Seamless steel pipe defect detection method, device and storage medium based on machine vision

Abstract

The present application discloses a seamless steel pipe defect detection method, device and storage medium based on machine vision, which belongs to the field of image analysis technology. The present application uses data from multiple detection methods (image, ultrasound, eddy current, and X-ray detection) to comprehensively locate potential defect areas, overcome the limitations that may exist in a single detection method, and improve the accuracy of detection. Operations such as determining key detection information blocks and constructing key detection attribute vectors from multi-mode detection data can deeply mine the effective information of the data and more accurately describe the state of potential defect areas. By constructing defect detection discriminant vectors through multiple detection attention levels and defect labels and outputting detection opinions, defects can be classified and evaluated in detail, which helps to provide an accurate basis for the quality control of seamless steel pipes, ensure their safety and reliability in use in various industries, and at the same time improve detection efficiency and reduce cost risks caused by misjudgment or omission of defects.

Description

Technical Field

[0001] The present application belongs to the field of image analysis technology, and specifically relates to a seamless steel pipe defect detection method, device and storage medium based on machine vision. Background Art

[0002] During the production and use of seamless steel pipes, it is crucial to ensure their quality. Traditional seamless steel pipe defect detection methods often use a single detection technology. For example, it is difficult to fully detect internal defects by relying solely on image detection, and ultrasonic detection alone is not sensitive enough to subtle surface defects. The limitations of this single detection method lead to inaccurate test results, which are prone to misjudgment or omission. In some fields with extremely high quality requirements for seamless steel pipes, such as aerospace, petrochemicals, etc., inaccurate test results may bring serious safety hazards. Moreover, it is difficult for traditional detection methods to classify and accurately evaluate defects, and it is impossible to provide an accurate basis for quality control. At the same time, it may also increase costs due to repeated inspections or improper use of defective steel pipes. Summary of the invention

[0003] The present application provides a method, device and storage medium for seamless steel pipe defect detection based on machine vision, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] The embodiment of the present application provides a seamless steel pipe defect detection method based on machine vision, which is applied to a seamless steel pipe defect detection device. The method comprises: locating a potential defect area of ​​the target seamless steel pipe according to a multi-mode detection data set of the target seamless steel pipe, and segmenting the multi-mode detection data set of the target seamless steel pipe to obtain multi-mode detection characterization information corresponding to the potential defect area; wherein the multi-mode detection characterization information comprises: image detection information, ultrasonic detection information, eddy current detection information and radiographic detection information; determining key detection information blocks from each multi-mode detection characterization information obtained by segmentation, and determining each of the key detection information blocks at each detection attention level in a set number of detection attention levels. for each of the several detection attention levels, determine the defect state characterization vector of the potential defect area at the targeted detection attention level according to the key detection attribute vector of each of the key detection information blocks at the targeted detection attention level; based on the defect state characterization vector of the potential defect area at each of the several detection attention levels, determine the defect detection discriminant vectors corresponding to the potential defect area under the set multiple defect labels; based on the defect detection discriminant vectors corresponding to the potential defect area under the multiple defect labels, output the defect detection viewpoint corresponding to the potential defect area.

[0005] An embodiment of the present application provides a seamless steel pipe defect detection device, comprising at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the above method.

[0006] An embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0007] This application uses data from multiple detection methods (image, ultrasound, eddy current, and X-ray detection) to comprehensively locate potential defect areas, overcome the limitations that may exist in a single detection method, and improve the accuracy of detection. Operations such as determining key detection information blocks and constructing key detection attribute vectors from multi-mode detection data can deeply mine the effective information of the data and more accurately describe the status of potential defect areas. By constructing defect detection discriminant vectors and outputting detection opinions through multiple detection attention levels and defect labels, defects can be classified and evaluated in detail, which helps to provide an accurate basis for the quality control of seamless steel pipes, ensure their safety and reliability in use in various industries, and at the same time improve detection efficiency and reduce cost risks caused by misjudgment or omission of defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flow chart of a seamless steel pipe defect detection method based on machine vision provided in an embodiment of the present application.

[0009] Figure 2 A schematic structural diagram of a seamless steel pipe defect detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0011] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are usually a class, and the number of objects is not limited. For example, the first object can be one or more. In addition, in this application, "and / or" represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0012] Figure 1 A seamless steel pipe defect detection method based on machine vision is shown, which is applied to a seamless steel pipe defect detection device. The method includes the following steps 101-105.

[0013] Step 101: Locate a potential defect region of a target seamless steel pipe according to a multimode detection data set of the target seamless steel pipe, and segment the multimode detection data set of the target seamless steel pipe to obtain multimode detection characterization information corresponding to the potential defect region.

[0014] The multi-mode detection characterization information includes: image detection information, ultrasonic detection information, eddy current detection information and X-ray detection information.

[0015] Step 102: Determine key detection information blocks from each multi-mode detection representation information obtained by segmentation, and determine key detection attribute vectors of each key detection information block at each detection attention level among several set detection attention levels.

[0016] Step 103: For each of the several detection attention levels, determine the defect state representation vector of the potential defect area at the targeted detection attention level based on the key detection attribute vector of each key detection information block at the targeted detection attention level.

[0017] Step 104: Based on the defect state representation vector of the potential defect area at each of the several detection attention levels, determine the defect detection discriminant vectors corresponding to the potential defect area under the set multiple defect labels.

[0018] Step 105: Based on the defect detection discriminant vectors corresponding to the potential defect area under the multiple defect labels, output the defect detection viewpoint corresponding to the potential defect area.

[0019] For the defect detection of seamless steel pipes, this application adopts a method of comprehensive multi-mode detection data set. First, in step 101, the multi-mode detection data set of the target seamless steel pipe is the basic data source for the entire detection. The multi-mode detection data set here contains data obtained by various detection methods, namely image detection information, ultrasonic detection information, eddy current detection information, and X-ray detection information. For example, the image detection information may come from the shooting results of the surface and interior of the steel pipe (if it is possible to obtain the internal image through a special imaging method) by a high-resolution industrial camera, which may contain tens of thousands of pixel data, each pixel has its corresponding color value (if it is a color image) or grayscale value (if it is a grayscale image), these data can reflect the texture, color, shape and other characteristics of the steel pipe surface, which is of great significance for detecting surface defects such as scratches, pits, and protrusions.

[0020] Ultrasonic testing information is the reflected wave and other related data received after the ultrasonic testing equipment transmits ultrasonic waves to the steel pipe. For example, when using ultrasonic waves with a frequency of 5MHz to test a seamless steel pipe with a length of 5 meters and a diameter of 10 centimeters, the ultrasonic testing equipment can determine whether there are defects inside the steel pipe based on the received reflected wave signal strength, reflected wave time and other data. If an abnormally strong reflected wave is received at a certain point in time, it may mean that there are internal defects in the ultrasonic propagation path, such as inclusions or cracks.

[0021] Eddy current detection information is data obtained using eddy current detection technology. When an alternating magnetic field of a specific frequency (e.g. 100kHz) and strength is applied to the surface of a steel pipe, the distribution of the generated eddy currents will change due to the different internal structures and characteristics of the steel pipe, and these changes are reflected by the impedance changes of the detection coil. The eddy current detection equipment records these impedance change data, which constitute the eddy current detection information, which can be used to detect areas of conductivity changes inside the steel pipe, such as possible holes or areas of uneven metal internal structure.

[0022] Radiographic detection information is the image data formed when X-rays or gamma rays penetrate steel pipes. For example, when using X-rays to detect steel pipes, the attenuation of the rays after penetrating the steel pipes is different, and images of different grayscales are formed on the imaging plate. If there are high-density inclusions inside the steel pipe, it will appear as a brighter area on the image; if there are cavities, the amount of radiation transmitted will increase, and it will appear as a darker area on the image. By analyzing these image data, the structural information inside the steel pipe can be obtained.

[0023] Based on these multi-mode detection data sets, the seamless steel pipe defect detection device locates the potential defect areas of the target seamless steel pipe. This process may involve complex data analysis and feature matching. For example, in the image detection information, through the pre-established defect-free steel pipe surface image model, when certain areas in the image are detected to be significantly different from the model (such as a sudden increase in texture complexity, abnormal color changes, etc.), these areas may be preliminarily located as potential defect areas. Then, the multi-mode detection characterization information corresponding to these potential defect areas is segmented from the multi-mode detection data set.

[0024] Go to step 102, determine the key detection information block from each multi-mode detection characterization information obtained by segmentation. This determination process needs to be carried out according to the characteristics and importance of each detection information. Taking image detection information as an example, for detecting surface defects, the pixel data block that may be related to the defect boundary and defect center area is the key detection information block; for ultrasonic detection information, the data such as the emission time and reflected wave intensity corresponding to the abnormal peak of the reflected wave constitute the key detection information block; for eddy current detection information, the eddy current distribution data corresponding to the time period when the impedance of the detection coil changes significantly is the key detection information block; for X-ray detection information, the part of the data in the image with obvious grayscale changes corresponding to the potential defect area is the key detection information block.

[0025] After determining the key detection information blocks, it is also necessary to determine the key detection attribute vectors of each key detection information block at each detection attention level in the set several detection attention levels. These detection attention levels can be set according to different detection purposes and data characteristics. For example, it can be set as a structural integrity level, a material uniformity level, a surface flatness level, etc. For each level, the key detection attribute vector is determined differently. Taking the structural integrity level as an example, for the key detection information block of image detection information, its key detection attribute vector may include attributes such as the shape characteristics of the defect area (such as circularity, aspect ratio, etc.), the contrast between the defect area and the surrounding area; for the key detection information block of ultrasonic detection information, the key detection attribute vector may include attributes such as the amplitude of the reflected wave anomaly and the position of the reflected wave anomaly area in the length direction of the steel pipe; for the key detection information block of eddy current detection information, the key detection attribute vector may include attributes such as the maximum value, minimum value and change gradient of the eddy current density change; for the key detection information block of ray detection information, the key detection attribute vector may include attributes such as the area of ​​the defect area in the image and the grayscale mean.

