Oil pipe inner wall defect detection method based on machine vision

Abstract

This application relates to the field of image processing, and more particularly to a machine vision-based method for detecting defects on the inner wall of an oil pipe. The method includes the following steps: obtaining a grayscale image of the inner wall of the oil pipe and an RGB image of a thermal image; calculating the iteration requirement for each pixel; and calculating the number of adaptive iterations for each pixel based on the iteration requirement. Based on the adaptive iteration number, a filtering algorithm is used to filter the grayscale image of the inner wall of the oil pipe before performing defect detection. By dynamically setting the number of iterations in the filtering algorithm, the present application improves denoising quality and efficiency.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method for detecting defects on the inner wall of an oil pipe based on machine vision. Background Art

[0002] During use, oil and gas pipelines may develop defects such as cracks due to the influence of the external environment and long-term loads. If these defects are not discovered and repaired in a timely manner, they may lead to serious accidents. Therefore, it is necessary to detect defects on the inner wall of the oil pipeline. Existing technology can detect defects on the inner wall of oil and gas pipelines through image detection technology. The image can be collected by a robot carrying a high-resolution camera into the interior of the pipeline or by inserting the camera into the pipeline using an endoscope to capture the image of the inner wall. The image is captured in real time and transmitted to an external device for analysis. For example, the Chinese patent with patent publication number CN116468674A and a method and system for detecting defects on the inner wall of a pipeline, which relates to oil and gas pipeline inspection technology, provides a detection method that uses image recognition and processing to detect defects.

[0003] When detecting cracks on the inner wall of an oil pipe, image quality directly impacts crack recognition accuracy. Images of the inner wall often contain noise, which can interfere with crack detection. To denoise these images, the Perona-Malik filter algorithm can be used to filter the inner wall image.

[0004] When the Perona-Malik filter algorithm denoises an image, it usually sets a fixed number of iterations for each pixel. However, the noise levels and detailed features in different areas of the image vary greatly. For flat areas in the image, too many iterations may lead to a waste of computing resources. Although the denoising effect in these areas may benefit, not having the necessary number of iterations will lead to unnecessary computing overhead and reduce processing efficiency. For the edge areas of the image, a uniform number of iterations may cause the boundaries of potential crack defects to become blurred, making it difficult to accurately identify the potential crack location and morphology later. For areas with high noise intensity, a uniform number of iterations may not be able to completely remove the noise effect. In summary, the existing technology cannot dynamically adjust the number of iterations according to the characteristics of different areas of the image, resulting in problems such as waste of computing resources, loss of edge details, and incomplete noise removal. Summary of the Invention

[0005] In order to solve the technical problem of how to dynamically set the number of iterations in a filtering algorithm, the present application provides a method for detecting defects on the inner wall of an oil pipe based on machine vision.

[0006] In a first aspect, the present application provides a method for detecting defects on the inner wall of an oil pipe based on machine vision, which adopts the following technical solutions:

[0007] A method for detecting defects on the inner wall of an oil pipe based on machine vision includes the following steps: obtaining a grayscale image of the inner wall of the oil pipe and an RGB image of a thermal image; calculating the iteration requirement of each pixel; calculating the number of adaptive iterations for each pixel based on the iteration requirement; and using a filtering algorithm to filter the grayscale image of the inner wall of the oil pipe according to the number of adaptive iterations before performing defect identification.

[0008] The method for calculating the degree of iteration requirement is as follows: calculate the degree of distortion of the grayscale value of each pixel in the grayscale image; set a reference area with each pixel on the grayscale image as the center, and calculate the information entropy of the reference area with respect to the target color channel in the corresponding area on the RGB image; calculate the discrete degree of the channel value in the target color channel according to the information entropy, calculate the possibility of cracks in the area where each pixel is located according to the information entropy and the discrete degree, normalize the negative correlation of the product of the information entropy and the possibility of cracks and adjust the value to obtain the possibility that the reference area where the pixel is located is a defective area; normalize the product of the degree of distortion and the possibility of the reference area where each pixel is located to be positively correlated and use it as the degree of iteration requirement.

