A thick-walled steel pipe weld detection method based on image recognition

Abstract

The present application relates to the technical field of image processing, and especially relates to a thick-walled steel pipe weld detection method based on image recognition. The present application uses coaxial light to vertically irradiate single weld surface, respectively determines weld ridge line and weld boundary line of each weld interval to divide the weld interval into a plurality of confidence intervals and a plurality of non-confidence intervals. It is determined whether to use low-angle grazing light to irradiate the confidence interval. It is determined whether the confidence interval is a defect interval based on edge defect characteristics. The type of non-confidence interval and the corresponding detection method are determined. Image fusion of the corresponding interval is carried out. It is determined whether the non-confidence interval is a defect interval based on the fusion image. It is determined whether the non-confidence interval is a defect interval based on the bevel defect characteristics. The defect rate of the confidence interval is determined to reduce the preset sliding distance or analyze the defect rate of the non-confidence interval. It is determined whether to reduce the preset distance. The present application improves the accuracy of thick-walled steel pipe weld detection based on image recognition.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method for detecting weld seams in thick-walled steel pipes based on image recognition. Background Technology

[0002] Thick-walled steel pipes are commonly used in high-pressure, high-temperature, or corrosive environments. The weld, as a weak point in the pipe structure, directly determines the safety and service life of the entire pipeline system. During the welding process of thick-walled steel pipes, to ensure joint strength and welding quality, multi-layer, multi-pass welding is typically employed. The bevel shapes are often V-shaped, U-shaped, or double-sided bevels, with the bevel depth and width increasing significantly with wall thickness. Due to the large wall thickness of thick-walled steel pipes, in the early and middle stages of welding, the filler weld has not yet completely filled the bevel, and the weld metal surface is significantly lower than the surface of the base material. The weld is actually located deep within the bevel, forming a narrow and deep internal space. Because the weld is inside the bevel, the line of sight of external industrial cameras or vision sensors is easily obstructed by the bevel sidewalls, limiting the imaging angle and making it difficult to simultaneously acquire images of the weld surface and the bevel. Therefore, conventional image acquisition methods struggle to obtain defect information in the area between the weld and the bevel during the early stages of multi-layer, multi-pass welding of thick-walled steel pipes, directly affecting the accuracy of subsequent image processing and welding quality assessment.

[0003] For example, Chinese patent application publication number CN120948617A discloses a defect identification method for steel pipe manufacturing, comprising: acquiring weld image data of the steel pipe surface and determining a weld morphology index based on the image data, wherein the image data includes weld width, weld reinforcement height, and undercut depth; determining the weld weld qualification based on the weld morphology index, and determining the cause of weld failure based on the average undercut depth of the weld if the weld is unqualified; determining an ultrasonic testing strategy based on the defect correspondence rate between weld width and weld reinforcement height if the weld is qualified, and obtaining defect feature values ​​inside the weld through ultrasonic testing; determining the steel pipe manufacturing qualification based on the defect feature values, and determining whether the steel pipe meets a preset standard based on the defect area ratio if the steel pipe manufacturing is unqualified. This invention improves the accuracy and efficiency of steel pipe defect identification.

[0004] The following problems still exist in the existing technology: Existing technologies cannot effectively detect welding defects between the edge and bevel of the filler layer weld in multi-layer, multi-pass welding. Summary of the Invention

[0005] Therefore, the present invention provides an image recognition-based method for detecting weld seams in thick-walled steel pipes, which overcomes the problem in the prior art that welding defects between the edge and bevel of the filler layer weld seam in multi-layer, multi-pass welds cannot be effectively detected.

[0006] To achieve the above objectives, the present invention provides a method for detecting weld seams in thick-walled steel pipes based on image recognition, comprising: The surface of a single weld is vertically irradiated with coaxial light, and images of several weld sections are captured by a preset sliding distance. Several detection points are set on each weld section along the weld direction at a preset distance. The weld ridge line and weld edge line of each weld section are determined to divide the weld section into several confidence intervals and several non-confidence intervals. Analyze the proportion of invalid points in the confidence interval, determine whether to use low-angle grazing light to illuminate the confidence interval to determine edge defect features based on the proportion of invalid points, and determine whether the confidence interval is a defect interval based on the edge defect features; Analyze the proportion of strong edge points in the non-confidence intervals, and determine the type of non-confidence interval and the corresponding detection method based on the proportion of strong edge points, including: The images obtained by vertically illuminating the weld surface with coaxial light and the images obtained by illuminating the non-confidence interval with low-angle grazing light are fused together for the corresponding intervals. Based on the fused image, it is determined whether the non-confidence interval is a defect interval. An endoscope is used to acquire weld edge bevel images of non-confidence intervals to determine bevel defect features, and based on the bevel defect features, it is determined whether the non-confidence interval is a defect interval. The preset sliding distance is reduced based on the defect rate within the confidence interval, or the defect rate within the non-confidence interval is analyzed, and the preset distance is reduced based on the defect rate within the non-confidence interval.

[0007] Furthermore, the process of determining the weld ridge line and weld edge line for each weld section includes, Several detection points are set at preset distances along the weld direction in each weld section; On each surface image, starting from the detection point, several points with the maximum gray value are obtained along the direction perpendicular to the weld. Connect several points with the highest gray values ​​to form a line to generate a welding ridge line; On each surface image, scan from the weld ridge line to both sides along the direction perpendicular to the weld seam, passing through each detection point, to determine strong edge points or weak edge points; A line is formed by connecting several strong edge points and several selected weak edge points to generate a weld edge line.

[0008] Furthermore, the process of dividing the weld interval into several confidence intervals and several non-confidence intervals includes, Set up corresponding inspection lines at each inspection point along the direction perpendicular to the weld seam; Calculate the set of distances between the intersection points of the detection line and the weld ridge line and the intersection points of the detection line and the weld edge line; The standard deviation of the distance set is calculated as the confidence level of the corresponding weld interval; The weld interval is divided into a confidence interval or a non-confidence interval based on the confidence level.