[0026] In step 103, for each of the several detection attention levels, the defect state characterization vector of the potential defect area at the detection attention level is determined according to the key detection attribute vector of each key detection information block at the detection attention level. For example, at the structural integrity level, the key detection attribute vectors of image detection information, ultrasonic detection information, eddy current detection information and radiographic detection information have been determined. For the shape feature in the key detection attribute vector of the image detection information, if the shape of the defect area is irregular (such as the aspect ratio is much greater than 1), this may indicate that there is a certain risk at the structural integrity level. This shape feature is added to the defect state characterization vector with a certain weight (the weight can be determined based on a large number of experiments and data analysis, such as the weight of the shape feature is 0.3); for the abnormal amplitude of the reflected wave in the key detection attribute vector of the ultrasonic detection information, if the abnormal amplitude of the reflected wave is large (for example, exceeding 50% of the normal amplitude of the reflected wave), this also indicates that there may be problems with the structural integrity. This abnormal amplitude of the reflected wave is added to the defect state characterization vector with a certain weight (such as 0.2); for the maximum value of the eddy current density change in the key detection attribute vector of the eddy current detection information, if this maximum value exceeds the set threshold (for example, 1.5 times the normal eddy current density), it is added to the defect state characterization vector with a weight of 0.2; for the area of ​​the defect area in the key detection attribute vector of the radiographic detection information, if the area is large (such as exceeding 10% of the cross-sectional area of ​​the steel pipe), it is added to the defect state characterization vector with a weight of 0.3. In this way, the key detection attribute vectors of each key detection information block at the structural integrity level are integrated to construct the defect state representation vector of the potential defect area at the structural integrity level.

[0027] In step 104, based on the defect state characterization vectors of the potential defect area at each detection attention level in several detection attention levels, the defect detection discriminant vectors corresponding to the potential defect area under the set multiple defect labels are determined. For example, the set defect labels are "cracks", "inclusions", "holes", etc. Taking the "crack" defect label as an example, in the defect state characterization vector at the structural integrity level, if the shape feature is displayed as a slender shape (such as an aspect ratio greater than 10), the abnormal amplitude of the ultrasonic detection reflection wave is large and continuous in a certain direction, the eddy current detection eddy current density changes linearly, and the ray detection image shows a linear grayscale change area, these features combined will make the defect detection discriminant vector under the "crack" defect label have a higher value. For the "inclusion" defect label, at the material uniformity level, if the ultrasonic detection reflection wave has multiple abnormal points in a small range, the eddy current detection eddy current density has discrete changes in a local area, and the ray detection image shows a small bright spot area, the defect detection discriminant vector under the "inclusion" defect label has a higher value. For each defect label, the corresponding defect detection discriminant vector is constructed by integrating the defect state representation vectors at each detection attention level.

[0028] Finally, in step 105, based on the defect detection discriminant vectors corresponding to the potential defect area under multiple defect labels, the defect detection viewpoint corresponding to the potential defect area is output. For example, if the value of the defect detection discriminant vector under the "crack" defect label is much higher than the values ​​of the discriminant vectors under other defect labels, then the output defect detection viewpoint may be "there is a crack defect in the potential defect area of ​​the target seamless steel pipe". If the values ​​of the defect detection discriminant vectors under the "inclusion" and "hole" defect labels are relatively close and higher than other labels, the defect detection viewpoint such as "there may be inclusions or hole defects in the potential defect area of ​​the target seamless steel pipe, which needs further confirmation" may be output.

[0029] Through such a complete process, the seamless steel pipe defect detection device can integrate multi-mode detection data, analyze and judge the potential defect areas of the seamless steel pipe from multiple levels and angles, and thus obtain more accurate and reliable defect detection results, which will help improve the quality inspection efficiency and accuracy of seamless steel pipes and ensure the safety and reliability of seamless steel pipes in various application fields (such as petrochemicals, building structures, etc.).

[0030] In the entire seamless steel pipe defect detection process, step 105 plays a key role in the final decision and output. It outputs the defect detection viewpoint corresponding to the potential defect area based on the defect detection discriminant vectors corresponding to the potential defect area under multiple defect labels. This process involves comprehensive consideration of the information obtained and processed in the previous multiple steps to draw an accurate and practical conclusion about the potential defects of the seamless steel pipe.

[0031] First, we need to deeply understand the nature and significance of the defect detection discriminant vectors. These discriminant vectors are constructed in step 104 based on the defect state representation vectors of the potential defect area at each detection attention level. For each set defect label (such as "crack", "inclusion", "hole", "corrosion", etc.), there is a corresponding defect detection discriminant vector. These vectors contain a lot of information about the relationship between the potential defect area and each defect label, and are a quantitative representation of the multi-mode detection data after layer-by-layer analysis and processing.

[0032] To illustrate with a practical example, for example, there is a seamless steel pipe with a diameter of 20 cm and a length of 5 meters. In the previous inspection process, for the defect label of "crack", at the structural integrity level, its defect state representation vector reflects that the potential defect area in the image detection information presents a slender shape, and the morphological characteristics similar to cracks are clearly reflected in the shape attributes; the ultrasonic detection information shows that there are continuous reflection wave anomalies in a specific direction, and the anomaly amplitude is large; the eddy current detection information shows that the eddy current density changes in the corresponding area present a specific pattern related to the existence of cracks; the image of the X-ray detection information shows linear grayscale changes, which is consistent with the performance of cracks in the image. The defect detection discriminant vector under the "crack" defect label constructed by combining this information will contain higher values, which are the result of the comprehensive effect of various detection information under this defect label.

[0033] For other defect labels, such as "inclusions", in the defect state characterization vector at the material uniformity level, if there are multiple scattered small reflection wave anomalies in the ultrasonic detection information, this may indicate the presence of small inclusions inside the steel pipe; the eddy current detection information shows that the eddy current density has discrete changes in a small local area, which corresponds to the local changes in the eddy current distribution caused by the presence of inclusions; the X-ray detection information shows some small, irregular bright spot areas in the image, which may be the manifestation of inclusions in the X-ray image. The defect detection discriminant vector under the "inclusion" defect label constructed based on this information will also have corresponding values.

[0034] When outputting defect detection opinions, the detection device needs to compare and analyze the defect detection discriminant vectors under these different defect labels. A common analysis method is to set a threshold or use a relative comparison method. For example, if the value of the defect detection discriminant vector under a certain defect label exceeds a preset threshold, it can be preliminarily determined that this type of defect exists in the potential defect area. The setting of this threshold is based on a large amount of experimental data, historical detection experience, and comprehensive consideration of the quality requirements of seamless steel pipes.

[0035] Continuing with the above example, if the value of the defect detection discriminant vector under the "crack" defect label exceeds 0.8 (the 0.8 here is a threshold value, which needs to be determined according to the specific situation in actual application), and the discriminant vector values ​​under other defect labels are far lower than this value, such as the discriminant vector value of "inclusion" is 0.2, the discriminant vector value of "hole" is 0.1, and the discriminant vector value of "corrosion" is 0.05, etc., then the defect detection viewpoint "there is a crack defect in the potential defect area of ​​the target seamless steel pipe" can be output with relative certainty.

[0036] However, the situation is not always so clear-cut. Sometimes there may be a situation where the discriminant vector values ​​of defect detection under multiple defect labels are relatively close. For example, the discriminant vector value of "inclusion" is 0.6, and the discriminant vector value of "hole" is 0.55. At this time, a more cautious analysis is required. The detection device may further check the situation of each component in each discriminant vector (that is, the contribution from different levels of detection attention). For example, in the discriminant vector of "inclusion", the component at the material uniformity level contributes more, while in the discriminant vector of "hole", the component at the structural integrity level has a certain influence. At this time, the detection device may consider re-examining the relevant detection data, or combining some additional information (such as the production process and usage history of the steel pipe) to make a more accurate judgment.

[0037] In some cases, the values ​​of the defect detection discriminant vectors may be relatively low. This may mean that although the potential defect area has been detected, the current test data is not sufficient to clearly determine it as a specific defect type. At this time, the output defect detection view may be "there are suspected defects in the potential defect area of ​​the target seamless steel pipe, which requires further testing or analysis." This provides a direction for subsequent testing work, such as using higher-resolution testing equipment, increasing the number of tests, or changing test parameters.

[0038] In addition, the output of the defect detection viewpoint is not just a simple determination of what kind of defect exists, but may also include a preliminary assessment of the severity of the defect. This can be achieved by comparing the value of the defect detection discriminant vector with the preset severity grading standard. For example, if the value of the defect detection discriminant vector under the "crack" defect label is very high (such as exceeding 0.95), it may indicate that the crack is relatively serious, which can be expressed in the output defect detection viewpoint as "there is a serious crack defect in the potential defect area of ​​the target seamless steel pipe"; and if the value is between 0.8-0.9, it can be expressed as "there is a more obvious crack defect", etc.

[0039] In actual industrial application scenarios, the quality requirements for seamless steel pipes are very strict, especially in some key fields such as aerospace, petrochemicals, etc. Therefore, the defect detection view output in step 105 has a crucial impact on subsequent decision-making. If it is determined that there are serious defects, the steel pipe may be judged as unqualified and need to be repaired or scrapped; if it is a suspected defect, it may prompt the initiation of further inspection processes to ensure that the quality of the steel pipe meets the requirements.

[0040] From the algorithmic point of view, in order to output defect detection opinions more accurately, some classification algorithms or decision models can be used. For example, a probability-based classification algorithm can be used to regard each defect detection discriminant vector as a probability representation of the existence of a certain defect. Then, according to probability calculation methods such as Bayes' theorem, the probability of all defect labels is comprehensively considered to obtain the most likely defect type. This method can handle the uncertainty and ambiguity of data to a certain extent.

[0041] Alternatively, a decision tree model can be used to take the components in the defect detection discriminant vector as the input features of the decision tree, and through pre-built decision tree rules (these rules are trained based on a large amount of sample data), the defect detection opinions are gradually judged and finally output. The advantage of the decision tree model is that it is highly interpretable, and the impact of each feature on the final result can be clearly seen.

[0042] When processing multi-mode inspection data, the weight allocation of different inspection methods will also affect the output of defect detection opinions. For example, in some cases, X-ray inspection information may be more accurate and reliable for detecting internal defects. In this case, a higher weight can be given to X-ray inspection information when constructing defect detection discriminant vectors and outputting defect detection opinions. This requires comprehensive evaluation and setting based on the accuracy, reliability, and detection capabilities of different inspection methods for different defect types.

