time monitoring state of the GIS shell on the production line; wherein the adaptive denoising processing comprises, for each target pixel point in the grayscale image, determining a noise attribute of the target pixel point based on a grayscale relationship between the target pixel point and its neighborhood pixel points; identifying at least one key region in the grayscale image representing a stress concentration part of the GIS shell, comprising: performing Hough circle transformation on the grayscale image to identify a key region having a circular contour in the grayscale image; determining a position importance of the target pixel point relative to the key region, comprising: determining a spatial proximity and an edge feature consistency of the target pixel point relative to the key region to determine the position importance of the target pixel point; wherein the position importance of the target pixel point is positively correlated with the spatial proximity and the edge feature consistency; the spatial proximity is represented by a Euclidean distance between the target pixel point and the nearest key region; and the edge feature consistency is represented by a variance of gradient amplitudes of all edge pixel points of the nearest key region relative to the target pixel point; combining the noise attribute and the position importance of the target pixel point to adaptively adjust a filter strength acting on the target pixel point in subsequent filter processing; the subsequent filter processing is PM filter processing; the filter strength is realized by adjusting a time step of the PM filter processing, comprising: generating a time step correction factor of the target pixel point, the time step correction factor being positively correlated with the noise attribute of the target pixel point and negatively correlated with the position importance of the target pixel point; and calculating an adaptive time step of the target pixel point based on the generated time step correction factor and a preset reference time step to complete the adjustment of the time step; the noise attribute of the target pixel point is quantified by a noise probability of the target pixel point; wherein the noise probability is positively correlated with a grayscale extremeness and a grayscale isolation of the target pixel point; The noise probability of the target pixel point satisfies a relationship: ; wherein, is the noise probability of the th target pixel point, , are the absolute difference between the gray value of the th target pixel point and the lower limit of the gray range, and the absolute difference between the gray value of the th target pixel point and the upper limit of the gray range, respectively; is the sum of the absolute difference between the gray value of the th target pixel point and the gray value of each pixel point in the neighborhood of the th target pixel point, is a preset infinitesimal value, is a standard normalization function.

2. The GIS enclosure production line real-time monitoring method of claim 1, wherein, The adaptive time step satisfies the relationship: ; in, It is the first Adaptive time step for each target pixel It is a preset fixed-size time step for each target pixel. It is the first Time step correction factor for each target pixel.

3. A GIS housing production line real-time monitoring system, characterized in that, The GIS shell production line real-time monitoring system comprises a processor and a memory, and the memory stores computer program instructions which, when executed by the processor, implement the GIS shell production line real-time monitoring method according to claim 1 or 2.

Images