[0009] Optionally, the method for calculating the degree of distortion is: calculating the difference between the maximum and minimum grayscale values ​​in the grayscale image; calculating the ratio of the grayscale value of each pixel to the difference, calculating the absolute difference between the ratio and a preset deviation factor and using it as the grayscale relative value of the pixel; calculating the first variance of the set consisting of the grayscale values ​​of each pixel and all pixels in its neighborhood; calculating the second variance of the set consisting of the grayscale values ​​of all pixels in the neighborhood of each pixel; normalizing the difference between the first variance and the second variance to obtain the variance contribution; and normalizing the product of the grayscale relative value and the variance contribution to obtain the degree of grayscale distortion of each pixel.

[0010] Optionally, the crack possibility is calculated by taking the difference between the maximum value and the minimum value of any pixel point in the reference area as the crack possibility.

[0011] Optionally, the product of information entropy and crack possibility is negatively correlated and normalized, and the value is adjusted to obtain the possibility that the reference area where the pixel point is located is a defective area. The value adjustment method is: setting a value adjustment factor, and adding the value adjustment factor and the result of normalizing the negative correlation of the product of information entropy and crack possibility.

[0012] Optionally, the destination color channel is the R channel.

[0013] Optionally, the adaptive number of iterations for each pixel is calculated based on the iteration requirement, including: taking the product of a preset fixed number of iterations and the iteration requirement as the initial number of iterations, and rounding the initial number of iterations to obtain the adaptive number of iterations.

[0014] Optionally, the rounding operation includes rounding, rounding up, rounding down, or truncating.

[0015] In a second aspect, the present application provides a machine vision-based oil pipe inner wall defect detection system, which adopts the following technical solutions:

[0016] The oil pipe inner wall defect detection system based on machine vision includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the oil pipe inner wall defect detection method based on machine vision is implemented.

[0017] The beneficial effect is that the above-mentioned oil pipe inner wall defect detection method based on machine vision is generated into a computer program and stored in a memory so as to be loaded and executed by a processor, thereby making a system based on the memory and the processor convenient to use.

[0018] This application has the following technical effects:

[0019] 1. The introduction of adaptive iterations dynamically adjusts the number of iterations based on the local characteristics of each pixel. In flat areas and image edges, the number of iterations is reduced, effectively avoiding unnecessary computational overhead while preserving image details for subsequent detection of potential cracks. In areas with severe noise, the number of iterations is increased for more effective noise removal. Adaptive calculation based on the iteration requirements of each pixel also improves processing efficiency and avoids wasted resources.

[0020] 2. By analyzing the noise characteristics of the grayscale image and the temperature distribution characteristics of the thermal image, the iteration requirement for each pixel is comprehensively calculated and the number of iterations is adaptively adjusted accordingly. Thermal images can reveal abnormal temperature distribution on the inner surface of the oil pipe, especially the characteristics of cracks or defects. This improves the denoising effect while preserving key image details (such as crack edges), providing high-quality image data support for subsequent crack detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of a method for detecting inner wall defects of oil pipes based on machine vision according to an embodiment of the present application.

[0022] Figure 2 This is a method flow chart of step S2 in the oil pipe inner wall defect detection method based on machine vision in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The present application discloses a machine vision-based method for detecting inner pipe wall defects. Prior to detecting potential crack defects on the inner pipe wall, the Perona-Malik filtering algorithm is used to denoise the inner pipe wall image. During this process, the number of Perona-Malik filtering iterations for each pixel is adaptively adjusted based on an analysis of the grayscale image and the characteristics of each pixel in the thermal image. The Perona-Malik Diffusion (PMD) filtering algorithm, an image smoothing method based on partial differential equations, can be used to filter inner pipe wall images to remove noise and preserve image edge information, facilitating the identification of inner pipe wall defects (such as cracks).