[0009] Furthermore, the process of determining whether to use low-angle grazing light to illuminate the confidence interval to determine edge defect features based on the proportion of invalid points includes, Obtain the number of invalid points within the confidence interval, and determine the proportion of invalid points based on the number of invalid points; Based on the proportion of invalid points, determine whether the confidence interval is normal or whether to illuminate the confidence interval using low-angle grazing light; The image obtained by illuminating the confidence interval with low-angle grazing light is analyzed to obtain the defect pixel area and average gray level difference in the preset pixel area on both sides of the weld edge; Edge defect features are determined based on the defect pixel area and the average gray level difference.

[0010] Furthermore, the process of determining whether the confidence interval is a defect interval based on the defect characteristics includes, If the defect feature exceeds a preset defect feature threshold, the confidence interval is marked as a defect interval.

[0011] Furthermore, the process of determining the type of non-confidence interval based on the proportion of strong edge points includes, Obtain the number of strong edge points within the non-confidence interval, and determine the proportion of strong edge points based on the number of strong edge points; Based on the proportion of strong edge points, the type of non-confidence interval is determined to be either clear edge or blurred edge.

[0012] Furthermore, the process of determining whether the non-confidence interval is a defect interval based on the fused image includes, For unconfidence regions with clear edges, the images obtained by vertically illuminating the weld surface with coaxial light and the images obtained by illuminating the unconfidence regions with low-angle grazing light are fused based on the weld ridge line for the corresponding regions. Calculate the grayscale value of each pixel point within a preset pixel area on both sides of the weld seam edge in the fused image, and count the abnormal image area covered by the pixel points whose grayscale values ​​are not within the grayscale threshold range; The abnormal area ratio is determined based on the abnormal image area, and the non-confidence interval is determined as a defect interval based on the abnormal area ratio.

[0013] Furthermore, the process of determining whether the non-confidence interval is a defect interval based on the bevel defect characteristics includes, For unconfidence intervals with blurred edges, use an endoscope to obtain weld edge bevel images of the unconfidence intervals; Based on the weld edge bevel image, obtain the bevel defect pixel area and average grayscale difference of the bevel within the preset pixel area on both sides of the weld edge line; The bevel defect features are determined based on the pixel area of ​​the bevel defect and the average gray level difference of the bevel. If the bevel defect features exceed the preset bevel defect feature threshold, then the non-confidence interval is marked as a defect interval.

[0014] Furthermore, the process of determining whether to reduce the preset sliding distance or analyze the defect rate in the non-confidence interval based on the defect rate of the confidence interval includes, The percentage of defective intervals within a confidence interval is defined as the defect rate of the confidence interval. If the defect rate of the confidence interval is greater than or equal to the confidence defect rate threshold, then the preset sliding distance is reduced. If the defect rate in the confidence interval is less than the confidence defect rate threshold, then the defect rate in the non-confidence interval is determined to be analyzed.

[0015] Furthermore, the process of determining whether to reduce the preset distance based on the defect rate within the non-confidence interval includes, The percentage of defective intervals within the non-confidence interval is defined as the defect rate of the non-confidence interval. If the defect rate of the unconfidence interval is greater than or equal to the unconfidence defect rate threshold, then the preset distance is reduced.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention first uses coaxial light to perpendicularly irradiate the surface of a single weld, dividing the single weld into several detection intervals. For each detection interval, weld ridges and edges are simulated and generated. A preliminary analysis of the weld morphology in each detection interval is performed. High-quality weld intervals with uniform morphology and minimal width variation are classified as confidence intervals, while low-quality weld intervals with less uniform morphology and significant width variation are classified as non-confidence intervals. Subsequently, different image acquisition methods are used for the confidence and non-confidence intervals. For the non-confidence intervals where defects are frequent, acquisition parameters are optimized and image details are supplemented. This effectively solves the problems of indiscriminate detection of weld quality intervals, missed defect detection, and large recognition deviations in traditional image acquisition technologies. It effectively improves the image quality and feature recognition accuracy of thick-walled steel pipe welds, increasing both detection efficiency and accuracy based on image recognition for thick-walled steel pipe weld detection.

[0017] In particular, this invention determines whether the weld corresponding to the confidence interval is normal by analyzing the number of invalid points within the confidence interval. Under the condition of abnormal weld quality, low-angle grazing light is used to illuminate the weld corresponding to the confidence interval to determine whether the confidence interval is normal or whether low-angle grazing light needs to be used for supplementary illumination. This avoids the redundancy and efficiency loss caused by adding light sources to all intervals indiscriminately. Since low-angle grazing light is more sensitive to defects such as small height changes, depressions, undercuts, lack of fusion, and steps at the weld edge, it can form obvious shadows or grayscale differences, making edge defects more prominent in the image, which is conducive to improving the defect detection rate. It effectively solves the problems of low contrast, blurred features, and easy missed detection of edge defects under traditional coaxial light illumination, and further improves the accuracy of thick-walled steel pipe weld detection based on image recognition.

[0018] In particular, this invention addresses the characteristics of uneven weld morphology and significant differences in edge quality within non-confidence intervals by constructing a classified and graded defect detection system. This invention categorizes non-confidence intervals into two types: those with clear edges and those with blurred edges, based on statistically analyzing the number and proportion of strong edge points within the intervals. This allows subsequent detection strategies to accurately match the actual edge quality characteristics of the intervals, avoiding the problem of insufficient adaptability of a single detection method to different morphological intervals. For non-confidence intervals with clear edges, this invention fuses the coaxial vertical illumination image and the low-angle grazing illumination image based on the weld ridge line for the corresponding interval, enabling the two... The imaging advantages of similar light sources are complementary and superimposed in the same coordinate system, which not only preserves the overall morphological information of the weld but also enhances the contrast of edge details, effectively improving the sensitivity and accuracy of defect identification in areas with clear edges. For non-confidence areas with blurred edges, this invention introduces an endoscope to directly acquire images of the weld edge bevel and determine the bevel defect characteristics. Since the endoscope is used to acquire images of the weld edge bevel, the internal and edge details of the bevel can be obtained at close range, effectively overcoming the detection difficulties caused by edge blurring, improving the detection rate of bevel defects, especially in scenarios with blurred edges, and further improving the accuracy of thick-walled steel pipe weld detection based on image recognition.

