Gis defect detection method and apparatus

Abstract

The application discloses a GIS defect detection method and device, which comprises the following steps of: obtaining an original image I Original of GIS equipment through X-ray detection of the GIS equipment grqy , performing gray processing on the original image I grqy to obtain a gray image I grqy , removing noise of the gray image I blur by adopting Gaussian blur processing to obtain a blurred image I blur , calculating horizontal and vertical degrees of the blurred image I blur by using a Sobel operator, and finally performing pixel value adjustment processing on the blurred image I target to obtain a target image I target ; performing edge extraction on the target image I target by using a Canny edge detection algorithm, performing a step of fitting a center line of a pair of pixels by using a least square method, and determining a current defect type by judging a relationship among a curvature k (x), a curvature change rate k i and a bifurcation number P. According to the GIS equipment, the present application analyzes possible defect types, and proposes a method combining the center line curvature, the curvature change and the bifurcation to identify the defect type, so that the efficiency is greatly improved compared with artificial identification and judgment by checking the images one by one.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of GIS device detection, in particular to the technical field of image post-processing after GIS device X-ray detection, and more particularly to the technical field of detecting whether a defect exists in an X-ray detection image analysis device and the type of defect, and specifically to a GIS defect detection method and device. BACKGROUND

[0002] In recent years, X-ray detection-based gas-insulated electrical equipment (GIS) defect inspection technology has been widely applied. GIS is widely used in high-voltage power systems due to its excellent insulation performance and compact design. However, due to the complex structure and working environment of the equipment, various defects may occur in GIS during long-term operation, such as gas leakage, insulation material aging, and mechanical damage. As a non-destructive testing (NDT) technology, X-ray detection can effectively penetrate the internal structure of the equipment and provide high-resolution images, thereby achieving early identification of potential defects. Currently, many power companies and detection agencies have integrated X-ray detection into the regular maintenance process to improve the reliability and safety of the equipment. In addition, with the development of image processing technology, automated defect recognition systems have gradually emerged, improving detection efficiency and accuracy. However, despite the continuous progress of technology, X-ray detection still faces some challenges in practical application, such as radiation safety, detection equipment cost, and professional training of operators, which affect its widespread promotion and application.

[0003] Analyzing the defect types in X-ray detection images is of great significance to ensuring the safety and reliability of GIS equipment. By identifying and classifying different types of defects such as bubbles, cracks, corrosion, and material inhomogeneity, maintenance personnel can develop targeted maintenance strategies to reduce equipment failure rates and extend equipment service life. In addition, defect type analysis can also provide a reference for equipment design improvement and material selection. However, existing technologies still face many challenges in defect analysis. First, image noise and artifacts may affect the accuracy of defect identification, making subsequent analysis difficult. Second, current automated defect recognition systems often rely on traditional image processing algorithms, which are difficult to cope with complex defect types and diverse equipment structures. In addition, the lack of sufficient labeled data and effective machine learning models also limits the application range of intelligent analysis technology. Therefore, improving the precision and robustness of image processing and analysis technology is a key direction for future research.

[0004] Prior art: patent document, publication number CN117740819B, subject name as a kind of GIS tank defect detection system and detection method provides a specific system for GIS tank defect detection and a method for judging the severity of defects, which can only detect the GIS tank, and can only perform qualitative detection on defects, and cannot further judge the defect type. In order to quickly and accurately analyze the X-ray detection image of GIS equipment, the present application is completed. SUMMARY

[0005] In order to solve the problems of defect detection, identification and classification in the post-processing of GIS equipment X-ray detection image, the present application provides a GIS defect detection method and device for automatically identifying and analyzing X-ray detection image and qualitatively classifying defect type, which can save a lot of manual film reading and manual screening time, and greatly improve the detection post-processing efficiency.

[0006] The present application can directly input the X-ray original image obtained by detection without image preprocessing, and directly judge the data by the detection method provided by the present application, so that the defect discrimination efficiency is high and a large number of image processing can be met.

[0007] The present application adopts unique curvature and bifurcation discrimination, and has high classification accuracy for the defect types often appearing in GIS equipment defects, and the misjudgment and omission probability is very low.