[0024] Reference Figure 1 The oil pipe inner wall defect detection method based on machine vision includes steps S1 to S3, which are specifically as follows:

[0025] S1: Obtain a grayscale image of the inner wall of the oil pipe and an RGB image of the thermal image.

[0026] In one embodiment, the present application uses an integrated thermal imaging and visible light camera to photograph the inner wall of an oil pipeline, obtaining a high-definition visible light image and a thermal image of the inner wall. The high-definition visible light image is grayscaled to obtain a grayscale image of the inner wall. The thermal image is pseudo-color processed to obtain an RGB (Red, Green, Blue) image. Existing techniques are not described in detail here.

[0027] S2: Calculate the iteration requirement of each pixel.

[0028] Reference Figure 2 The method for calculating the iteration requirement degree includes steps S20-S21, which are as follows:

[0029] S20: Calculate the degree of distortion of the grayscale value of each pixel in the grayscale image.

[0030] The calculation method of the grayscale value distortion degree is as follows: calculate the difference between the maximum and minimum grayscale values ​​in the grayscale image; calculate the ratio of the grayscale value of each pixel to the difference, calculate the absolute difference between the ratio and the preset deviation factor and use it as the grayscale relative value of the pixel; calculate the first variance of the set consisting of the grayscale values ​​of each pixel and all pixels in its neighborhood; calculate the second variance of the set consisting of the grayscale values ​​of all pixels in the neighborhood of each pixel; normalize the difference between the first variance and the second variance to obtain the variance contribution; normalize the product of the grayscale relative value and the variance contribution to obtain the grayscale value distortion degree of each pixel.

[0031] In one embodiment, the mathematical expression of the grayscale value distortion degree may be:

[0032] Where Indicates the The degree of grayscale distortion of each pixel, Indicates the The absolute value of the difference between the gray value of a pixel and the minimum gray value in the image, Represents the difference between the maximum grayscale value and the minimum grayscale value in the image. Indicates the The relative size of the grayscale value of a pixel in the image. The larger or smaller the value is, the more extreme the grayscale value of a pixel is in the image. This is an example value of the preset deviation factor. You can adjust this value according to the actual application scenario. This application prefers 0.5. 0.5 is selected as the deviation factor because it represents the middle position of the grayscale value. By calculating the degree of deviation between the grayscale value of the pixel and the middle value (0.5), you can more intuitively evaluate whether it is an extreme value. Grayscale relative value The larger the The more extreme the relative size of the grayscale value of a pixel in the image is, the more likely it is to be affected by noise, and the greater the degree of distortion of the corresponding grayscale value.

[0033] Indicates the The first variance of the grayscale value of a pixel and all pixels in its 8 neighborhoods (including pixels), Indicates the The second variance of the set of gray values ​​of all pixels in the neighborhood of 8 pixels (excluding the pixels). Represents the standard normalization function, and the prior art will not be described in detail. Indicates the The contribution of a pixel to the variance of the grayscale value of the pixels in the neighborhood. The larger the value, the greater the contribution of the pixel to the variance of the grayscale value of the pixels in the neighborhood. The more extreme the grayscale value of a pixel, the greater its credibility, indicating that it is more likely to be affected by noise, and the corresponding grayscale value distortion is greater. The neighborhood of this application can be selected as 8 neighborhoods or 16 neighborhoods, etc., and 8 neighborhoods are preferred because they can capture the texture changes of the local area while quickly calculating statistics such as variance, with low computational complexity. At the same time, if the neighborhood range is too large, the real crack features may be misjudged as noise; and if the neighborhood range is too small, it may not be able to effectively distinguish between noise and real details. The size of 8 neighborhoods is moderate, which can effectively remove noise interference while retaining the real defect features.

[0034] S21: Set a reference area with each pixel on the grayscale image as the center, and calculate the information entropy of the reference area with respect to the target color channel in the corresponding area on the RGB image; calculate the discreteness of the channel value in the target color channel based on the information entropy, and calculate the possibility of cracks in the area where each pixel is located based on the information entropy and the discreteness, normalize the negative correlation between the product of the information entropy and the possibility of cracks, and adjust the value to obtain the possibility that the reference area where the pixel is located is a defective area; normalize the positive correlation between the product of the distortion degree and the possibility of the reference area where each pixel is located and use it as the iteration requirement.

