Welded pipeline precipitate detection system and method based on image recognition

Abstract

The invention provides a welded pipeline precipitate detection system and method based on image recognition. The system comprises an image preprocessing assembly, a feature extraction and fusion assembly and a decision and feedback optimization assembly. The method comprises the following steps: preprocessing an original welding pipeline image, dynamically adjusting denoising parameters, identifying a welding seam edge and a precipitate area by analyzing local gradient information, and performing contrast enhancement; extracting multi-level feature information from the welding pipeline image after comparison enhancement; meanwhile, capturing features of different scales, and automatically focusing on a key area; making a decision based on a detection result of the key area; and dynamically adjusting a detection strategy according to a detection result, and meanwhile, generating a detailed detection report including the type, distribution, size and potential risk level of precipitates. According to the invention, accurate and efficient detection of the precipitates of the welded pipeline is ensured; the safety detection level of the welded pipeline can be remarkably improved, potential safety risks are prevented, and smooth industrial production is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of object analysis and detection, and particularly relates to a welding pipeline precipitate detection system and method based on image recognition. Background Art

[0002] The precipitates in the welded pipeline mainly include residues of zinc and welding agent. When welding galvanized pipes, due to the low melting point of zinc, a large amount of zinc gas and steam will be generated during the welding process. At the same time, the welding agent will also evaporate and decompose under the action of heat, forming some residues. These residues of zinc and welding agent are the slag produced by welding galvanized pipes. The main hazards of the precipitates in the welded pipeline include damage to the respiratory system, nervous system, skin and eyes, as well as potential carcinogenic risks. At present, the detection means of the precipitates include non-destructive testing, chemical analysis, physical analysis, etc. These measures require the assistance of certain equipment, not only with a long detection period, but also poor real-time performance.

[0003] Prior Art One, Application No.: CN202410187081.4 discloses a steel plate with excellent toughness in the heat affected zone of large heat input welding and its manufacturing method. The mass percentage of the steel plate components is: C 0.05 - 0.09%, Si 0.10 - 0.30%, Mn 1.2 - 1.6%, P ≤ 0.015%, S 0.001 - 0.01%, Ni 0.2 - 0.4%, Ti 0.005 - 0.03%, Nb 0.001 - 0.015%, Mg 0.0005 - 0.004%, N 0.001 - 0.006%, Al 0.004 - 0.036%, B 0.0005 - 0.005%, O 0.0005 - 0.004%, Ca ≤ 0.005%, REM ≤ 0.02%, and the balance includes Fe and unavoidable impurities; and it satisfies: 1 < effective [Ti] / [Nb] < 8, the free Nb content in the steel plate < -0.03%, and the average particle size of all precipitates is less than 25 nm. Although it can inhibit the growth of austenite grains at high temperatures and promote the formation of fine and tough polygonal ferrite structures in the heat affected zone of welding at room temperature, improving the low-temperature toughness of the heat affected zone of welding; however, it does not detect the type and potential hazards of the precipitates, which increases the danger of the welded steel plate to a certain extent.

[0004] Prior art 2, application number: CN202410187060.2 discloses a calcium-treated thick steel plate for improving high-wire energy welding and a manufacturing method thereof, wherein the steel plate composition weight percentage is: C 0.045-0.075%, Si 0.10-0.30%, Mn 1.3-1.6%, P≤0.015%, S 0.001-0.01%, Ti 0.005-0.035%, Nb 0.002-0.02%, Ca0.0005-0.005%, N 0.001-0.01%, O 0.001-0.004%, Al 0.001~0.06%, REM≤0.02%, Zr≤0.02%, the balance includes Fe and other unavoidable impurities; and 2≤Ti / N≤6; effective oxygen content combined with Ti OTi,eff; effective Ti content combined with N TiN,eff: TiN,eff=Ti-2OTi,eff, should meet 2≤TiN,eff / N≤4.5; effective Ti content TiN,eff and Nb content must also meet TiN,eff / Nb≥1.5. The volume density of precipitates less than 500nm in the steel plate is greater than 9.0×107 / mm 3 , the proportion of precipitates smaller than 100nm is greater than 75%. Although the good high-energy line welding performance of the welding heat affected zone is improved, it is limited to the investigation and proportion adjustment of the precipitates, and there is no detection of the precipitates, lacking safety reminders during the welding process.