[0019] In particular, this invention obtains the defect rate of a confidence interval by statistically analyzing the proportion of defective intervals within that interval, and compares it with a confidence defect rate threshold. When the confidence interval defect rate is greater than or equal to the threshold, it indicates a high concentration of defects in that weld segment, making the welding process prone to quality defects. This invention reduces the preset sliding distance, thereby increasing the detection sampling density and narrowing the interval between adjacent detection intervals, effectively avoiding missed defects due to excessively large sliding steps. When the confidence interval defect rate is less than the threshold, the defect rate of the non-confidence interval is further analyzed, using the proportion of defective intervals within the non-confidence interval as the judgment criterion. If the non-confidence interval defect rate is greater than or equal to the non-confidence defect rate threshold, the preset distance reduction operation is also triggered. This automatically densifies sampling and improves recognition resolution in high-defect areas, while maintaining a reasonable step size and avoiding redundant detection in low-defect areas. Under different weld quality conditions, it maintains high detection accuracy and stability, further improving the accuracy of image recognition-based thick-walled steel pipe weld detection. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the steps of an image recognition-based weld seam detection method for thick-walled steel pipes according to an embodiment of the present invention. Figure 2 A flowchart illustrating the steps for determining edge defect features in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the steps for determining the type of non-confidence interval in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the steps of determining whether a non-confidence interval is a defect interval based on bevel defect characteristics in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0022] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0023] It should be noted that in the description of this invention, the terms "upper," "lower," "inner," "outer," etc., which indicate the direction or positional relationship, are based on the direction or positional relationship shown in the drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0024] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0025] Please see Figure 1 The diagram illustrates the steps of an image recognition-based weld seam detection method for thick-walled steel pipes according to an embodiment of the present invention. The method includes: Step S1: Use coaxial light to vertically illuminate the surface of a single weld seam, take pictures at a preset sliding distance, and obtain surface images of several weld seam intervals. Set several detection points on each weld seam interval at a preset distance along the weld seam direction, and determine the weld seam ridge line and weld seam edge line of each weld seam interval to divide the weld seam interval into several confidence intervals and several non-confidence intervals. Specifically, in coaxial lighting, the light source is aligned with the camera's optical axis via a beam splitter, and the light rays illuminate the surface perpendicularly. Most of the reflected light from the planar area returns to the camera, forming a uniform bright field, while the vertices of the curved surface, i.e., the weld ridges, reflect the light directly back to the camera, creating bright lines. This is existing technology and will not be elaborated further.

[0026] Specifically, the preset sliding distance is the length of the weld used to calculate the standard deviation of weld width fluctuation. Its setting needs to balance local sensitivity and statistical representativeness. Specifically, the preset sliding distance needs to include a sufficient number of sampling points so that the calculated standard deviation of weld width fluctuation is statistically significant and can reflect the average fluctuation level of the distance between the weld ridge and the weld edge in the weld interval. If the window is too short and contains too few sample points, the calculated standard deviation will fluctuate wildly due to accidental noise and cannot represent the true quality of the weld interval. On the other hand, the sliding distance cannot be too long, otherwise, it will average out these real local fluctuations, causing the standard deviation value to be diluted and unable to effectively capture local forming defects. For example, the preset sliding distance can account for 5% to 20% of the weld length.

[0027] Specifically, the preset distance is the distance between adjacent detection points within the weld interval set to simulate the weld ridge and edge. Its setting needs to balance the accuracy of geometric feature description and data processing efficiency. Specifically, a smaller preset distance is needed to restore the direction of the weld. If the preset distance is too large, the spatial geometric details will be smoothed out, causing the fitted ridge and edge to deviate too much from the actual physical position, and the error of subsequent calculations will increase significantly. On the other hand, the preset distance cannot be too small, because the pixel coordinates in the image itself contain quantization noise, and there may be random interference from spatter or grayscale transition areas at the weld edge. If the step size is too short, the fitting curve will over-track these noise points, generating high-frequency oscillations, which will cause the extraction of weld ridge or edge to be unstable. At the same time, excessively dense sampling will linearly increase the computational burden, which is not conducive to real-time processing in long weld detection. For example, the preset distance can account for 0.1% to 1% of the total length of the effective weld interval.

[0028] Step S2: Analyze the proportion of invalid points in the confidence interval, determine whether to use low-angle grazing light to illuminate the confidence interval to determine edge defect features based on the proportion of invalid points, and determine whether the confidence interval is a defect interval based on the edge defect features. Specifically, low-angle grazing light is a light source that grazes the surface of the base material from all sides. The ring light source or strip light group is installed 10 to 30 mm away from the workpiece surface. The angle is adjusted so that the angle between the light and the workpiece surface is 10° to 15°. This ensures that the light grazes the weld surface evenly from all sides, producing a non-directional shadow. The non-directional shadow is an optical effect that is achieved by using ring / multi-directional low-angle lighting to make the small geometric undulations of the weld surface produce a uniform shadow in all directions. This will not be elaborated on here.

[0029] Step S3: Analyze the proportion of strong edge points in the non-confidence interval, and determine the type of non-confidence interval and the corresponding detection method based on the proportion of strong edge points, including... The images obtained by vertically illuminating the weld surface with coaxial light and the images obtained by illuminating the non-confidence interval with low-angle grazing light are fused together for the corresponding intervals. Based on the fused image, it is determined whether the non-confidence interval is a defect interval. Step S4: Use an endoscope to acquire weld edge bevel images of the non-confidence interval to determine bevel defect features, and determine whether the non-confidence interval is a defect interval based on the bevel defect features. Step S5: Based on the defect rate in the confidence interval, determine whether to reduce the preset sliding distance or analyze the defect rate in the non-confidence interval, and based on the defect rate in the non-confidence interval, determine whether to reduce the preset distance.

[0030] Specifically, this invention first uses coaxial light to perpendicularly illuminate the surface of a single weld, dividing the weld into several detection intervals. For each detection interval, weld ridges and edges are simulated and generated. A preliminary analysis of the weld morphology in each detection interval is performed. High-quality weld intervals with uniform morphology and minimal width variation are designated as confidence intervals, while low-quality weld intervals with less uniform morphology and significant width variation are designated as non-confidence intervals. Subsequently, different image acquisition methods are used for the confidence and non-confidence intervals. For the non-confidence intervals where defects are frequent, acquisition parameters are optimized and image details are supplemented. This effectively solves the problems of indiscriminate detection of weld quality intervals, missed defect detection, and large recognition deviations in traditional image acquisition techniques. It effectively improves the image quality and feature recognition accuracy of thick-walled steel pipe welds, increasing both detection efficiency and accuracy based on image recognition for thick-walled steel pipe weld detection.