[0035] Relying solely on the degree of grayscale distortion of a pixel cannot fully determine the number of iterations required for filtering a pixel. This is because when a pixel that is most affected by noise (i.e., has a greater degree of grayscale distortion) is located in an area with complex textures or in an area prone to cracks on the inner wall of the oil pipe, in order to ensure clearer image data support for subsequent potential crack defect detection, it is necessary to balance the degree of grayscale distortion and the image area to which the pixel belongs, and set a more reasonable number of iterations for filtering.

[0036] Specifically, the color characteristics of each pixel in the thermal imaging image of the inner wall of the oil pipe are analyzed, and the degree of grayscale distortion of each pixel is combined to determine the iteration requirement for each pixel. Therefore, when analyzing the iteration requirement for each pixel, the greater the likelihood that the region to which each pixel belongs belongs to an area with complex texture in the image or an area on the inner wall of the oil pipe that is prone to cracks, the fewer iterations need to be set for this pixel to ensure that its detailed information can be retained as much as possible during filtering. Conversely, the less likely a pixel belongs to an area with complex texture in the image or an area on the inner wall of the oil pipe that is prone to cracks, the more iterations need to be set for this pixel to ensure that noise can be effectively removed.

[0037] When analyzing the likelihood that a pixel belongs to a complex texture region in an image or a crack-prone region on the inner wall of an oil pipe, the different areas of the inner wall of the oil pipe may be affected by the external environment, pressure, fluid flow rate, or oil properties, resulting in uneven heat distribution. Thermal images of complex texture regions and crack-prone areas typically exhibit large temperature gradients, with significant local differences. Therefore, the more discrete the numerical distribution of pixels in the R channel within a pixel reference area in a thermal image, the greater the likelihood that it belongs to a complex texture region or a crack-prone region, and the correspondingly smaller the number of iterations required, requiring fewer iterations to preserve as much detail as possible. Conversely, the greater the number of iterations required, the smaller the number of iterations required.

[0038] In one embodiment, a 15×15 pixel area around each pixel being analyzed is selected as a reference area. The size of the reference area can be adaptively adjusted according to actual conditions.

[0039] In one embodiment, the target color channel can be the R channel in the RGB image, and the method for calculating the information entropy is a prior art and will not be repeated here. The R channel can directly reflect the distribution characteristics of the high-temperature area in the thermal imaging image, and the high-temperature area is closely related to the crack or defect area on the inner wall of the oil pipe. The discrete degree of the numerical distribution of the R channel can effectively determine whether the pixel point is located in an area with complex texture or an area prone to cracks. The G channel is suitable for analyzing the intermediate temperature area, and the B channel is suitable for analyzing the low-temperature area, and its applicability in crack detection is relatively weak. The information entropy mean of the three RGB channels can also be taken to replace the information entropy of the R channel to participate in the calculation, but it will increase the computational complexity. Therefore, in this application, the preferred R channel not only meets the needs of crack detection, but also maintains the efficiency and simplicity of the algorithm.

[0040] In one embodiment, the crack probability is calculated by taking the difference between the maximum and minimum values ​​of any pixel within the reference region as the crack probability. The crack probability intuitively reflects the degree of temperature dispersion and the magnitude of the temperature gradient within the reference region. Cracks or regions with complex textures typically exhibit large temperature gradients, so the larger the difference, the higher the crack probability.

[0041] In one embodiment, the mathematical expression for the probability of a defective area is:

[0042] Where Indicates the The degree of iteration required for each pixel, Indicates the The degree of grayscale distortion of each pixel. Indicates the The information entropy value of the reference area of ​​​​the pixel point in its corresponding thermal imaging image about the R channel value, Indicates the The maximum value of the R channel value in the corresponding thermal imaging image of the reference area of ​​the pixel point is Indicates the The reference area of ​​​​the pixel points has the minimum value of the R channel value in its corresponding thermal imaging image. Indicates An exponential function with base .