[0008] In order to achieve the above purpose, the technical scheme adopted by the present application is:

[0009] The GIS defect detection method provided by the present application comprises the following detection steps:

[0010] Step STP100, the original image of X-ray detection of GIS equipment is subjected to gray scale processing to obtain a gray scale image , The gray scale image is subjected to Gaussian blur processing to remove noise and obtain a blurred image , The horizontal and vertical degrees of the blurred image are calculated by using Sobel operator , Finally, the blurred image is subjected to pixel value adjustment processing to obtain a target image ;

[0011] Step STP200, the target image is subjected to edge extraction by using Canny edge detection algorithm to obtain a plurality of edge objects , , … Among them n≥1;

[0012] Step STP300 involves taking each edge object obtained in step STP200. , … The least squares method is used to fit the center lines of paired pixels to obtain the center lines of the edge objects. , … Corresponding multiple center lines , ,in, x, y For target image Pixel coordinates;

[0013] Step STP400 involves taking each centerline obtained in step STP300. , Progress Curvature k ( x ) and rate of change of curvature Δ calculate;

[0014] Step STP500 involves determining the curvature. k ( x ) and set bifurcation threshold The relationship between the two determines the current centerline. Number of branches P By judging curvature k ( x ), rate of change of curvature Δ and number of branches P The relationship between the two determines the type of defect currently detected:

[0015] The defect is hollowness: satisfying the curvature k ( x ) < K, and the actual number of bifurcations P=0 K is the preset curvature value;

[0016] Defects are impurities: rate of change of curvature Δ < and < , i ∈ n ;

[0017] The defect is a fatigue crack: rate of change of curvature Δ , And P∈[1,3], i ∈ n ;in, It is the threshold of the rate of change of curvature. It is the threshold for the cumulative change in curvature;

[0018] The defect is fracture: satisfying the curvature k ( x )>K, and the actual number of bifurcations P≥3.

[0019] To improve the accuracy of subsequent edge recognition of defect areas, preferably, the Sobel operator is used to calculate the blurred image in step STP100. The method for determining horizontal and vertical dimensions involves processing a blurred image after Gaussian filtering. The Sobel operator uses two convolution kernels to calculate the gradients in the horizontal and vertical directions, respectively. Specifically, the horizontal Sobel operator... Represented as:

[0020] =

[0021] Vertical Sobel operator Represented as:

[0022] =

[0023] Each convolution kernel is compared with the blurred input image. After performing convolution, the gradients in the horizontal and vertical directions are calculated as follows:

[0024] Horizontal gradient Represented as:

[0025] Vertical gradient Represented as:

[0026] in, It is the input blurred image exist The pixel values; also including the gradient magnitude and direction calculated from the horizontal and vertical gradients, as follows:

[0027] gradient magnitude Represented as = ;

[0028] The gradient direction θ is denoted as θ = .

[0029] To further improve the accuracy of defect identification and extraction, preferably, in step STP100, the blurred image is... Pixel value adjustment processing is performed to obtain the target image. The process includes steps such as adjusting the gradient magnitude of pixel values ​​and pixel classification, specifically including:

[0030] Gradient magnitude adjustment: Adjust the gradient direction θ of any pixel. Convert the pixel to discrete directional angles of 0° / 45° / 90° / 135°, and then compare the gradient magnitudes of adjacent pixels based on the direction of each pixel. ,like If the value is greater than that of the adjacent pixels, retain the gradient magnitude of the current pixel; otherwise, set it to 0, and iterate through the entire blurred image. Pixels;

[0031] Pixel classification: Setting a high gradient threshold and low gradient threshold Based on the current pixel gradient magnitude With high gradient threshold and low gradient threshold The relationship between edges determines the edge type:

[0032] satisfy > Then Classified as a strong edge;

[0033] satisfy > > Then Classified as weakly marginal;

[0034] satisfy < Then Image pixels classified as non-edge.

[0035] The present invention also provides a detection device, including a processor and a memory, wherein the memory stores a program for executing the GIS defect detection method described above, and the processor is used to call the program in the memory to execute the GIS defect detection method.