[0031] Specifically, in step S1, the process of determining the weld ridge line and weld edge line of each weld section includes, Several detection points are set at preset distances along the weld direction in each weld section; On each surface image, starting from the detection point, several points with the maximum gray value are obtained along the direction perpendicular to the weld. Connect several points with the highest gray values ​​to form a line to generate a welding ridge line; Specifically, after obtaining several points with the highest grayscale values, some of these points may deviate from the actual ridge line due to noise or reflection interference. Median filtering can be used, with a window size of 5-11, to smooth the peak position curve. Simultaneously, points with grayscale values ​​less than the peak grayscale threshold are removed. A peak grayscale threshold is set to improve the accuracy of welding ridge line generation. The peak grayscale threshold is set to 70% of the maximum grayscale value in the image. Since areas with grayscale values ​​greater than the peak grayscale threshold are bright and clear, the extracted points with the highest grayscale values ​​are more reliable. The peak grayscale threshold can be determined by those skilled in the art based on the accuracy requirements for drawing the welding ridge line; the higher the accuracy requirement, the higher the peak grayscale threshold should be. Since an excessively high threshold may cause ridge lines in the area to be marked as invalid, preferably, the set peak grayscale threshold does not exceed 85%, which will not be elaborated further here.

[0032] On each surface image, scan from the weld ridge line to both sides along the direction perpendicular to the weld seam, passing through each detection point, to determine the strong edge points or weak edge points; Specifically, the image is scanned along the direction perpendicular to the weld seam, starting from the weld ridge and moving outwards to both sides. The Sobel operator is used to calculate the first derivative of the image in the horizontal and vertical directions, thereby obtaining the gradient magnitude of each pixel. Points with the maximum gradient magnitude are selected, and those with a maximum gradient magnitude greater than or equal to a second gradient threshold are determined as strong edge points. If the maximum gradient magnitude is less than the second gradient threshold, then points with a maximum gradient magnitude greater than or equal to a first gradient threshold and adjacent to at least one strong edge point are determined as weak edge points. If the maximum gradient magnitude is less than the first gradient threshold, then the points with the maximum gradient magnitude are determined to be invalid.

[0033] As can be understood, the gradient is a vector representing the change in grayscale value of each pixel in an image. It contains two mutually perpendicular directions: the horizontal and vertical directions, representing the rate of change in grayscale value. The direction of this vector points in the direction of the fastest change in grayscale value, while the length of the vector, the gradient magnitude, represents the abruptness of the grayscale change. In an image, regions with gentle grayscale changes have smaller gradient magnitudes, while areas with abrupt grayscale changes, such as weld edges, have larger gradient magnitudes. Therefore, the gradient can effectively reflect the location of weld edges in an image, which will not be elaborated further here.

[0034] It is understandable that for each pixel in the image, the Sobel operator is first used to calculate the amount of gray change in the horizontal and vertical directions respectively. Then, the gray change in the horizontal and vertical directions is squared respectively. The two squared values ​​are added together, and the square root of the sum is the value of the gradient magnitude of the pixel. This will not be elaborated further here.

[0035] It is understood that the first gradient threshold is less than the second gradient threshold. In this embodiment, the first and second gradient thresholds are used to quantify the severity of grayscale changes at the maximum gradient amplitude point. Then, by comparing the maximum gradient amplitude with the first and second gradient thresholds respectively, the maximum gradient amplitude point is determined to be a strong edge point, a weak edge point, or an invalid point. In this embodiment, the second gradient threshold is determined based on the maximum image gradient amplitude, and the first gradient threshold is determined based on the second gradient threshold. Preferably, the second gradient threshold is 70% of the maximum image gradient amplitude, and the first gradient threshold is 50% of the second gradient threshold.

[0036] Connect several strong edge points and several selected weak edge points to form a line to generate the weld edge line.

[0037] Specifically, in step S1, the process of dividing the weld interval into several confidence intervals and several non-confidence intervals includes, Set up corresponding inspection lines at each inspection point along the direction perpendicular to the weld seam; Calculate the set of distances between the intersections of the inspection line and the weld ridge line and the intersections of the inspection line and the weld edge line; The standard deviation of the distance set is used as the confidence level for the corresponding weld interval; The weld interval is divided into a confidence interval or an unconfidence interval based on the confidence level.

[0038] Specifically, if the confidence level is less than or equal to the confidence level threshold, it indicates that the distance from the weld edge to the weld ridge fluctuates little, the weld edge is straight and uniform, and the quality is good. In this case, the weld interval is divided into a confidence interval. If the confidence level is greater than the confidence level threshold, it indicates that the distance from the weld edge to the weld ridge fluctuates greatly, and there may be weld defects such as undercut and weld beads. In this case, the weld interval is divided into a non-confidence interval. Specifically, in this embodiment, the confidence threshold is used to quantify the degree of distance fluctuation from the weld edge to the weld ridge. Then, by comparing the confidence level with the confidence threshold, the weld interval is divided into a confidence interval or a non-confidence interval. In this embodiment, the higher the user's requirement for the accuracy of the confidence interval determination, the smaller the confidence threshold. Preferably, the confidence threshold is the product of the confidence threshold reference value and the confidence deviation factor. The confidence threshold reference value is the mean of the standard deviations of the distance sets corresponding to several standard welds under the same welding conditions in historical data. The same welding conditions include the same pipe material, the same pipe wall thickness, the same welding process, etc. The confidence deviation factor can be set by those skilled in the art based on their requirement for the accuracy of the confidence interval determination. The higher the accuracy requirement, the smaller the value should be. The value range can be [1.0, 1.1], preferably 1.05.