[0043] It can be seen that step 105 is a complex process of combining multi-mode inspection data, multiple inspection attention levels and different defect labels for analysis and decision-making. It needs to take into account multiple factors such as quantitative representation of data, threshold setting, processing of different situations and combination with actual application scenarios to output accurate and useful defect detection views, thereby providing strong support for quality control and safe use of seamless steel pipes.

[0044] In summary, this application uses data from multiple detection methods (image, ultrasound, eddy current, and X-ray detection) to comprehensively locate potential defect areas, overcome the limitations that may exist in a single detection method, and improve the accuracy of detection. Operations such as determining key detection information blocks and constructing key detection attribute vectors from multi-mode detection data can deeply mine the effective information of the data and more accurately describe the status of potential defect areas. By constructing defect detection discriminant vectors through multiple detection attention levels and defect labels and outputting detection opinions, defects can be classified and evaluated in detail, which helps to provide an accurate basis for the quality control of seamless steel pipes, ensure their safety and reliability in use in various industries, and at the same time improve detection efficiency and reduce cost risks caused by misjudgment or omission of defects.

[0045] In some exemplary technical ideas, the multimode detection characterization information corresponding to the potential defect area is segmented from the multimode detection data set of the target seamless steel pipe, including: determining the regional positioning mask and the data mask associated with the potential defect area; extracting the surface detection results belonging to the regional positioning mask and pointed to by the data mask from the multimode detection data set of the target seamless steel pipe as the multimode detection characterization information associated with the potential defect area.

[0046] In the technical field of seamless steel pipe defect detection, the technical solution of the present application revolves around segmenting multi-mode detection characterization information corresponding to potential defect areas from a multi-mode detection data set of a target seamless steel pipe.

[0047] First, determining the region location mask and data mask associated with the potential defect area is a key starting step. The region location mask can be regarded as a location mark of the potential defect area in the entire detection data space. For example, when inspecting a seamless steel pipe with a length of 5 meters and a diameter of 10 centimeters, if the inspection data set represents the inspection data of each position of the steel pipe in three-dimensional coordinates (length direction, circumferential direction, and pipe diameter direction), the region location mask may be a three-dimensional matrix, in which the value of the coordinate position corresponding to the potential defect area is 1, and the other positions are 0. The size of this matrix depends on the resolution of the inspection data and the size of the steel pipe. For example, if the resolution is one data point per centimeter, then there are 500 data points in the length direction, if the circumferential direction is divided into 360 degrees and one data point per degree, and if the pipe diameter direction is divided into 10 layers (representing inspection data of different depths), then the region location mask matrix is ​​500×360×10 in size.

[0048] The data mask plays a role of pointing and screening. It determines which data is relevant and valid to the potential defect area based on the characteristics and requirements of the detection data. For example, in the image detection information, if the image is in color, the data mask may filter out the data channels related to the color features to analyze the color changes in the potential defect area. For example, if the color image uses the RGB color mode, the data mask may select the data of the R channel (red channel) and the G channel (green channel), because in the detection of a certain type of specific defects (such as surface oxidation), the data of these two channels are more critical for judging defects. For ultrasonic detection information, the data mask may select the data part related to the reflection wave intensity and reflection time. If the sampling frequency of ultrasonic detection is 5MHz, the reflection wave data received within a specific time window (such as 0.1 milliseconds to 1 millisecond) may be selected by the data mask, because the reflection wave data within this time window is more correlated with the potential defect area.

[0049] From the multi-mode detection data set of the target seamless steel pipe, the surface detection results belonging to the regional positioning mask and pointed to by the data mask are extracted as the multi-mode detection characterization information associated with the potential defect area. Take eddy current detection as an example. For example, when the above-mentioned seamless steel pipe is subjected to eddy current detection, the detection data contains the eddy current density distribution data of the entire steel pipe surface. The regional positioning mask determines the position of the potential defect area on the surface of the steel pipe, and the data mask points to the data part where the eddy current density variation exceeds a certain threshold (such as 1.5 times the normal eddy current density). Then, what is extracted from the multi-mode detection data set are the data within the potential defect area and the eddy current density variation exceeds 1.5 times the normal eddy current density. These data constitute the eddy current detection part of the multi-mode detection characterization information associated with the potential defect area.

[0050] For X-ray detection information, the same extraction principle is followed. If the image data resolution obtained by X-ray detection is one pixel per millimeter, and the image size is 500 mm × 500 mm (corresponding to a certain detection area of ​​the steel pipe), the regional positioning mask determines the position range of the potential defect area in this image, and the data mask may select data related to the grayscale value change rate. For example, if the data mask selects pixel data with a grayscale value change rate exceeding 10%, then what is extracted from the X-ray detection data set is the pixel data in the potential defect area and with a grayscale value change rate exceeding 10%, and this data becomes the X-ray detection part of the multi-mode detection characterization information.

[0051] In terms of image detection information, for example, a high-resolution industrial camera is used to photograph the surface of a steel pipe, and the image resolution is 1000×1000 pixels. The region positioning mask determines the location of the potential defect area in the image, and the data mask selects data related to texture features. For example, the data mask selects the data of the pixel area whose texture complexity exceeds a certain threshold (the threshold obtained by analyzing the texture of a large number of defect-free steel pipe images). Then, what is extracted from the image detection data set is the pixel data in the potential defect area whose texture complexity exceeds the threshold. These data constitute the image detection part of the multi-mode detection representation information.

[0052] For ultrasonic testing, for example, the ultrasonic probe detects the steel pipe at a certain interval (such as every 2 cm). After determining the regional positioning mask and data mask, the data related to the reflected wave characteristics (such as the reflected wave amplitude exceeding the normal amplitude by 30%) in the potential defect area are extracted from the ultrasonic testing data set. These data become the ultrasonic testing part of the multi-mode detection characterization information.

[0053] The multi-mode detection characterization information obtained in this way contains data closely related to potential defect areas in various aspects of image detection, ultrasonic detection, eddy current detection and X-ray detection. These data are obtained through careful screening and positioning, and can more accurately reflect the characteristics of potential defect areas, laying a solid foundation for the subsequent determination of key detection information blocks from the various multi-mode detection characterization information obtained from the segmentation. In this way, subsequent steps can operate on these more targeted data. For example, when determining the key detection information blocks, the more critical parts for defect characterization can be found in these screened data, and then the key detection attribute vectors at each detection attention level can be determined. This series of operations makes the entire seamless steel pipe defect detection process more accurate and efficient, and can more accurately determine the status of potential defect areas under each defect label, and finally output accurate defect detection views.

[0054] In this way, by determining the regional positioning mask and data mask to extract the multi-mode detection characterization information, it is possible to accurately locate and filter the data related to the potential defect area. In terms of numerical values, the data is filtered according to specific thresholds under different detection methods (such as the amplitude of the ultrasonic detection reflection wave exceeds the normal amplitude by 30%, etc.), which improves the validity of the data. This method avoids the interference of irrelevant data and makes subsequent operations such as determining key detection information blocks and constructing key detection attribute vectors more targeted. This improves the accuracy of the entire seamless steel pipe defect detection process, reduces misjudgments and missed judgments, and provides a more reliable basis for the quality control of seamless steel pipes.

[0055] In an optional embodiment, each multi-mode detection characterization information obtained by segmentation is used as target multi-mode detection characterization information, and the step of determining a key detection information block from the target multi-mode detection characterization information includes: performing modal decomposition processing on the target multi-mode detection characterization information to obtain at least one single-mode detection characterization information included in the target multi-mode detection characterization information; for each single-mode detection characterization information in the at least one single-mode detection characterization information, determining a set key detection information block paired with the targeted single-mode detection characterization information in a key detection information pool, and using the set key detection information block paired with the targeted single-mode detection characterization information as the key detection information block determined from the target multi-mode detection characterization information.

[0056] Furthermore, determining whether there is a set key detection information block paired with the targeted single-mode detection characterization information in the key detection information pool includes: determining the detection feature difference between the targeted single-mode detection characterization information and the set key detection information block in the key detection information pool; when there is a target set key detection information block in the key detection information pool whose detection feature difference satisfies the first feature difference index, using the target set key detection information block as the set key detection information block paired with the targeted single-mode detection characterization information.

[0057] In the framework of seamless steel pipe defect detection technology, the technical solution of this embodiment revolves around determining key detection information blocks from each multi-mode detection characterization information obtained by segmentation.

[0058] Multi-mode detection characterization information includes data from multiple detection methods (such as image detection, ultrasonic detection, eddy current detection, and X-ray detection). These multi-mode detection characterization information are processed as target multi-mode detection characterization information. First, modal decomposition is performed. This process aims to decompose complex multi-mode detection characterization information into at least one single-mode detection characterization information that is easier to analyze. For example, for a certain target multi-mode detection characterization information, if it contains data from multiple modes such as texture data in image detection, reflected wave data in ultrasonic detection, eddy current density data in eddy current detection, and image grayscale data in X-ray detection, after modal decomposition processing, single-mode detection characterization information corresponding to image detection, ultrasonic detection, eddy current detection, and X-ray detection will be obtained.

[0059] After obtaining the single-mode detection characterization information, for each single-mode detection characterization information, search for the set key detection information block that matches it in the key detection information pool. Taking the single-mode detection characterization information of image detection as an example, this information may include the texture features and color features of the steel pipe surface. The key detection information pool is a pre-built resource library that stores various possible key detection information blocks. For image detection, the set key detection information blocks in the key detection information pool may include texture feature blocks of specific shapes (such as linear texture feature blocks related to scratches), color feature blocks of specific color change patterns, etc.

[0060] When determining whether there is a set key detection information block paired with the single-mode detection characterization information in the key detection information pool, it is necessary to determine the detection feature difference between the single-mode detection characterization information and the set key detection information block in the key detection information pool. For example, for the texture feature part in the single-mode detection characterization information of image detection, it is compared with the texture feature block in the key detection information pool. For example, when detecting a seamless steel pipe with a length of 5 meters and a diameter of 10 centimeters, the texture features in the collected image detection data are represented in a certain quantitative way (such as texture complexity value, texture direction angle value, etc.). A texture feature block in the key detection information pool is represented as a specific texture complexity range (such as between 3-5) and a texture direction angle range (such as an angle of 30-60 degrees with the horizontal direction). When the texture feature value in the single-mode detection characterization information is within this range, the detection feature difference meets the first feature difference index.