[0036] Beneficial effects:

[0037] 1. This invention analyzes the possible defect types of existing GIS equipment and proposes a method that combines centerline curvature, curvature changes, and bifurcation to identify defect types. Compared with manual image review and judgment, this method is much more efficient and can handle the defect identification task of massive inspection images, providing a good solution for image post-processing tasks of X-ray inspection of GIS equipment.

[0038] 2. Before identifying defects in the detected image, this invention performs pixel-level processing on the head image, especially quantitative identification and processing of pixel gradients and directions, which can greatly improve the efficiency and accuracy of defect processing. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are some embodiments of this application, but not all embodiments.

[0040] Example 1:

[0041] This embodiment provides a GIS defect detection method, including the following detection steps:

[0042] Step STP100: The raw image from the X-ray inspection of the GIS equipment. Obtain a grayscale image by performing grayscale processing. grayscale image Use Gaussian blur to remove grayscale images Obtaining a blurred image from noise Then, the Sobel operator is used to calculate the blurred image. The horizontal and vertical dimensions are then used to blur the image. Pixel value adjustment processing is performed to obtain the target image. ;

[0043] Step STP200 involves using the Canny edge detection algorithm on the target image. Perform edge extraction to obtain multiple edge objects. , … ,in n ≥1;

[0044] Step STP300 involves taking each edge object obtained in step STP200. , … The least squares method is used to fit the center lines of paired pixels to obtain the center lines of the edge objects. , … Corresponding multiple center lines , ,in, x, y For target image Pixel coordinates;

[0045] Step STP400 involves taking each centerline obtained in step STP300. , Progress Curvature k ( x ) and rate of change of curvature Δ calculate;

[0046] Step STP500 involves determining the curvature. k (x ) and set bifurcation threshold The relationship between the two determines the current centerline. Number of branches P By judging curvature k ( x ), rate of change of curvature Δ and number of branches P The relationship between the two determines the type of defect currently detected:

[0047] The defect is hollowness: satisfying the curvature k ( x ) < K, and the actual number of bifurcations P=0 K is the preset curvature value;

[0048] Defects are impurities: rate of change of curvature Δ < and < , i ∈ n ;

[0049] The defect is a fatigue crack: rate of change of curvature Δ , And P∈[1,3], i ∈ n ;in, It is the threshold of the rate of change of curvature. It is the threshold for the cumulative change in curvature;

[0050] The defect is fracture: satisfying the curvature k ( x )>K, and the actual number of bifurcations P≥3.

[0051] In this embodiment, steps STP300-STP500 are particularly important parts of the invention. Compared to existing manual image interpretation or computer recognition, they offer advantages in efficiency and accuracy. Specifically, this embodiment combines curvature, rate of change of curvature, whether there is bifurcation, and the number of bifurcations for comprehensive judgment. Traditional defect recognition methods mainly rely on threshold-based segmentation techniques and template matching, which are often sensitive to image noise and lighting changes, resulting in low recognition accuracy. This embodiment utilizes edge extraction technology, especially edge detection based on Canny or Sobel operators, to effectively capture detailed defect information, particularly against complex backgrounds. By accurately identifying the edges of defects, subsequent centerline fitting can more accurately reflect the shape characteristics of the defects, thereby improving overall recognition accuracy. Furthermore, existing defect recognition methods typically focus on simple features, such as the area or length of the defect, which limits in-depth analysis of the defect's essence. By fitting the centerline of the defect and analyzing its curvature, curvature changes, and bifurcation, richer geometric features can be extracted. These characteristics not only help identify the type of defect, but also reflect the defect's generation mechanism and development trend, providing an important basis for subsequent fault prediction and equipment maintenance. For example, fatigue cracks will extend with the equipment's usage time and operating conditions. Based on the existing defect situation and type, the possible failure time can be scientifically inferred. This allows for advance prediction and maintenance or replacement of equipment before the failure is likely to occur, thus avoiding power outages caused by sudden equipment failures, mitigating unnecessary power loss losses, and preventing power outage accidents.