[0039] Please see Figure 2 The diagram illustrates the steps for determining edge defect features according to an embodiment of the present invention. In step S2, the process of determining whether to use low-angle grazing light to illuminate the confidence interval to determine edge defect features based on the proportion of invalid points includes... Step S201: Obtain the number of invalid points within the confidence interval, and determine the proportion of invalid points based on the number of invalid points; Specifically, in this embodiment, the invalid point ratio is the ratio of the number of invalid points within the confidence interval to the total number of detection points.

[0040] Step S202: Determine whether the confidence interval is normal or illuminated with low-angle grazing light based on the proportion of invalid points; Specifically, if the proportion of invalid points is less than or equal to the threshold for the proportion of invalid points, it indicates that there are many weld edge points constituting the confidence interval, and the weld edge is relatively stable, thus the confidence interval is determined to be normal. If the proportion of invalid points is greater than the threshold for invalid points, it indicates that there are fewer weld edge points constituting the confidence interval and the weld edge is less stable. In this case, low-angle grazing light is used to illuminate the confidence interval for imaging to further examine the weld edge area. Specifically, in this embodiment, the invalid point percentage threshold is used to quantify the stability of the weld edge of the confidence interval. Then, by comparing the invalid point percentage with the invalid point percentage threshold, it is determined whether to use low-angle grazing light to illuminate the confidence interval. In this embodiment, the higher the user's requirement for the accuracy of determining the normality of the confidence interval, the smaller the invalid point percentage threshold. Preferably, the invalid point percentage threshold is the product of the invalid point percentage threshold reference value and the invalid point percentage deviation factor. The invalid point percentage threshold reference value is the average of several invalid point percentage thresholds under the same welding conditions in historical data. The invalid point percentage deviation factor can be set by those skilled in the art based on their requirements for the accuracy of determining the normality of the confidence interval. The higher the accuracy requirement, the smaller the value should be. The value range can be [1.0, 1.3], preferably 1.15.

[0041] Step S203: Analyze the image obtained by illuminating the confidence interval with low-angle grazing light to obtain the defect pixel area and average gray level difference in the preset pixel area on both sides of the weld edge. Specifically, the preset pixel is the side length of the pixel region used to calculate the defect pixel area and the average grayscale difference. Its setting needs to take into account both local sensitivity and spatial detail preservation. Specifically, the preset pixel region needs to be large enough to ensure that most pixels in the region belong to the normal background rather than the defect itself. If the window is too small, the defect region may occupy most of the region, causing the background estimation value to be biased towards the defect grayscale, thereby weakening the ratio contrast and reducing the detection sensitivity. At the same time, the preset pixel region also needs to be small enough to adapt to local features such as uneven lighting, reflection changes, or texture gradients. If the window is too large, it will cross multiple different grayscale regions, causing the background estimation to be distorted and unable to correctly reflect the local background grayscale. For example, the preset pixel can be 15 pixels.

[0042] Specifically, within the preset pixel area on both sides of the weld edge, morphological top cap or high cap transformation is used to highlight the dark band or bright convexity. Connected component labeling is used to extract all suspicious areas. The pixel area of ​​each suspicious area is recorded as the defect pixel area. The average grayscale difference of each suspicious area is also recorded. These details will not be elaborated here.

[0043] Step S204: Determine edge defect features based on defect pixel area and average grayscale difference.

[0044] Specifically, edge defect features are used to determine whether defects occur in the region adjacent to the weld edge line within the confidence interval. These features are determined based on the defect pixel area and the average grayscale difference. In this embodiment, the edge defect features are defined using a formula... Determine, where D is the edge defect feature, G is the average grayscale difference, and A is the defect pixel area (unit: square pixel). To normalize the average grayscale difference, multiply by This involves introducing an equivalent radius based on geometric dimensions to balance small but prominent defects with large but inconspicuous defects. The above formula only considers numerical values ​​during calculation.

[0045] Specifically, in step S2, the process of determining whether a confidence interval is a defect interval based on defect characteristics includes: If the defect features exceed the preset defect feature threshold, the confidence interval will be marked as a defect interval.

[0046] Specifically, in this embodiment, the defect feature threshold is used to quantify the severity of defects in the adjacent areas of the two sides of the confidence interval. Then, by comparing the defect features with the defect feature threshold, it is determined whether the confidence interval is marked as a defect interval. In this embodiment, the defect feature threshold is calibrated experimentally. Preferably, the defect feature threshold is the average value of the defect feature thresholds corresponding to several standard welds under the same welding conditions in historical data.

[0047] Specifically, this invention determines whether the weld corresponding to the confidence interval is normal by analyzing the number of invalid points within the confidence interval. Under the condition of abnormal weld quality, low-angle grazing light is used to illuminate the weld corresponding to the confidence interval to determine whether the confidence interval is normal or whether low-angle grazing light needs to be used for supplementary illumination. This avoids the redundancy and efficiency loss caused by adding light sources to all intervals indiscriminately. Since low-angle grazing light is more sensitive to defects such as small height changes, depressions, undercuts, lack of fusion, and steps at the weld edge, it can form obvious shadows or grayscale differences, making edge defects more prominent in the image, which is conducive to improving the defect detection rate. It effectively solves the problems of low contrast, blurred features, and easy missed detection of edge defects under traditional coaxial light illumination, and further improves the accuracy of thick-walled steel pipe weld detection based on image recognition.

[0048] Please see Figure 3 The diagram illustrates the steps for determining the type of a non-confidence interval in an embodiment of the present invention. In step S3, the process of determining the type of a non-confidence interval based on the proportion of strong edge points includes: Step S301: Obtain the number of strong edge points within the non-confidence interval, and determine the proportion of strong edge points based on the number of strong edge points; Specifically, in this embodiment, the proportion of strong edge points is the ratio of the number of strong edge points within the non-confidence interval to the total number of detection points.

[0049] Step S302: Determine whether the type of the non-confidence interval is clear edge type or blurry edge type based on the proportion of strong edge points.