[0061] Taking the single-mode detection characterization information of ultrasonic testing as an example, it contains data such as the intensity and reflection time of ultrasonic reflected waves. The set key detection information block in the key detection information pool may be the reflection wave intensity and reflection time pattern for a specific type of defect (such as internal holes). For example, the sampling frequency of ultrasonic testing is 5MHz. For normal steel pipes, the reflection wave intensity is within a certain range (such as 1-2 volts) and the reflection time is also in a specific interval (such as 0.1-0.5 milliseconds). When the single-mode detection characterization information of seamless steel pipes is detected, the reflection wave intensity is 1.5 volts and the reflection time is 0.3 milliseconds, and there is a set key detection information block in the key detection information pool indicating that the reflection wave intensity is 1.4-1.6 volts and the reflection time is 0.2-0.4 milliseconds, which satisfies the situation that the detection feature difference meets the first feature difference index, and this set key detection information block is determined as the key detection information block paired with the single-mode detection characterization information of the ultrasonic test.

[0062] For the single-mode detection characterization information of eddy current detection, for example, it contains the distribution data of eddy current density. The set key detection information block in the key detection information pool may be the eddy current density change pattern for the internal structural non-uniformity of the steel pipe. If during eddy current detection of seamless steel pipes, the eddy current density change in a certain area is detected (expressed as the rate of change of eddy current density relative to normal density, for example, the normal eddy current density is 100 units, the eddy current density in the detection area is 120 units, and the rate of change is 20%), and there is a set key detection information block in the key detection information pool indicating that the rate of change of eddy current density is between 15% and 25%, this indicates that the detection feature difference meets the first feature difference index, and the set key detection information block becomes the key detection information block paired with the single-mode detection characterization information of the eddy current detection.

[0063] For single-mode detection characterization information of X-ray detection, such as grayscale value, grayscale change rate and other data of the image. The set key detection information block in the key detection information pool may be an image grayscale feature block for specific defects (such as internal inclusions). For example, the resolution of the X-ray detection image is one pixel per millimeter, the grayscale value of a certain area is 150 (expressed in the grayscale value range of 0-255), and the grayscale change rate is 10%. There is a set key detection information block in the key detection information pool indicating that the grayscale value is between 140-160 and the grayscale change rate is between 8%-12%. This meets the requirement that the detection feature difference meets the first feature difference index. The set key detection information block is determined as the key detection information block paired with the single-mode detection characterization information of the X-ray detection.

[0064] In this way, key detection information blocks are determined from the target multi-mode detection characterization information. These key detection information blocks will provide a basis for subsequent operations such as determining key detection attribute vectors, so that the entire seamless steel pipe defect detection process can more accurately analyze and evaluate potential defect areas.

[0065] In this way, by decomposing the multi-mode detection characterization information into single-mode detection characterization information through modal disassembly processing, the data of different detection methods can be analyzed more carefully. When determining the key detection information block, the method of matching with the key detection information pool can accurately locate the key detection information block that matches the single-mode detection characterization information based on the detection feature difference and the first feature difference index. The accuracy of this matching method is illustrated by specific numerical examples (such as numerical comparisons of ultrasonic detection reflection wave intensity, eddy current density change rate, and X-ray detection grayscale value). This helps to improve the accuracy of subsequent analysis of potential defect areas, reduce misjudgments caused by improper data processing, and provide a more reliable basis for defect detection of seamless steel pipes.

[0066] In the actual application process, the key detection attribute vectors of each of the determined key detection information blocks at each of the set detection attention levels specifically include: for each of the determined key detection information blocks, obtaining the preset key detection attribute vectors of the set key detection information blocks paired with the targeted key detection information blocks at each of the set detection attention levels; based on the obtained preset key detection attribute vectors, determining the key detection attribute vectors of the targeted key detection information blocks at each of the detection attention levels of the set detection attention levels.

[0067] In the seamless steel pipe defect detection technology system, this embodiment focuses on determining the key detection attribute vectors of each determined key detection information block at several set detection attention levels.

[0068] First, for each key detection information block that has been determined, it is necessary to obtain the preset key detection attribute vector of each detection attention level of the set key detection information block matched with the key detection information block. Take the key detection information block previously determined in the image detection as an example, for example, this key detection information block is a texture feature block related to the scratches on the surface of the steel pipe. In the set detection attention level, there may be a surface texture integrity level, a color contrast level, etc. For the set key detection information block (pre-stored and defined) matched with the key detection information block, the preset key detection attribute vector at the surface texture integrity level may include the regularity of the texture (expressed as a numerical value, for example, the regularity of the texture is 0.8, 0 means completely irregular, 1 means completely regular), the consistency of the texture direction (for example, expressed as an angle deviation range, such as a deviation within 10 degrees means good consistency, the value is 0.9) and other attributes. The preset key detection attribute vector at the color contrast level may include the color contrast value with the surrounding area (such as the contrast is 0.6, 0 means no contrast, 1 means maximum contrast) and other attributes.

[0069] For the key detection information blocks determined in ultrasonic testing, for example, the key detection information blocks of reflection wave characteristics related to the internal holes of steel pipes. At the structural integrity detection attention level, the preset key detection attribute vector of the paired key detection information blocks may include attributes such as the abnormal amplitude of the reflection wave (for example, expressed as a multiple relative to the normal reflection wave amplitude, such as the abnormal amplitude is 1.5 times the normal amplitude), the length of the abnormal reflection wave area (in centimeters, such as the length of the abnormal area is 2 centimeters). At the material uniformity detection attention level, it may include attributes such as the fluctuation frequency of the reflection wave in the abnormal area (for example, 10 fluctuations per second).

[0070] In terms of eddy current testing, for example, the key detection information block determined is the key detection information block of eddy current density change related to the uneven internal structure of the steel pipe. At the eddy current distribution uniformity detection attention level, the preset key detection attribute vector of the paired key detection information block may include the maximum value of the eddy current density change (for example, the normal eddy current density is 100 units and the maximum value is 130 units), the minimum value (such as 90 units) and the gradient of the eddy current density change (such as a change of 5 units per centimeter). At the energy loss detection attention level, it may include attributes such as the proportion of energy loss caused by the change in eddy current density (such as 0.1, i.e. 10% energy loss).

[0071] For the key detection information blocks determined in the radiographic detection, for example, the image grayscale key detection information blocks related to the internal inclusions of the steel pipe. At the image clarity detection attention level, the preset key detection attribute vector of the paired key detection information block may include the blur degree of the image of the inclusion area (expressed as a blur coefficient, such as a blur coefficient of 0.3, 0 means completely clear, 1 means completely blurry), the smoothness of the grayscale change (such as a smoothness value of 0.7, 0 means completely unsmooth, 1 means completely smooth) and other attributes. At the internal structure display integrity detection attention level, it may include attributes such as the area ratio of inclusions in the image (such as 0.05, i.e. 5% area ratio).

[0072] Based on the obtained preset key detection attribute vector, the key detection attribute vector of the key detection information block at each detection attention level in several detection attention levels is determined. For the key detection information block of scratch texture features in image detection, at the surface texture integrity level, since the texture regularity of the paired key detection information block is set to 0.8, the texture regularity of this key detection information block in the key detection attribute vector at this level will also be close to 0.8 (it may be fine-tuned according to specific adjustment rules, such as adjustment according to the confidence of the detection data. If the confidence is high, it is directly adopted. If the confidence is low, it is adjusted within a certain range. For example, here the confidence is high and 0.8 is directly adopted). At the color contrast level, because the contrast of the key detection information block is set to 0.6, the contrast of the key detection attribute vector of this key detection information block at this level is also 0.6.

[0073] For the key detection information block of hole reflection wave characteristics in ultrasonic testing, at the structural integrity detection attention level, since the abnormal amplitude of the reflection wave of the paired key detection information block is set to 1.5 times the normal amplitude, the abnormal amplitude of the reflection wave in the key detection attribute vector of this key detection information block at this level is also 1.5 times the normal amplitude. At the material uniformity detection attention level, the reflection wave fluctuation frequency of the key detection information block is set to 10 times per second, so the fluctuation frequency of this key detection information block in the key detection attribute vector at this level is also 10 times per second.

[0074] In the key detection information block of structural non-uniform eddy current density change in eddy current detection, at the eddy current distribution uniformity detection attention level, since the maximum value of the eddy current density change of the key detection information block is set to 130 units, the minimum value is 90 units, and the gradient is 5 units per centimeter, then the maximum value, minimum value and gradient of the eddy current density change in the key detection attribute vector of the key detection information block at this level are also 130 units, 90 units and 5 units per centimeter respectively. At the energy loss detection attention level, the energy loss ratio of the key detection information block is set to 0.1, and the energy loss ratio of the key detection attribute vector of the key detection information block at this level is also 0.1.

[0075] For the inclusion image grayscale key detection information block in the X-ray detection, at the image clarity detection attention level, because the fuzziness coefficient of the key detection information block is set to 0.3, the fuzziness coefficient of the key detection information block in the key detection attribute vector at this level is also 0.3. At the internal structure display integrity detection attention level, the inclusion area ratio of the key detection information block is set to 0.05, and the inclusion area ratio of the key detection information block in the key detection attribute vector at this level is also 0.05.

[0076] In this way, the key detection attribute vector can be accurately determined for each key detection information block at different detection attention levels, thereby laying a solid foundation for the subsequent determination of the defect state characterization vector of the potential defect area at each detection attention level, making the entire seamless steel pipe defect detection process more accurate and orderly.

[0077] Therefore, the key detection attribute vector of the key detection information block itself is determined by obtaining the preset key detection attribute vector of the set key detection information block at each detection attention level that is paired with the key detection information block. This method has strong reference and accuracy. Numerical examples (such as texture regularity, reflection wave abnormal amplitude, eddy current density change value, inclusion area ratio, etc.) in different detection methods (image, ultrasound, eddy current, and X-ray detection) show that it can effectively utilize the pre-defined standard attribute vector, avoid the blindness in the attribute vector determination process, and improve the accuracy of the description of the key detection information block attributes, thereby providing a strong guarantee for the accuracy of the entire seamless steel pipe defect detection.