[0043] The larger the When analyzing the grayscale features of each pixel in the image, the greater the possibility of being affected by noise, the more iterations are needed to denoise it, so the corresponding iteration requirement is greater. Indicates the The probability that the area to which the pixel belongs belongs to the complex texture area in the image or the area where the inner wall of the oil pipe is prone to cracks is high. The larger the value, the higher the probability that the pixel belongs to. The area to which the pixel belongs should be given more attention. The smaller the number of iterations for each pixel, the more detailed information can be retained during filtering, so the less iteration is required. The larger the The more discrete the distribution of the R channel value of the reference area of ​​the pixel point in the corresponding thermal imaging image, the more discrete the distribution of the R channel value of the reference area of ​​the pixel point in the corresponding thermal imaging image. The greater the possibility that the reference area of ​​the pixel belongs to the complex texture area in the image or the area where the inner wall of the oil pipe is prone to cracks, the higher the probability that the reference area of ​​the pixel belongs to the complex texture area in the image or the area where the inner wall of the oil pipe is prone to cracks. The smaller the number of pixels, the less iteration is required; The larger the The more discrete the distribution of the R channel value of the reference area of ​​the pixel point in the corresponding thermal imaging image, the greater the credibility. The greater the possibility that the reference area of ​​the pixel belongs to the complex texture area in the image or the area where the inner wall of the oil pipe is prone to cracks, the higher the probability that the reference area of ​​the pixel belongs to the complex texture area in the image or the area where the inner wall of the oil pipe is prone to cracks. The more pixels there are, the less iteration is required.

[0044] The value adjustment factor that represents the degree of iteration required for each pixel point can be adjusted by adjusting value, ensure The value range of always falls within a reasonable range. For example, , and finally the iteration requirement range of each pixel will be adjusted to . A value of 0.5 ensures The lower limit is 0.5 to avoid the calculated result being too small. The introduction of can better balance the denoising effect and detail preservation, which can be achieved by fine-tuning The value of flexibly adapts to image features in different scenarios.

[0045] S3: The adaptive iteration number of each pixel is calculated based on the degree of iteration requirement. According to the adaptive iteration number, the grayscale image of the inner wall of the oil pipe is filtered using a filtering algorithm and then defect identification is performed.

[0046] The greater the degree of iteration required for a pixel, the more affected it is by noise, the simpler the texture of the image region it is in, and the more likely it is a region where cracks are less likely to occur on the inner wall of the oil pipe. Therefore, a greater number of iterations is required to ensure noise removal. Conversely, the smaller the degree of iteration required for a pixel, the less affected it is by noise, the more complex the texture of the image region it is in, and the more likely it is a region where cracks have already occurred on the inner wall of the oil pipe. Therefore, fewer iterations are required to save computing resources and preserve detailed information.

[0047] The product of the preset fixed number of iterations and the iteration requirement is used as the initial number of iterations, and the initial number of iterations is rounded to obtain the adaptive number of iterations. Exemplary rounding operations include rounding, rounding up, rounding down, or truncation. The prior art will not be described in detail here. This application prefers rounding.

[0048] In one embodiment, the mathematical expression of the adaptive iteration number can be:

[0049] Where Indicates the The number of iterations after pixel adaptation, Indicates the number of iterations of the algorithm with a fixed size preset for all pixels. For example, , Indicates the The iteration requirement of each pixel is finally used The number of iterations after each pixel adaptation is rounded up. Adding 0.5 and then rounding down can achieve the rounding effect.

[0050] After obtaining the number of adaptive iterations, the image of the inner wall surface of the oil pipe is filtered according to the existing steps of the Perona-Malik filtering algorithm. The number of iterations for each pixel can be calculated according to the core logic of this solution. The contrast parameter corresponding to each pixel is preset to an empirical value of 15, and the time step corresponding to each pixel is preset to an empirical value of 0.1.