[0050] Specifically, if the proportion of strong edge points is less than or equal to the threshold of the proportion of strong edge points, it indicates that there are fewer weld edge points constituting the non-confidence interval, the weld edge is less stable, and the type of non-confidence interval is determined to be the edge fuzzy type. If the proportion of strong edge points is greater than the threshold for the proportion of strong edge points, it indicates that there are many weld edge points constituting the confidence interval, the weld edge is relatively stable, and the type of non-confidence interval is determined to be the clear edge type. Specifically, in this embodiment, the strong edge point proportion threshold is used to quantify the stability of the weld edge line in the non-confidence interval. Then, by comparing the strong edge point proportion with the strong edge point proportion threshold, the non-confidence interval type is determined to be either a clear edge type or a blurry edge type. In this embodiment, the higher the user's requirement for the accuracy of the determination of the clear edge type, the larger the strong edge point proportion threshold. Preferably, the strong edge point proportion threshold is the product of the strong edge point proportion threshold reference value and the strong edge point proportion deviation factor. The strong edge point proportion threshold reference value is the average of several strong edge point proportion thresholds under the same welding conditions in historical data. The strong edge point proportion deviation factor can be set by those skilled in the art based on their requirements for the accuracy of the determination of the clear edge type. The higher the accuracy requirement, the larger the value should be. The value range can be [0.9, 1.0], and preferably, it can be 0.95.

[0051] Specifically, in step S3, the process of determining whether a non-confidence interval is a defect interval based on the fused image includes, Step S303: For the unconfidence region with clear edges, the images obtained by vertically illuminating the weld surface with coaxial light and the images obtained by illuminating the unconfidence region with low-angle grazing light are fused based on the weld ridge line to form the corresponding region images. It is understandable that, using the weld ridge line as a reference, the coaxial light image as a reference image, and the low-angle grazing light image as the image to be registered, the weld ridge line point sets extracted from the two images are matched, and the transformation matrix, which includes rotation, translation, and scaling parameters, is solved by the least squares method or the iterative nearest point algorithm. The transformation matrix is ​​then applied to the low-angle grazing light image, and resampling is performed using bilinear interpolation or cubic convolution interpolation to make its ridge line completely coincide with the ridge line of the coaxial light image. This is existing technology and will not be elaborated here.

[0052] Step S304: Calculate the gray value of each pixel in the preset pixel area on both sides of the weld edge in the fused image, and count the abnormal image area covered by the pixel whose gray value is not in the gray value threshold range. Specifically, the grayscale threshold range is used to quantify the degree of abnormality of each pixel point within a preset pixel area on both sides of the weld edge. Then, by comparing the grayscale value with the grayscale threshold range, the area of ​​the abnormal image is determined. In this embodiment, preferably, the initial values ​​of the minimum and maximum values ​​of the grayscale threshold range can be calibrated experimentally. Under the same working conditions, the average of the minimum grayscale values ​​of several standard normal welds is set as the minimum grayscale threshold range, and the average of the maximum grayscale values ​​of several standard normal welds is set as the maximum grayscale threshold range.

[0053] Step S305: Determine the abnormal area ratio based on the abnormal image area, and determine whether the non-confidence interval is a defect interval based on the abnormal area ratio.

[0054] Specifically, in this embodiment, the abnormal area ratio is the ratio of the area of ​​the abnormal image to the area of ​​the preset pixel region on both sides of the weld edge.

[0055] Specifically, if the abnormal area ratio is less than or equal to the abnormal area ratio threshold, it indicates that the preset pixel areas on both sides of the weld edge are relatively stable, and the non-confidence interval is determined not to be a defect interval. If the abnormal area ratio is greater than the abnormal area ratio threshold, it indicates that the preset pixel areas on both sides of the weld edge are relatively unstable, and the non-confidence interval is determined as the defect interval. Specifically, in this embodiment, the abnormal area ratio threshold is used to quantify the stability of the weld edge in the non-confidence interval. Then, by comparing the abnormal area ratio with the abnormal area ratio threshold, it is determined whether the non-confidence interval is a defect interval. In this embodiment, the higher the user's requirement for accuracy in determining whether the non-confidence interval is a defect interval, the smaller the abnormal area ratio threshold. Preferably, the abnormal area ratio threshold is the product of the abnormal area ratio threshold reference value and the abnormal area ratio deviation factor. The abnormal area ratio threshold reference value is the average of several abnormal area ratio thresholds under the same welding conditions in historical data. The abnormal area ratio deviation factor can be set by those skilled in the art based on their requirements for accuracy in determining the normality of the confidence interval. The higher the accuracy requirement, the smaller the value should be. The value range can be [0.9, 1.0], preferably 0.95.

[0056] Please see Figure 4 The diagram illustrates the steps of determining whether a non-confidence interval is a defect interval based on bevel defect features according to an embodiment of the present invention. In step S4, the process of determining whether a non-confidence interval is a defect interval based on bevel defect features includes... Step S401: For the edge-blurred non-confidence interval, use an endoscope to obtain the weld edge bevel image of the non-confidence interval; Step S402: Based on the weld edge bevel image, obtain the bevel defect pixel area and average gray level difference of the bevel within the preset pixel area on both sides of the weld edge line; Step S403: Determine the bevel defect features based on the bevel defect pixel area and the average gray level difference of the bevel. Specifically, the bevel defect feature is used to determine whether a defect occurs in the area adjacent to the weld edge in the non-confidence interval. It is determined based on the bevel defect pixel area and the average grayscale difference of the bevel. In this embodiment, the bevel defect feature is determined using a formula... Determine, where F represents the bevel defect feature, H represents the average grayscale difference of the bevel, and B represents the pixel area of ​​the bevel defect, in square pixels. To normalize the average grayscale difference, multiply by This involves introducing an equivalent radius based on geometric dimensions to balance small but prominent defects with large but inconspicuous defects. The above formula only considers numerical values ​​during calculation.

[0057] Step S404: If the bevel defect features exceed the preset bevel defect feature threshold, then the non-confidence interval is marked as a defect interval.

[0058] Specifically, in this embodiment, the bevel defect feature threshold is used to quantify the severity of defects in the adjacent areas of the two sides of the non-confidence interval. Then, by comparing the bevel defect features with the bevel defect feature threshold, it is determined whether the non-confidence interval is marked as a defect interval. In this embodiment, the bevel defect feature threshold is calibrated experimentally. Preferably, the bevel defect feature threshold is the average value of the bevel defect feature thresholds corresponding to several standard welds under the same welding conditions in historical data.