[0078] In some preferred embodiments, for each of the several detection attention levels, the defect state characterization vector of the potential defect area at the targeted detection attention level is determined according to the key detection attribute vector of each of the key detection information blocks at the targeted detection attention level, including: for each of the several detection attention levels, the fully connected state attribute vector of the potential defect area at the targeted detection attention level is determined according to the key detection attribute vector of each of the key detection information blocks at the targeted detection attention level; based on the fully connected state attribute vector of the potential defect area at the targeted detection attention level, the defect state characterization vector of the potential defect area at the targeted detection attention level is determined.

[0079] In the framework of the seamless steel pipe defect detection technical solution, this preferred embodiment focuses on how to determine the defect state characterization vector of the potential defect area at the detection attention level based on the key detection attribute vector of the key detection information block.

[0080] First, for each of the several detection attention levels, the fully connected state attribute vector of the potential defect area at the level should be determined based on the key detection attribute vector of each key detection information block at the detection attention level. Taking image detection as an example, at the detection attention level of surface texture integrity, there are two key detection information blocks: one is the texture feature key detection information block related to scratches, whose texture regularity degree in the key detection attribute vector is 0.3 (indicating that the texture is relatively irregular) and texture direction consistency is 0.6 (indicating that the direction consistency is general); the other is the key detection information block related to the texture change around the pit, whose texture regularity degree is 0.2 and texture direction consistency is 0.5. When determining the fully connected state attribute vector, it is necessary to comprehensively consider the attribute vectors of these two key detection information blocks.

[0081] For ultrasonic testing, under the structural integrity testing attention level, for example, there are three key detection information blocks. The first is the key detection information block of the reflection wave characteristics related to the internal holes, whose reflection wave abnormal amplitude is 1.5 times the normal amplitude and the reflection wave abnormal area length is 3 cm; the second is the key detection information block related to the uneven thickness of the steel pipe wall, whose reflection wave fluctuation frequency in the abnormal area is 8 times per second, and the reflection wave abnormal amplitude is 1.2 times the normal amplitude; the third is the key detection information block related to the small cracks, whose reflection wave abnormal amplitude is 1.3 times the normal amplitude and the reflection wave abnormal area length is 1.5 cm. When constructing the fully connected state attribute vector, the attribute vectors of these key detection information blocks will be comprehensively considered. For example, the values ​​of the reflection wave abnormal amplitude will be combined in some way, which may be the average value ((1.5+1.2+1.3) / 3=1.33 times the normal amplitude), and the reflection wave abnormal area length will also be similarly combined ((3+1.5) / 2=2.25 cm, where the key detection information block of the small crack has a lower weight, so it is calculated proportionally when calculating the average length).

[0082] At the eddy current distribution uniformity detection attention level of eddy current detection, for example, there are two key detection information blocks. One is the eddy current density change key detection information block related to internal structural unevenness, with a maximum eddy current density change of 130 units, a minimum of 90 units, and a gradient of 5 units per centimeter; the other is the key detection information block related to metal impurities, with a maximum eddy current density change of 140 units, a minimum of 80 units, and a gradient of 6 units per centimeter. When determining the fully connected state attribute vector, the maximum value of the eddy current density change may take the maximum value of the two (140 units), the minimum value takes the minimum value (80 units), and the gradient may be calculated based on the influence weights of the two (for example, if the weights of the two are the same, the gradient is (5+6) / 2=5.5 units per centimeter).

[0083] For ray detection, at the image clarity detection attention level, for example, there are two key detection information blocks. One is the image grayscale key detection information block related to internal inclusions, with a fuzzy coefficient of 0.3 and a grayscale change smoothness of 0.7; the other is the key detection information block related to internal voids, with a fuzzy coefficient of 0.4 and a grayscale change smoothness of 0.6. When constructing the fully connected state attribute vector, the fuzzy coefficients may be combined according to a certain rule (such as weighted average, for example, the weight of the key detection information block of inclusions is 0.6, and that of voids is 0.4, then the combined fuzzy coefficient is 0.3×0.6+0.4×0.4=0.34), and the smoothness of grayscale changes is similarly calculated (0.7×0.6+0.6×0.4=0.66).

[0084] According to the fully connected state attribute vector of the potential defect area at the targeted detection attention level, the defect state representation vector of the potential defect area at the detection attention level is determined. Continuing with the surface texture integrity level of image detection as an example, for example, the texture regularity degree in the fully connected state attribute vector is 0.25, and the texture direction consistency is 0.55. Based on pre-set rules, the situation where the texture regularity degree is lower than 0.5 and the texture direction consistency is lower than 0.6 may be defined as a serious surface texture integrity problem. Then in the defect state representation vector, this state will be represented in a specific encoding method, such as using binary encoding 01 (0 for normal, 1 for problems) to indicate that there is a problem with the texture regularity degree, and using 01 to indicate that there is a problem with the texture direction consistency.

[0085] At the structural integrity detection attention level of ultrasonic testing, for example, the abnormal amplitude of the reflected wave in the fully connected state attribute vector is 1.33 times the normal amplitude, and the length of the abnormal area of ​​the reflected wave is 2.25 cm. According to the predetermined standard, if the abnormal amplitude of the reflected wave exceeds 1.2 times the normal amplitude and the length of the abnormal area exceeds 2 cm, it means that there is a greater risk of structural integrity. In the defect state characterization vector, 01 may be used to indicate that there is a risk of abnormal amplitude of the reflected wave, and 01 may be used to indicate that there is a risk of abnormal length of the reflected wave.

[0086] For the eddy current distribution uniformity detection of eddy current detection, the maximum value of the eddy current density change in the fully connected state attribute vector is 140 units, the minimum value is 80 units, and the gradient is 5.5 units per centimeter. According to the pre-setting, if the maximum value of the eddy current density change exceeds 130 units, the minimum value is less than 90 units, and the gradient exceeds 5 units, it means that there is a problem with the uniformity of the eddy current distribution. In the defect state characterization vector, 01 may be used to indicate that there is a problem with the maximum value, 01 indicates that there is a problem with the minimum value, and 01 indicates that there is a problem with the gradient.

[0087] At the image clarity detection attention level of X-ray detection, the fuzzy coefficient in the fully connected state attribute vector is 0.34, and the smoothness of grayscale change is 0.66. If the pre-set fuzzy coefficient exceeds 0.3 and the smoothness of grayscale change is lower than 0.7, it means that there is a problem with image clarity. In the defect state representation vector, 01 may be used to indicate that there is a problem with the fuzzy coefficient, and 01 may be used to indicate that there is a problem with the smoothness of grayscale change.

[0088] In this way, the conversion from the fully connected state attribute vector to the defect state representation vector can more accurately describe the state of the potential defect area at each detection attention level, and provide effective intermediate data for the subsequent determination of the defect detection discriminant vector of the potential defect area under different defect labels, thereby ensuring the accuracy and reliability of the entire seamless steel pipe defect detection process.

[0089] This design, by first determining the fully connected state attribute vector and then determining the defect state representation vector, can fully integrate the attribute information of each key detection information block at the detection attention level. Numerical examples in different detection methods (such as texture regularity, reflection wave abnormal amplitude, eddy current density change value, fuzzy coefficient, etc.) show that this method can accurately quantify and encode the state of potential defect areas at each detection attention level, making the description of the defect state more accurate and comprehensive, providing a more reliable basis for subsequent defect identification, and improving the accuracy of seamless steel pipe defect detection.

[0090] In one possible example, the defect label includes a defect category distribution label, and the defect detection discriminant vector of the potential defect area under the defect category distribution label includes a defect category distribution thermal vector; the defect detection discriminant vectors corresponding to the potential defect area under the set multiple defect labels are determined based on the defect state characterization vector of the potential defect area at each of the several detection attention levels, including: for each of the several detection attention levels, based on the defect state characterization vector of the potential defect area at each of the several detection attention levels and a set credibility coefficient condition, determining the defect category distribution thermal vector of the potential defect area under the defect category distribution label.

[0091] In the in-depth discussion of seamless steel pipe defect detection technology, this example focuses on the defect category distribution thermal vector in the defect detection discriminant vector that determines the potential defect area under the defect category distribution label.

[0092] First of all, it is clear that the defect category distribution label here covers various types of defects that may exist in seamless steel pipes, such as surface scratches, internal holes, inclusions, uneven wall thickness, etc. The defect detection discriminant vector of the potential defect area under the defect category distribution label contains the defect category distribution thermal vector, which will express the possibility or thermal distribution of the potential defect area belonging to different defect categories in a quantitative way.

[0093] For each of several detection attention levels, the defect category distribution thermal vector is determined based on the defect state representation vector of the potential defect area at these detection attention levels and the set credibility coefficient conditions. Take the detection attention level of surface texture integrity in image detection as an example. For example, in the defect state representation vector, the value of the texture regularity is 0.3 (indicating that the texture is relatively irregular), and the value of the texture direction consistency is 0.5 (indicating that the direction consistency is general). For the defect category of surface scratches, if the credibility coefficient conditions set based on historical data and experience show that when the texture regularity is lower than 0.5 and the texture direction consistency is lower than 0.6, the credibility of the surface scratch defect category increases.

[0094] At the structural integrity detection attention level of ultrasonic testing, for example, the abnormal amplitude of the reflected wave in the defect state characterization vector is 1.5 times the normal amplitude, and the length of the abnormal area of ​​the reflected wave is 3 cm. For the internal hole defect category, the set credibility coefficient condition may stipulate that when the abnormal amplitude of the reflected wave exceeds 1.2 times the normal amplitude and the length of the abnormal area exceeds 2 cm, the credibility of the internal hole defect category increases significantly. This increase in credibility will be reflected in the defect category distribution thermal vector, and a larger value may be added to the component corresponding to the internal hole.

[0095] At the eddy current distribution uniformity detection attention level of eddy current testing, for example, the maximum value of the eddy current density change in the defect state characterization vector is 130 units, the minimum value is 90 units, and the gradient is 5 units per centimeter. For the defect category of inclusions, if the set credibility coefficient condition is that the maximum value of the eddy current density change exceeds 120 units, the minimum value is less than 95 units, and the gradient exceeds 4 units, the credibility of the inclusion defect category is improved. This improvement will also be reflected in the defect category distribution thermal vector, and the corresponding numerical adjustment will be made on the component corresponding to the inclusion.