[0052] While some computer-based classification software exists in the existing technology, these identification programs largely rely on machine learning models, typically requiring a large amount of labeled data for training. Furthermore, their classification effectiveness is limited by the rationality of feature selection. On the other hand, model training often remains theoretical. In practical applications, especially for training on inspection images of GIS equipment, it's nearly impossible to accumulate enough image samples to train a mature model. Therefore, limited by the number of samples, even with optimal training modes, mechanisms, and parameters, it's almost impossible to obtain a mature model capable of autonomously determining the existence and type of defects in GIS equipment. This embodiment, however, constructs a more refined classification standard by analyzing the curvature and number of bifurcations of the defect centerline. For example, different types of cracks, voids, or impurities exhibit significant differences in curvature characteristics, allowing for effective classification by setting thresholds. This method not only reduces reliance on training data but also adapts more quickly to new defect types in practical applications. In real-world applications, GIS equipment may contain multiple defects simultaneously, and the morphology and distribution of these defects may influence each other. Traditional methods often struggle to handle such complex defects, easily leading to identification errors or omissions. By combining edge extraction and centerline fitting, multiple defects can be effectively distinguished and analyzed. During the analysis, the relationships between different defects (such as the number and location of bifurcation points) can be detected to determine their mutual influence, thus providing a more comprehensive fault analysis. Currently, many defect detection methods rely on human experience for defect identification and classification, which is easily influenced by subjective factors. Automated processing through edge extraction and centerline fitting can significantly reduce the need for human intervention, improving detection efficiency and consistency. This not only improves the reliability of detection but also reduces the consumption of human resources during large-scale inspections, enhancing overall work efficiency.

[0053] Finally, the edge extraction and centerline fitting method exhibits good scalability, adapting to different types of GIS equipment and defect features. In existing technologies, many methods are highly dependent on specific equipment or defect types, while the detection method provided in this embodiment, by extracting geometric features, can be easily transferred to other fields or other types of equipment, possessing strong compatibility and application potential. This makes the method widely applicable in future equipment maintenance and inspection. Based on the implementation of this invention, a foundation is laid for establishing a map library of GIS equipment status for the entire industry. Because this invention can quickly and autonomously classify defects in detected images, massive amounts of data can be clustered and analyzed, allowing the detection images of all GIS equipment nationwide to be incorporated into the map library. Real-time previews of the current status of all in-service GIS equipment in the entire industry can be achieved, which has an immeasurable and significant role and practical importance in maintaining the entire power industry.

[0054] Example 2:

[0055] This embodiment refines the detected image before defect identification, based on Embodiment 1, aiming to further improve the accuracy and efficiency of defect identification, analysis, and classification. Specifically, in this embodiment, the Sobel operator is used to calculate the blurred image in step STP100. The method for determining horizontal and vertical dimensions involves processing a blurred image after Gaussian filtering. The Sobel operator uses two convolution kernels to calculate the gradients in the horizontal and vertical directions, respectively. Specifically, the horizontal Sobel operator... Represented as:

[0056] =

[0057] Vertical Sobel operator Represented as:

[0058] =

[0059] Each convolution kernel is compared with the blurred input image. After performing convolution, the gradients in the horizontal and vertical directions are calculated as follows:

[0060] Horizontal gradient Represented as:

[0061] Vertical gradient Represented as:

[0062] in, It is the input blurred image exist The pixel values; also including the gradient magnitude and direction calculated from the horizontal and vertical gradients, as follows:

[0063] gradient magnitude Represented as = ;

[0064] The gradient direction θ is denoted as θ = .

[0065] Meanwhile, in step STP100, the blurred image Pixel value adjustment processing is performed to obtain the target image. The process includes steps such as adjusting the gradient magnitude of pixel values ​​and pixel classification, specifically including:

[0066] Gradient magnitude adjustment: Adjust the gradient direction θ of any pixel. Convert the pixel to discrete directional angles of 0° / 45° / 90° / 135°, and then compare the gradient magnitudes of adjacent pixels based on the direction of each pixel. ,like If the value is greater than that of the adjacent pixels, retain the gradient magnitude of the current pixel; otherwise, set it to 0, and iterate through the entire blurred image. Pixels;

[0067] Pixel classification: Setting a high gradient threshold and low gradient threshold Based on the current pixel gradient magnitude With high gradient threshold and low gradient threshold The relationship between edges determines the edge type:

[0068] satisfy > Then Classified as a strong edge;

[0069] satisfy > > Then Classified as weakly marginal;

[0070] satisfy < Then Image pixels classified as non-edge.