[0059] Specifically, this invention addresses the characteristics of uneven weld morphology and significant edge quality differences in non-confidence intervals by constructing a classified and graded defect detection system. This invention categorizes non-confidence intervals into two types: those with clear edges and those with blurred edges, based on statistically analyzing the number and proportion of strong edge points within the intervals. This allows subsequent detection strategies to accurately match the actual edge quality characteristics of the intervals, avoiding the problem of insufficient adaptability of a single detection method to different morphological intervals. For non-confidence intervals with clear edges, this invention fuses images of corresponding intervals based on the weld ridge line using coaxial vertical illumination and low-angle grazing illumination. The imaging advantages of the two types of light sources are complementary and superimposed in the same coordinate system, which not only preserves the overall morphological information of the weld but also enhances the contrast of edge details, effectively improving the sensitivity and accuracy of defect identification in areas with clear edges. For non-confidence areas with blurred edges, this invention introduces an endoscope to directly acquire images of the weld edge bevel and determine the bevel defect characteristics. Since the weld edge bevel image is acquired using an endoscope, the internal and edge details of the bevel can be obtained at close range, effectively overcoming the detection difficulties caused by edge blurring and improving the detection rate of bevel defects, especially in scenarios with blurred edges, further improving the accuracy of thick-walled steel pipe weld detection based on image recognition.

[0060] Specifically, in step S5, the process of determining whether to reduce the preset sliding distance or analyze the defect rate in the non-confidence interval based on the defect rate of the confidence interval includes, The percentage of defective intervals within a confidence interval is defined as the defect rate of the confidence interval. If the defect rate in the confidence interval is greater than or equal to the confidence defect rate threshold, then the preset sliding distance is reduced. If the defect rate in the confidence interval is less than the confidence defect rate threshold, then the defect rate in the non-confidence interval is determined for analysis.

[0061] Specifically, in this embodiment, the confidence defect rate threshold is used to quantify the severity of defects in several confidence intervals contained in the weld. Then, by comparing the defect rate of the confidence interval with the confidence defect rate threshold, it is determined whether to reduce the preset sliding distance. If the defect rate of the confidence interval is too large, it is determined that the preset sliding distance is too large, leading to inaccurate classification of the confidence interval in the early stages. In this embodiment, the higher the user's requirement for the accuracy of the confidence interval classification, the smaller the confidence defect rate threshold. Preferably, the confidence defect rate threshold is the product of the confidence defect rate threshold reference value and the confidence defect rate deviation factor. The confidence defect rate threshold reference value is the average of several confidence defect rate thresholds under the same welding conditions in historical data. The confidence defect rate deviation factor can be set by those skilled in the art based on their requirements for the accuracy of the confidence interval classification; the higher the accuracy requirement, the smaller the value should be. The value range can be [1.0, 1.5], preferably 1.25.

[0062] It is understandable that the reduction in the preset sliding distance can be determined based on the degree of deviation between the defect rate of the confidence interval and the confidence defect rate threshold. For example, the reduction ratio of the preset sliding distance is the ratio of the confidence deviation value to the confidence defect rate threshold, where the confidence deviation value is the absolute value of the deviation between the defect rate of the confidence interval and the confidence defect rate threshold.

[0063] Specifically, in step S5, the process of determining whether to reduce the preset distance based on the defect rate of the non-confidence interval includes, The percentage of defective intervals within the non-confidence interval is defined as the defect rate of the non-confidence interval. If the defect rate in the unconfidence interval is less than the unconfidence defect rate threshold, then the preset distance is reduced.

[0064] Specifically, in this embodiment, the unconfidence defect rate threshold is used to quantify the severity of defects in several unconfidence intervals contained in the weld. Then, by comparing the defect rate of the unconfidence interval with the unconfidence defect rate threshold, it is determined whether to reduce the preset distance. If the defect rate of the unconfidence interval is too small, fewer defects will be detected in subsequent inspections for unconfidence intervals with higher defect risks during the initial detection and classification. Therefore, increasing the number of detection points ensures the accuracy of the detection. In this embodiment, the higher the user's requirement for the accuracy of defect detection in the unconfidence interval, the larger the unconfidence defect rate threshold. Preferably, the unconfidence defect rate threshold is the product of the unconfidence defect rate threshold reference value and the unconfidence defect rate deviation factor. The unconfidence defect rate threshold reference value is the average of several unconfidence defect rate thresholds under the same welding conditions in historical data. The unconfidence defect rate deviation factor can be set by those skilled in the art based on their requirements for the accuracy of defect detection in the unconfidence interval. The higher the accuracy requirement, the larger the value should be. The value range can be [0.5, 1.0], preferably 0.75.

[0065] It is understandable that the reduction of the preset distance can be determined based on the degree of deviation between the defect rate of the unconfidence interval and the unconfidence defect rate threshold. For example, the reduction ratio of the preset distance is the ratio of the unconfidence deviation value to the unconfidence defect rate threshold, and the unconfidence deviation value is the absolute value of the deviation between the defect rate of the unconfidence interval and the unconfidence defect rate threshold.

[0066] Specifically, this invention obtains the defect rate of a confidence interval by statistically analyzing the proportion of defective intervals within that interval, and compares it with a confidence defect rate threshold. When the confidence interval defect rate is greater than or equal to the threshold, it indicates a high concentration of defects in that weld segment, making the welding process prone to quality defects. This invention reduces the preset sliding distance to increase the detection sampling density, narrow the interval between adjacent detection intervals, and effectively avoid missed defects due to excessively large sliding steps. When the confidence interval defect rate is less than the threshold, the defect rate of the non-confidence interval is further analyzed, using the proportion of defective intervals within the non-confidence interval as the judgment criterion. If the non-confidence interval defect rate is greater than or equal to the non-confidence defect rate threshold, the preset distance reduction operation is also triggered. This automatically densifies sampling and improves recognition resolution in high-defect-occurrence areas, while maintaining a reasonable step size and avoiding redundant detection in low-defect-occurrence areas. Under different weld quality conditions, it maintains high detection accuracy and stability, further improving the accuracy of image recognition-based thick-walled steel pipe weld detection.