[0096] At the attention level of image clarity detection of X-ray detection, for example, the fuzziness coefficient in the defect state representation vector is 0.3 and the smoothness of grayscale change is 0.7. For the internal void defect category, the set credibility coefficient condition may be that when the fuzziness coefficient exceeds 0.3 and the smoothness of grayscale change is less than 0.8, the credibility of the internal void defect category increases, which in turn affects the component value corresponding to the internal void in the defect category distribution thermal vector.

[0097] When combining the information of each detection attention level, each level may be assigned different weights according to its importance in defect detection. For example, for some key detection attention levels, such as the structural integrity level of ultrasonic detection, a higher weight may be assigned because it is very critical for detecting internal defects. Some levels of image detection may have relatively low weights. When calculating the defect category distribution thermal vector, the credibility contribution of different defect categories obtained under the credibility coefficient conditions at each detection attention level will be accumulated according to their respective weights.

[0098] To illustrate with specific numerical values, for example, in the category of surface scratch defects, the credibility contribution of the image detection layer to the surface scratch calculated according to the above conditions is 0.2 (after weight adjustment), the credibility contribution of the ultrasonic detection layer is 0.05 because no strong correlation features directly related to the surface scratches are found, the credibility contribution of the eddy current detection layer is 0.1, and the credibility contribution of the X-ray detection layer is 0.05. After accumulating these values, the component value corresponding to the surface scratch in the defect category distribution thermal vector is 0.4.

[0099] For the internal hole defect category, for example, the credibility contribution of the image detection layer is 0.05, the credibility contribution calculated based on the reflection wave characteristics of the ultrasonic detection layer is 0.3 (after weight adjustment), the credibility contribution of the eddy current detection layer is 0.1, and the credibility contribution of the X-ray detection layer is 0.05. After accumulation, the component value of the internal hole in the defect category distribution thermal vector is 0.5.

[0100] In this way, similar calculations are performed for all set defect categories (such as inclusions, uneven wall thickness, etc.) to determine the defect category distribution thermal vector of the potential defect area under the defect category distribution label. This vector can intuitively show the degree of correlation or thermal distribution between the potential defect area and different defect categories, providing an important quantitative basis for the subsequent accurate judgment of the defect type of the potential defect area.

[0101] In this way, by determining the defect category distribution thermal vector based on the defect state characterization vector and the credibility coefficient condition, the information of each detection attention level can be effectively integrated. Numerical examples in different detection methods (such as texture regularity, reflection wave abnormal amplitude, eddy current density change value, fuzzy coefficient, etc.) show that this method can accurately quantify the degree of association between potential defect areas and different defect categories, making the judgment of defect types more objective and accurate, and providing strong support for the accuracy of seamless steel pipe defect detection results.

[0102] In some other possible technical ideas, the defect state characterization vector of the potential defect area at each of the several detection attention levels is used to determine the defect detection discrimination vectors corresponding to the potential defect area under the set multiple defect labels, including: for each of the several detection attention levels, determining the error quantization characteristics between the key detection attribute vectors of each of the key detection information blocks at the current detection attention level and the defect state characterization vectors at the current detection attention level; based on the error quantization characteristics determined at the current detection attention level, determining the defect detection discrimination vectors of the potential defect area corresponding to the set multiple defect labels at the current detection attention level.

[0103] In the first application scenario of the above technical ideas, the defect label includes a defect trend change label, and the defect detection discriminant vector of the potential defect area under the defect trend change label includes a defect trend change vector; the error quantization characteristics determined at the current detection attention level determine the defect detection discriminant vectors of the potential defect area corresponding to the set multiple defect labels at the current detection attention level, including: determining the discrete measurement value at the current detection attention level based on the error quantization characteristics determined at the current detection attention level and the determined number of key detection information blocks; determining the defect trend change vector of the potential defect area corresponding to the defect trend change label at the current detection attention level based on the discrete measurement value at the current detection attention level.

[0104] In the second application scenario of the above technical ideas, the defect label includes a crack state label, and the defect detection discriminant vector of the potential defect area under the crack state label includes a crack state embedding vector; the error quantization characteristics determined at the current detection attention level determine the defect detection discriminant vectors of the potential defect area corresponding to the set multiple defect labels at the current detection attention level, and also include: based on the error quantization characteristics determined at the current detection attention level, from each of the determined key detection information blocks, select the target key detection information block indicated by the error quantization characteristics that meets the set error requirements; based on the number of the target key detection information blocks, determine the crack area noise characteristics at the current detection attention level; based on the crack area noise characteristics at the current detection attention level, determine the crack state embedding vector of the potential defect area corresponding to the crack state label at the current detection attention level.

[0105] In the third application scenario of the above technical ideas, the defect label includes a dirt state label, and the defect detection discrimination vector of the potential defect area under the dirt state label includes a dirt state description vector; the error quantization feature determined at the current detection attention level determines the defect detection discrimination vector of the potential defect area corresponding to the set multiple defect labels at the current detection attention level, and also includes: based on the mapping value of the error quantization feature determined at the current detection attention level, determining the dirt state description vector of the potential defect area corresponding to the dirt state label at the current detection attention level. Based on the third application scenario, the defect label includes an inclusion state label, and the defect detection discriminant vector of the potential defect area under the inclusion state label includes a dynamic description vector and a steady-state description vector; the error quantization feature determined at the current detection attention level is used to determine the defect detection discriminant vectors of the potential defect area corresponding to the set multiple defect labels at the current detection attention level, and further includes: determining the defect trend feature of each of the key detection information blocks determined based on the error quantization feature determined at the current detection attention level; the defect trend feature includes a first trend feature and a second trend feature; determining the first number of the key detection information blocks in the first trend feature, and determining the dynamic description vector of the potential defect area corresponding to the inclusion state label at the current detection attention level based on the first number and the dirt state description vector at the current detection attention level; determining the second number of the key detection information blocks in the second trend feature, and determining the steady-state description vector of the potential defect area corresponding to the inclusion state label at the current detection attention level based on the second number and the dirt state description vector at the current detection attention level.

[0106] In the further in-depth discussion of the technical solution for seamless steel pipe defect detection, this technical idea focuses on the defect state characterization vector based on the potential defect area at each detection attention level, and then determines the defect detection discriminant vector corresponding to multiple defect labels.

[0107] First, for each of the several detection attention levels, determining the error quantification characteristics between the key detection attribute vector of each key detection information block at the current detection attention level and the defect state representation vector at this level is the starting point of the entire process. In this process, it is necessary to conduct in-depth analysis of the key detection attribute vector and the defect state representation vector.

[0108] Taking ultrasonic testing as an example, at the level of structural integrity, the key detection attribute vector may include relevant values ​​such as the amplitude of the reflected wave, the arrival time of the reflected wave, and the frequency of the reflected wave. For example, in a specific detection case, the value of the reflected wave amplitude is 1.5 volts (relative to the normal reflected wave amplitude of 1 volt), the reflected wave arrival time is 0.3 milliseconds (normal range is 0.1-0.5 milliseconds), and the reflected wave frequency is 5MHz (normal frequency range is 4-6MHz). The defect state characterization vector may be a quantitative representation after integrating the information of multiple key detection information blocks. For example, it may exist in a coded form, in which each digit represents a different characteristic state, such as the first bit indicates whether the reflected wave amplitude exceeds the normal range (0 means normal, 1 means exceeded), the second bit indicates whether the arrival time is at the edge of the normal interval (0 means inside, 1 means at the edge), etc.

[0109] The error quantization feature is determined by comparing the corresponding elements in these two vectors. For the reflected wave amplitude, its error quantization feature element may be calculated by (1.5-1) / 1=0.5 (relative error calculation method is used here, and other methods such as absolute value difference can also be used). For the reflected wave arrival time, for example, the center value of the normal interval is 0.3 milliseconds, calculate (0.3-0.3) / 0.3=0 (indicating that the arrival time is exactly in the center of the normal interval). The collection of error quantization feature values ​​of these different elements constitutes the error quantization feature of the key detection information block related to ultrasonic detection at this detection attention level.

[0110] Similarly, at the eddy current distribution uniformity detection attention level of eddy current testing, the eddy current density value in the key detection attribute vector may be 120 units (normal is 100 units), the eddy current density change gradient is 5 units per centimeter (normal range is 3-7 units per centimeter), etc. The defect state characterization vector may represent the overall state of the eddy current distribution in different quantitative ways, such as using a comprehensive value to represent the degree of deviation of the eddy current density from the normal and the overall trend of the eddy current density change. When calculating the error quantification feature, for the eddy current density, (120-100) / 100=0.2; for the eddy current density change gradient, for example, the normal center value is 5 units per centimeter, (5-5) / 5=0. These values, together with the error quantification feature values ​​of other detection methods such as ultrasonic testing, constitute the complete error quantification feature at this detection attention level.

[0111] In the first application scenario, when the defect label is a defect trend change label and the corresponding defect detection discriminant vector is a defect trend change vector, the discrete measurement value at the current detection attention level is determined based on the error quantization characteristics determined at the current detection attention level and the number of key detection information blocks determined.

[0112] Continuing with the above-mentioned ultrasonic testing and eddy current testing as examples, at this level of detection attention, ultrasonic testing and eddy current testing are the only two key detection information blocks (the number is 2). It has been calculated that the quantized eigenvalue of the ultrasonic detection reflected wave amplitude error is 0.5, the arrival time is 0, and the quantized eigenvalue of the eddy current density error of the eddy current detection is 0.2, and the density change gradient is 0. When calculating the discrete metric value, one possible calculation method is to add the absolute values ​​of these error quantized eigenvalues ​​and divide them by the number of key detection information blocks, that is, (|0.5|+|0|+|0.2|+|0|) / 2=0.35.

[0113] Then, based on this discrete metric value, the defect trend change vector corresponding to the defect trend change label of the potential defect area at the current detection attention level is determined. If the discrete metric value is set to be less than 0.1, it means that the defect trend is very stable, 0.1-0.3 means relatively stable, 0.3-0.5 means a medium change trend, and above 0.5 means a rapid change trend. In this example, since the discrete metric value is 0.35, it may be indicated in the defect trend change vector that the defect trend of the potential defect area is in a medium change trend, which is encoded with a specific numerical value (such as 0.6 means a medium change trend, and different numerical values ​​between 0-1 represent different trend states).