[0071] Of course, the method provided in this embodiment is only one of the preferred methods and is not exclusive. Those skilled in the art can still extend or combine it within the framework of the inventive concept disclosed in this invention.

[0072] Example 3:

[0073] This embodiment provides a detection device, including a processor and a memory. The memory stores a program for executing the GIS defect detection method described in any of the above embodiments, and the processor is used to call the program in the memory to execute the GIS defect detection method.

[0074] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A GIS defect detection method, characterized in that, The following testing steps are included: Step STP100: The raw image from the X-ray inspection of the GIS equipment. Grayscale image obtained by grayscale conversion grayscale image Use Gaussian blur to remove grayscale images Obtaining a blurred image from noise Then, the Sobel operator is used to calculate the blurred image. The horizontal and vertical dimensions are then used to blur the image. Pixel value adjustment processing is performed to obtain the target image. ; Step STP200 involves using the Canny edge detection algorithm on the target image. Perform edge extraction to obtain multiple edge objects. , … ,in n ≥1; Step STP300 involves taking each edge object obtained in step STP200. , … The least squares method is used to fit the center lines of paired pixels to obtain the center lines of the edge objects. , … Corresponding multiple center lines , ,in, x, y For target image Pixel coordinates; Step STP400 involves taking each centerline obtained in step STP300. , Progress Curvature k ( x ) and rate of change of curvature Δ calculate; Step STP500 involves determining the curvature. k ( x ) and set bifurcation threshold The relationship between the two determines the current centerline. Number of branches P By judging curvature k ( x ), rate of change of curvature Δ and number of branches P The relationship between the two determines the type of defect currently detected: The defect is hollowness: satisfying the curvature k ( x ) < K, and the actual number of bifurcations P=0 K is the preset curvature value; Defects are impurities: rate of change of curvature Δ < and < , i ∈ n ; The defect is a fatigue crack: rate of change of curvature Δ , And P∈[1,3], i ∈ n ;in, It is the threshold of the rate of change of curvature. It is the threshold for the cumulative change in the rate of curvature change; The defect is fracture: satisfying the curvature k ( x )>K, and the actual number of bifurcations P≥3.

2. The GIS defect detection method according to claim 1, characterized in that, In step STP100, the Sobel operator is used to calculate the blurred image. The method for determining horizontal and vertical dimensions involves processing a blurred image after Gaussian filtering. The Sobel operator uses two convolution kernels to calculate the gradients in the horizontal and vertical directions, respectively. Specifically, the horizontal Sobel operator... Represented as: = Vertical Sobel operator Represented as: = Each convolution kernel is compared with the blurred input image. After performing convolution, the gradients in the horizontal and vertical directions are calculated as follows: Horizontal gradient Represented as: Vertical gradient Represented as: in, It is the input blurred image exist The pixel values; also includes the gradient magnitude and direction calculated from the horizontal and vertical gradients, as follows: gradient magnitude Represented as = ; The gradient direction θ is denoted as θ = .

3. The GIS defect detection method according to claim 2, characterized in that, In step STP100, the blurred image will be... Pixel value adjustment processing is performed to obtain the target image. The process includes steps such as adjusting the gradient magnitude of pixel values ​​and pixel classification, specifically including: Gradient magnitude adjustment: Adjust the gradient direction θ of any pixel. Convert the pixel to discrete directional angles of 0° / 45° / 90° / 135°, and then compare the gradient magnitudes of adjacent pixels based on the direction of each pixel. If the gradient magnitude of the current pixel is greater than that of its neighboring pixels, then retain the gradient magnitude of the current pixel; otherwise, set it to 0. Repeat this process for the entire blurred image. Pixels; Pixel classification: Setting a high gradient threshold and low gradient threshold Based on the current pixel gradient magnitude With high gradient threshold and low gradient threshold The relationship between edges determines the edge type: satisfy > Then Classified as a strong edge; satisfy > > Then Classified as weakly marginal; satisfy < Then Image pixels classified as non-edge.

4. A detection device, comprising a processor and a memory, characterized in that, The memory stores a program for executing the GIS defect detection method according to any one of claims 1-3, and the processor is used to call the program in the memory to execute the GIS defect detection method.

Images