[0051] After completing the denoising process on the oil pipe inner wall surface image, a CNN (Convolutional Neural Networks) model can be used to train and analyze the oil pipe inner wall surface image to automatically learn the characteristics of potential crack defect areas on the oil pipe inner wall surface and identify potential crack defects. The existing technology will not be repeated here.

[0052] An embodiment of the present application also discloses a machine vision-based oil pipe inner wall defect detection system, including a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the machine vision-based oil pipe inner wall defect detection method according to the present application is implemented.

[0053] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0054] In this application, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory, a dynamic random access memory, a static random access memory, or the like, or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of, accessible to, or connectable to a device.

[0055] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A method for detecting inner wall defects of oil pipes based on machine vision, characterized in that: The method includes the following steps: obtaining a grayscale image of the inner wall of the oil pipe and an RGB image of the thermal image; calculating the iteration requirement of each pixel point; calculating the number of adaptive iterations of each pixel point based on the iteration requirement; and using a filtering algorithm to filter the grayscale image of the inner wall of the oil pipe according to the number of adaptive iterations and then performing defect identification. The calculation method for the degree of iteration requirement is: Calculate the degree of distortion of the grayscale value of each pixel in the grayscale image; A reference area is set with each pixel on the grayscale image as the center, and the information entropy of the reference area with respect to the target color channel in the corresponding area on the RGB image is calculated. The discreteness of the channel value in the target color channel is calculated based on the information entropy. The crack probability of the area where each pixel is located is calculated based on the information entropy and the discreteness. The product of the information entropy and the crack probability is normalized and the value is adjusted to obtain the probability that the reference area where the pixel is located is a defective area. The probability refers to the probability that the reference area where the pixel is located is a defective area. The product of the distortion degree and the probability of each pixel in the reference area is normalized and used as the iteration requirement; Calculate the difference G between the maximum and minimum grayscale values ​​in the grayscale image; Calculate the absolute value Gn of the difference between the grayscale value of each pixel and the minimum grayscale value in the grayscale image, calculate the ratio of the absolute value of the difference Gn to the difference G, calculate the absolute difference between the ratio and the preset deviation factor and use it as the relative grayscale value of the pixel; calculate the first variance of the set consisting of the grayscale values ​​of each pixel and all pixels in its neighborhood; Calculate the second variance of the set of grayscale values ​​of all pixels in the neighborhood of each pixel; normalize the difference between the first variance and the second variance to obtain the variance contribution; The degree of distortion of the grayscale value of each pixel is obtained by normalizing the product of the grayscale relative value and the variance contribution; The calculation method of crack possibility is: the difference between the maximum and minimum values ​​of any pixel point in the reference area is used as the crack possibility; Calculate the number of adaptive iterations for each pixel based on the degree of iteration required, including: The product of the preset fixed number of iterations and the iteration requirement degree is used as the initial number of iterations, and the initial number of iterations is rounded to obtain the adaptive number of iterations.

2. The method for detecting inner wall defects of oil pipes based on machine vision according to claim 1, characterized in that: After normalizing the negative correlation between the product of information entropy and the possibility of cracks and adjusting the value, the possibility that the reference area where the pixel point is located is a defect area is obtained. The value adjustment method is: setting a value adjustment factor, and adding the value adjustment factor and the result of normalizing the negative correlation between the product of information entropy and the possibility of cracks.

3. The method for detecting inner wall defects of oil pipes based on machine vision according to claim 1, characterized in that: The destination color channel is the R channel.

4. The method for detecting inner wall defects of oil pipes based on machine vision according to claim 1, characterized in that: Rounding operations include rounding, ceiling, floor, or truncation.

5. The oil pipe inner wall defect detection system based on machine vision is characterized by: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the method for detecting inner wall defects of an oil pipe based on machine vision according to any one of claims 1 to 4 is implemented.

Images