[0114] In the second application scenario, when the defect label is a crack state label and the corresponding defect detection discriminant vector is a crack state embedding vector, based on the error quantization features determined at the current detection attention level, the target key detection information block indicated by the error quantization features that meet the set error requirements is selected from the determined key detection information blocks.

[0115] For example, the error requirement is set to that the element value in the error quantization feature is less than 0.1 (the absolute value of the relative error is used here). In the error quantization feature values ​​calculated previously, the error quantization feature value of the arrival time of the reflected wave of the ultrasonic detection is 0, and the error quantization feature value of the gradient of the eddy current density change of the eddy current detection is 0. These two key detection information blocks meet the requirements and are selected as the target key detection information blocks.

[0116] Next, the crack area noise characteristics at the current detection attention level are determined based on the number of target key detection information blocks. In this example, the number of target key detection information blocks is 2, and the total number of key detection information blocks is 2 (ultrasonic and eddy current detection only). The crack area noise characteristics may be expressed as the ratio of these two values ​​(2 / 2=1) or a value after some transformation. For example, if an exemplary transformation method is used, this ratio is multiplied by 0.5 to obtain 0.5, which is used as the crack area noise characteristics.

[0117] Finally, based on the crack area noise feature, the crack state embedding vector corresponding to the crack state label of the potential defect area at the current detection attention level is determined. For example, if the crack area noise feature is set to be less than 0.3, it means that the crack development is in a low noise (stable) state, 0.3-0.7 means a medium noise state, and above 0.7 means a high noise (unstable) state. In this example, since the crack area noise feature is 0.5, the crack state embedding vector may correspond to a code that the crack development is in a medium noise state (such as 0.5 represents a medium noise state, and different values ​​between 0-1 represent different noise levels).

[0118] In the third application scenario, when the defect label is a dirt state label and the corresponding defect detection discriminant vector is a dirt state description vector, the dirt state description vector is determined based on the mapping value of the error quantization feature determined at the current detection attention level.

[0119] For example, the mapping of error quantization features is to convert the element values ​​in the error quantization features into element values ​​in the fouling state description vector through a nonlinear function relationship. For example, the error quantization feature value of the reflected wave amplitude in ultrasonic testing is 0.5, and the corresponding element value in the fouling state description vector is 0.35 through the mapping function (such as the quadratic function y=0.5x²+0.1). For the error quantization feature value of the eddy current density in eddy current testing is 0.2, and the corresponding element value is 0.18 through the same mapping function. These element values ​​together constitute the fouling state description vector to describe the fouling state of the potential defect area.

[0120] Based on this third application scenario, for inclusion state labels, the corresponding defect detection discriminant vectors include dynamic description vectors and steady-state description vectors. Based on the error transfer vector of the error quantization feature determined at the current detection attention level, the defect trend characteristics of each key detection information block are determined.

[0121] The error transfer vector may be a new vector obtained by a series of complex operations on the error quantization characteristics. For example, the error quantization characteristic values ​​of ultrasonic testing and eddy current testing are arranged in a certain order, and then weighted summation, normalization and other operations are performed to obtain the error transfer vector. For example, this error transfer vector is analyzed to find that the key detection information block has two trend characteristics, namely the first trend characteristic and the second trend characteristic.

[0122] Determine the first number of key detection information blocks in the first trend feature, for example, 1 (for example, the ultrasonic detection key detection information block is in the first trend feature), and determine the dynamic description vector of the potential defect area corresponding to the inclusion state label at the current detection attention level based on this number and the dirt state description vector at the current detection attention level (for example, the value of an element in the dirt state description vector is 0.35) through a complex calculation method. For example, the calculation method may be to multiply the number by the element value in the dirt state description vector, and then normalize it, and then add an offset (such as 0.1) based on experience. For example, the calculation result is 0.45, and this value becomes an element value in the dynamic description vector.

[0123] Similarly, determine the second number of key detection information blocks in the second trend characteristic, such as 1 (for example, the eddy current detection key detection information block is in the second trend characteristic), and determine the steady-state description vector corresponding to the inclusion state label of the potential defect area at the current detection attention level based on this number and the dirt state description vector. For example, the calculation method may be to multiply the number by the element value in the dirt state description vector, then square it, and then subtract an adjustment value (such as 0.05) based on experience. For example, the calculation result is 0.12, and this value becomes an element value in the steady-state description vector.

[0124] Through such a series of processing methods under different defect labels based on error quantification characteristics, a more accurate and detailed description of the potential defect areas of seamless steel pipes is made from multiple angles and in multiple quantitative forms. This comprehensive and in-depth analysis method provides a rich basis for the final accurate judgment of the defect type and state, thereby significantly improving the accuracy and reliability of the entire seamless steel pipe defect detection. The processing methods under different defect labels are interrelated and have their own emphasis, and together they construct a complete seamless steel pipe defect detection and evaluation system. This system can make full use of the information in various test data, avoid the risk of misjudgment caused by a single data source or simple judgment method, and make the detection of seamless steel pipe defects more scientific, rigorous and comprehensive.

[0125] Based on the above technical ideas and application scenarios, by determining the defect detection discriminant vector under different defect labels based on the error quantization characteristics, the relationship between the key detection information block and the defect state characterization vector can be deeply explored. Various numerical examples (such as error quantization feature element values, the number of key detection information blocks, detection values ​​under different detection methods, etc.) show that this method can accurately describe the state of potential defect areas from multiple dimensions. This helps to more accurately judge the defect conditions of seamless steel pipes, including defect trends, crack conditions, dirt conditions, and inclusion conditions, etc., and improves the comprehensiveness and accuracy of defect detection. This multi-dimensional and refined detection method helps to detect defects in a timely manner during the production and use of seamless steel pipes, ensure the quality and safety of seamless steel pipes, and reduce various risks and losses that may be caused by defects not being discovered in a timely manner.

[0126] In an independent embodiment, the defect label includes a coating uniformity label, and the defect detection discrimination vector of the potential defect area under the coating uniformity label includes a coating uniformity quantization vector; the defect detection discrimination vectors corresponding to the potential defect area under the set multiple defect labels are determined based on the defect state characterization vector of the potential defect area at each detection attention level in the several detection attention levels, including: obtaining pyramid coating uniformity quantization vectors corresponding to each of the set multiple coating areas; for each coating area in the multiple coating areas, determining the spatial correlation between the potential defect area and the targeted coating area according to the pyramid coating uniformity quantization vector of the targeted coating area and the defect state characterization vector of the potential defect area at each detection attention level in the several detection attention levels; selecting a target coating area from the multiple coating areas according to the spatial correlation between the potential defect area and each coating area in the multiple coating areas; and determining the coating uniformity quantization vector of the potential defect area under the coating uniformity label based on the target coating area.

[0127] Within the scope of the seamless steel pipe defect detection technical solution, this independent embodiment focuses on determining the defect detection discrimination vector of the potential defect area under the coating uniformity label, that is, the coating uniformity quantization vector.

[0128] First, obtaining the pyramid coating uniformity quantization vectors corresponding to each of the set multiple coating areas is an important starting operation. These pyramid coating uniformity quantization vectors are a quantitative representation of the coating uniformity of each coating area. For example, for a certain coating area, the pyramid coating uniformity quantization vector may contain multiple levels of information, describing the uniformity of the coating from macro to micro. In a specific numerical example, this vector may represent the uniformity of the coating thickness at a certain level with a numerical value, such as 0.8 (0 means completely uneven, 1 means completely uniform), and at another level, it may represent the uniformity of the coating component distribution as 0.7, etc.

[0129] Next, for each coating area in the multiple coating areas, the spatial correlation between the potential defect area and the targeted coating area is determined based on the pyramid coating uniformity quantization vector of the targeted coating area and the defect state characterization vector of the potential defect area at several detection attention levels. In this process, the defect state characterization vector contains state information about the potential defect area at different detection attention levels. Taking image detection as an example, at the surface flatness detection attention level, the defect state characterization vector may contain information such as surface roughness value; at the color consistency detection attention level, it may contain color deviation value, etc. For a certain coating area, for example, the coating thickness uniformity in its pyramid coating uniformity quantization vector is 0.8, and the coating component distribution uniformity is 0.7, while the surface roughness value of the potential defect area at the surface flatness detection attention level is 0.3 (indicating relatively rough), and the color deviation value at the color consistency detection attention level is 0.4 (indicating a certain color deviation). These values ​​are combined using a specific calculation method that is based on a deep understanding of the relationship between coatings and defects and a large amount of experimental data to produce a spatial correlation value between the potential defect area and the coating area, such as 0.45.

[0130] In the same way, such calculation is performed for each of the multiple coating regions, thereby obtaining the spatial correlation value between the potential defect region and each coating region. Then, according to the spatial correlation between the potential defect region and each coating region in the multiple coating regions, the target coating region is selected from the multiple coating regions. For example, if the coating region with a spatial correlation value greater than 0.5 is set as the target coating region, among the multiple calculated spatial correlation values, such as 0.45, 0.55, 0.6, etc., the coating regions with spatial correlation values ​​of 0.55 and 0.6 will be selected as the target coating regions.

[0131] Finally, based on the target coating area, the coating uniformity quantification vector of the potential defect area under the coating uniformity label is determined. This determination process is based on the characteristics of the target coating area and their relationship with the potential defect area. Continuing with the above numerical example, if the coating thickness uniformity in the pyramid coating uniformity quantification vector of the target coating area is 0.8 and 0.9, and the coating component distribution uniformity is 0.7 and 0.8, by further analyzing and calculating these values ​​(such as taking the average or calculating according to a certain weight), the coating uniformity quantification vector of the potential defect area under the coating uniformity label is obtained. For example, after calculation, the value of the coating thickness uniformity in the coating uniformity quantification vector is 0.85 (obtained by (0.8+0.9) / 2), and the coating component distribution uniformity value is 0.75 (obtained by (0.7+0.8) / 2).

[0132] Different levels of detection attention play an important role in determining spatial correlation. For example, at the level of ultrasonic detection of the internal structure of the coating, the propagation characteristic value of the reflected wave in the coating in the defect state characterization vector (such as the reflected wave attenuation value of 0.2, indicating a certain attenuation) is combined with the internal structure uniformity value of the coating in the pyramid coating uniformity quantization vector of the coating area (such as 0.6, indicating that the structural uniformity is general), and the spatial correlation is calculated according to specific calculation rules. For example, the calculation result is 0.35, which is also involved in the selection of the target coating area and the determination of the final coating uniformity quantization vector.