[0067] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for inspecting weld seams in thick-walled steel pipes based on image recognition, characterized in that, include: The surface of a single weld is vertically irradiated with coaxial light, and images of several weld sections are captured by a preset sliding distance. Several detection points are set on each weld section along the weld direction at a preset distance. The weld ridge line and weld edge line of each weld section are determined to divide the weld section into several confidence intervals and several non-confidence intervals. Analyze the proportion of invalid points in the confidence interval, determine whether to use low-angle grazing light to illuminate the confidence interval to determine edge defect features based on the proportion of invalid points, and determine whether the confidence interval is a defect interval based on the edge defect features; Analyze the proportion of strong edge points in the non-confidence intervals, and determine the type of non-confidence interval and the corresponding detection method based on the proportion of strong edge points, including: The images obtained by vertically illuminating the weld surface with coaxial light and the images obtained by illuminating the non-confidence interval with low-angle grazing light are fused together for the corresponding intervals. Based on the fused image, it is determined whether the non-confidence interval is a defect interval. An endoscope is used to acquire weld edge bevel images of non-confidence intervals to determine bevel defect features, and based on the bevel defect features, it is determined whether the non-confidence interval is a defect interval. The preset sliding distance is reduced based on the defect rate within the confidence interval, or the defect rate within the non-confidence interval is analyzed, and the preset distance is reduced based on the defect rate within the non-confidence interval.

2. The image recognition-based weld seam detection method for thick-walled steel pipes according to claim 1, characterized in that, The process of determining the weld ridge line and weld edge line for each weld section includes, Several detection points are set at preset distances along the weld direction in each weld section; On each surface image, starting from the detection point, several points with the maximum gray value are obtained along the direction perpendicular to the weld. Connect several points with the highest gray values ​​to form a line to generate a welding ridge line; On each surface image, scan from the weld ridge line to both sides along the direction perpendicular to the weld seam, passing through each detection point, to determine strong edge points or weak edge points; A line is formed by connecting several strong edge points and several selected weak edge points to generate a weld edge line.

3. The image recognition-based weld seam detection method for thick-walled steel pipes according to claim 2, characterized in that, The process of dividing the weld seam interval into several confidence intervals and several non-confidence intervals includes: Set up corresponding inspection lines at each inspection point along the direction perpendicular to the weld seam; Calculate the set of distances between the intersection points of the detection line and the weld ridge line and the intersection points of the detection line and the weld edge line; The standard deviation of the distance set is calculated as the confidence level of the corresponding weld interval; The weld interval is divided into a confidence interval or a non-confidence interval based on the confidence level.

4. The image recognition-based weld seam detection method for thick-walled steel pipes according to claim 3, characterized in that, The process of determining whether to use low-angle grazing light to illuminate the confidence interval to determine edge defect characteristics based on the proportion of invalid points includes the following steps: Obtain the number of invalid points within the confidence interval, and determine the proportion of invalid points based on the number of invalid points; Based on the proportion of invalid points, determine whether the confidence interval is normal or whether to illuminate the confidence interval using low-angle grazing light; The image obtained by illuminating the confidence interval with low-angle grazing light is analyzed to obtain the defect pixel area and average gray level difference in the preset pixel area on both sides of the weld edge; Edge defect features are determined based on the defect pixel area and the average gray level difference.

5. The image recognition-based weld seam detection method for thick-walled steel pipes according to claim 4, characterized in that, The process of determining whether the confidence interval is a defect interval based on the defect characteristics includes: If the defect feature exceeds a preset defect feature threshold, the confidence interval is marked as a defect interval.

6. The image recognition-based weld seam detection method for thick-walled steel pipes according to claim 5, characterized in that, The process of determining the type of non-confidence interval based on the proportion of strong edge points includes: Obtain the number of strong edge points within the non-confidence interval, and determine the proportion of strong edge points based on the number of strong edge points; Based on the proportion of strong edge points, the type of non-confidence interval is determined to be either clear edge or blurred edge.

7. The image recognition-based weld seam detection method for thick-walled steel pipes according to claim 6, characterized in that, The process of determining whether the non-confidence interval is a defect interval based on the fused image includes: For unconfidence regions with clear edges, the images obtained by vertically illuminating the weld surface with coaxial light and the images obtained by illuminating the unconfidence regions with low-angle grazing light are fused based on the weld ridge line for the corresponding regions. Calculate the grayscale value of each pixel point within a preset pixel area on both sides of the weld seam edge in the fused image, and count the abnormal image area covered by the pixel points whose grayscale values ​​are not within the grayscale threshold range; The abnormal area ratio is determined based on the abnormal image area, and the non-confidence interval is determined as a defect interval based on the abnormal area ratio.

8. The image recognition-based weld seam detection method for thick-walled steel pipes according to claim 7, characterized in that, The process of determining whether the non-confidence interval is a defect interval based on the bevel defect characteristics includes: For unconfidence intervals with blurred edges, use an endoscope to obtain weld edge bevel images of the unconfidence intervals; Based on the weld edge bevel image, obtain the bevel defect pixel area and average grayscale difference of the bevel within the preset pixel area on both sides of the weld edge line; The bevel defect features are determined based on the pixel area of ​​the bevel defect and the average gray level difference of the bevel. If the bevel defect features exceed the preset bevel defect feature threshold, then the non-confidence interval is marked as a defect interval.

9. The image recognition-based weld seam inspection method for thick-walled steel pipes according to claim 8, characterized in that, The process of determining the reduction of the preset sliding distance or analyzing the defect rate in the non-confidence interval based on the defect rate of the confidence interval includes, The percentage of defective intervals within a confidence interval is defined as the defect rate of the confidence interval. If the defect rate of the confidence interval is greater than or equal to the confidence defect rate threshold, then the preset sliding distance is reduced. If the defect rate in the confidence interval is less than the confidence defect rate threshold, then the defect rate in the non-confidence interval is determined to be analyzed.

10. The image recognition-based weld seam detection method for thick-walled steel pipes according to claim 9, characterized in that, The process of determining whether to reduce the preset distance based on the defect rate of the non-confidence interval includes, The percentage of defective intervals within the non-confidence interval is defined as the defect rate of the non-confidence interval. If the defect rate of the unconfidence interval is greater than or equal to the unconfidence defect rate threshold, then the preset distance is reduced.

Images