[0133] At the coating conductivity detection attention level of eddy current testing, the relevant values ​​of the eddy current in the defect state characterization vector generated in the coating (such as the eddy current density change value of 0.3, indicating a certain eddy current density change) are calculated together with the coating conductivity uniformity value in the pyramid coating uniformity quantification vector of the coating area (such as 0.7, indicating good conductivity uniformity). For example, the calculated spatial correlation value is 0.4, which also affects the selection of the target coating area and the determination of the coating uniformity quantification vector.

[0134] At the coating density detection attention level of X-ray detection, the relevant value of the ray after penetrating the coating in the defect state characterization vector (such as the ray attenuation coefficient value of 0.4, indicating that the ray has a certain attenuation) and the coating density uniformity value in the pyramid coating uniformity quantification vector of the coating area (such as 0.8, indicating that the density uniformity is good) are calculated for spatial correlation. For example, the obtained spatial correlation value is 0.3, which also plays a corresponding role in the whole process.

[0135] By comprehensively and integratedly considering the characteristics of each detection attention level and the coating area, the coating uniformity quantification vector of the potential defect area under the coating uniformity label can be accurately determined, thereby providing an accurate quantitative basis for coating uniformity defect detection of seamless steel pipes.

[0136] It can be seen that by obtaining the pyramid coating uniformity quantification vector of the coating area, the spatial correlation is determined by combining the defect state characterization vector of the potential defect area at different detection attention levels, and then the target coating area is selected to determine the coating uniformity quantification vector. The uniformity of coating thickness, uniformity of coating component distribution, and various numerical examples at the detection attention level (such as surface roughness, color deviation, reflected wave attenuation, eddy current density change, ray attenuation coefficient, etc.) show that this method can comprehensively consider the relationship between the coating and the potential defect area, accurately quantify the coating uniformity, and improve the accuracy of seamless steel pipe coating uniformity defect detection.

[0137] Further, Figure 2 Schematic diagram of a seamless steel pipe defect detection device 200 provided in an embodiment of the present application. Figure 2 The seamless steel pipe defect detection device 200 shown includes a processor 210, and the processor 210 can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0138] Alternatively, if Figure 2 As shown, the seamless steel pipe defect detection device 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.

[0139] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0140] Alternatively, if Figure 2 As shown, the seamless steel pipe defect detection device 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, may send information or data to other devices, or receive information or data sent by other devices.

[0141] Optionally, the seamless steel pipe defect detection device 200 can implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0142] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities.

[0143] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0144] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0145] It should be noted that, in this article, the term "comprises", "includes" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the method and device in the embodiment of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0146] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0147] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the present application, all of which are within the protection of the present application.

Claims

1. A seamless steel pipe defect detection method based on machine vision, characterized in that: The method is applied to a seamless steel pipe defect detection device, and the method comprises: Locating a potential defect area of ​​the target seamless steel pipe according to a multi-mode detection data set of the target seamless steel pipe, and segmenting the multi-mode detection data set of the target seamless steel pipe to obtain multi-mode detection characterization information corresponding to the potential defect area; wherein the multi-mode detection characterization information includes: image detection information, ultrasonic detection information, eddy current detection information and radiographic detection information; Determine key detection information blocks from each multi-mode detection representation information obtained by segmentation, and mine key detection attribute vectors of each of the determined key detection information blocks at each of the set detection attention levels; For each of the several detection attention levels, according to the key detection attribute vectors of each of the key detection information blocks at the targeted detection attention level, determine the defect state representation vector of the potential defect area at the targeted detection attention level; Determine, based on the defect state representation vector of the potential defect area at each of the plurality of detection attention levels, the defect detection discriminant vectors corresponding to the potential defect area under the set plurality of defect labels; Outputting a defect detection viewpoint corresponding to the potential defect area based on the defect detection discriminant vectors respectively corresponding to the potential defect area under the multiple defect labels; The method of determining the defect detection discriminant vectors corresponding to the potential defect area under the set multiple defect labels based on the defect state characterization vectors of the potential defect area at each of the several detection attention levels includes: for each of the several detection attention levels, mining the error quantization features between the key detection attribute vectors of each of the key detection information blocks determined at the current detection attention level and the defect state characterization vectors at the current detection attention level; determining the defect detection discriminant vectors corresponding to the set multiple defect labels of the potential defect area at the current detection attention level based on the error quantization features determined at the current detection attention level; The defect label includes a defect trend change label, and the defect detection discriminant vector of the potential defect area under the defect trend change label includes a defect trend change vector; the error quantization characteristics determined at the current detection attention level are used to determine the defect detection discriminant vectors of the potential defect area corresponding to the set multiple defect labels at the current detection attention level, including: determining the discrete measurement value at the current detection attention level based on the error quantization characteristics determined at the current detection attention level and the determined number of key detection information blocks; determining the defect trend change vector of the potential defect area corresponding to the defect trend change label at the current detection attention level based on the discrete measurement value at the current detection attention level.

2. The method according to claim 1, characterized in that The multi-mode detection characterization information corresponding to the potential defect area is obtained by segmenting the multi-mode detection data set of the target seamless steel pipe, including: determining a region location mask and a data mask associated with the potential defect region; From the multimode inspection data set of the target seamless steel pipe, the surface inspection results belonging to the regional positioning mask and pointed to by the data mask are extracted as the multimode inspection characterization information associated with the potential defect area.

3. The method according to claim 1, characterized in that The steps of using each multimode detection characterization information obtained by segmentation as target multimode detection characterization information and determining a key detection information block from the target multimode detection characterization information include: Performing modal decomposition processing on the target multi-mode detection representation information to obtain at least one single-mode detection representation information included in the target multi-mode detection representation information; For each single-mode detection characterization information in the at least one single-mode detection characterization information, determine a set key detection information block paired with the single-mode detection characterization information in the key detection information pool, and use the set key detection information block paired with the single-mode detection characterization information as the key detection information block determined from the target multi-mode detection characterization information; The determining whether there is a set key detection information block matched with the single-mode detection characterization information in the key detection information pool includes: Determine a detection feature difference between the single-mode detection characterization information and a set key detection information block in a key detection information pool; When there is a target setting key detection information block in the key detection information pool whose detection feature difference satisfies the first feature difference index, the target setting key detection information block is used as a setting key detection information block paired with the targeted single-mode detection characterization information.

4. The method according to claim 1, characterized in that The key detection attribute vectors of each of the key detection information blocks determined by the mining at each of the set multiple detection attention levels include: For each of the determined key detection information blocks, obtaining a preset key detection attribute vector of a set key detection information block paired with the key detection information block at each of the set multiple detection attention levels; Based on the obtained preset key detection attribute vector, the key detection attribute vector of the targeted key detection information block at each detection attention level among the several detection attention levels is determined.

5. The method according to claim 1, characterized in that The step of determining, for each of the plurality of detection attention levels, a defect state representation vector of the potential defect area at the detection attention level according to the key detection attribute vector of each of the key detection information blocks at the detection attention level, includes: For each of the several detection attention levels, according to the key detection attribute vectors of each of the key detection information blocks at the targeted detection attention level, determine the fully connected state attribute vector of the potential defect area at the targeted detection attention level; According to the fully connected state attribute vector of the potential defect area at the targeted detection attention level, a defect state representation vector of the potential defect area at the targeted detection attention level is determined.

6. The method according to claim 1, characterized in that The defect label includes a defect category distribution label, and the defect detection discriminant vector of the potential defect area under the defect category distribution label includes a defect category distribution thermal vector; the defect state characterization vector of the potential defect area at each of the several detection attention levels is determined based on the defect state characterization vector of the potential defect area, and the defect detection discriminant vectors corresponding to the potential defect area under the set multiple defect labels respectively include: For each of the several detection attention levels, based on the defect state characterization vector of the potential defect area at each of the several detection attention levels and the set credibility coefficient condition, the defect category distribution thermal vector of the potential defect area under the defect category distribution label is determined.

7. The method according to claim 1, characterized in that The defect label includes a crack state label, and the defect detection discriminant vector of the potential defect area under the crack state label includes a crack state embedding vector; the error quantization feature determined at the current detection attention level determines the defect detection discriminant vectors of the potential defect area corresponding to the set multiple defect labels at the current detection attention level, and also includes: based on the error quantization feature determined at the current detection attention level, from each of the determined key detection information blocks, select the target key detection information block indicated by the error quantization feature that meets the set error requirement; based on the number of the target key detection information blocks, determine the crack area noise feature at the current detection attention level; based on the crack area noise feature at the current detection attention level, determine the crack state embedding vector of the potential defect area corresponding to the crack state label at the current detection attention level; The defect label includes a dirt state label, and the defect detection discrimination vector of the potential defect area under the dirt state label includes a dirt state description vector; the error quantization feature determined at the current detection attention level determines the defect detection discrimination vectors of the potential defect area corresponding to the set multiple defect labels at the current detection attention level, and also includes: based on the mapping value of the error quantization feature determined at the current detection attention level, determining the dirt state description vector of the potential defect area corresponding to the dirt state label at the current detection attention level; the defect label includes an inclusion state label, and the defect detection discrimination vector of the potential defect area under the inclusion state label includes a dynamic description vector and a steady-state description vector; the error quantization feature determined at the current detection attention level determines the defect detection discrimination vector of the potential defect area at the current detection attention level The defect detection discrimination vectors respectively corresponding to the set multiple defect labels also include: based on the error transfer vector of the error quantization feature determined at the current detection attention level, mining the defect trend features of each of the key detection information blocks determined; the defect trend features include a first trend feature and a second trend feature; determining the first number of the key detection information blocks in the first trend feature, and determining the dynamic description vector of the potential defect area corresponding to the inclusion state label at the current detection attention level based on the first number and the dirt state description vector at the current detection attention level; determining the second number of the key detection information blocks in the second trend feature, and determining the steady-state description vector of the potential defect area corresponding to the inclusion state label at the current detection attention level based on the second number and the dirt state description vector at the current detection attention level.

8. A seamless steel pipe defect detection device, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 7.

9. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images