[0005] Prior art three, application number: CN 202410217114.5 discloses a high yield rate direct rolling controlled cooling method, in which the ratio of niobium, vanadium and titanium elements is optimized and graphite is added during the direct rolling controlled cooling process. Although the strength, toughness, corrosion resistance and welding performance of the steel are enhanced; and adjusting the temperature, duration, speed and tension of the rolling mill to refine the grains can improve the distribution of precipitates, increase the dislocation density and promote phase transformation, thereby enhancing the mechanical properties of the steel; however, the degree of intelligence of its technical means is not explained, and the active detection and identification of precipitates is not realized, so the accuracy of precipitate detection has a lot of room for improvement.

[0006] At present, the existing technologies 1, 2 and 3 lack detection technology means for welding management precipitates and cannot effectively ensure the safety during welding. Therefore, the present invention provides a welding pipeline precipitate detection system and method based on image recognition. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a welding pipeline precipitate detection system based on image recognition, comprising the following steps:

[0008] The image preprocessing component is responsible for preprocessing the original welded pipeline image, dynamically adjusting the denoising parameters, identifying the weld edges and precipitate regions by analyzing the local gradient information, and enhancing the contrast;

[0009] The feature extraction and fusion component is responsible for extracting multi-level feature information from the contrast-enhanced welded pipeline image, including local details and global structures; at the same time, capturing features at different scales and automatically focusing on key areas;

[0010] The decision-making and feedback optimization component is responsible for making decisions based on the detection results of the key areas; dynamically adjusting the detection strategy according to the detection results, and generating a detailed detection report at the same time.

[0011] Optionally, the image preprocessing component includes:

[0012] The adaptive denoising module is responsible for intelligently adjusting the denoising parameters according to the local features of the original welded pipeline image; by real-time monitoring the local gradient changes of the original welded pipeline image, identifying which regions contain weld edge and precipitate information;

[0013] The gradient analysis module is responsible for adopting multi-scale gradient analysis during the identification process, capturing the details of the original welded pipeline image at different resolution levels, and identifying the weld edges and precipitate regions;

[0014] The parameter feedback optimization module is responsible for setting up a dynamic parameter feedback optimization mechanism, collecting effect feedback in real time, and dynamically adjusting the denoising and contrast enhancement parameters according to the feedback results.

[0015] Optionally, the feature extraction and fusion component includes:

[0016] The region screening module is responsible for initially analyzing the preprocessed welded pipeline image, identifying regions with significant gradient changes, which are marked as potential key areas;

[0017] The feature integration module is responsible for layer-by-layer extracting feature information from the welded pipeline image marked with potential key areas;

[0018] The automatic evaluation module is responsible for dynamically adjusting the priority of the analysis regions, automatically evaluating the potential importance of each key area based on the results of gradient perception and multi-scale feature extraction; giving priority to locking key areas containing significant precipitates or abnormal structures.

[0019] Optionally, the region screening module sets the priority of the key areas according to the activity of the gradient change.

[0020] Optionally, the region screening module includes:

[0021] The threshold preset sub-module is responsible for setting the dynamic saliency threshold, which is used to identify the gradient changes in the welded pipeline image. At the same time, the dynamic saliency threshold is adaptively adjusted according to the overall gradient distribution and local activity of the welded pipeline image.

[0022] The change capture sub-module is responsible for globally scanning the pre-processed welded pipeline image through gradient perception, capturing the gradient changes of each pixel point in the image, and comparing with the dynamic saliency threshold to confirm whether it belongs to the region with significant gradient changes.

[0023] The region division sub-module is responsible for evaluating the gradient activity of each region. The gradient activity is based on the change frequency and local consistency of the gradient amplitude and gradient direction, distinguishing the active region and the stable region. The active region is marked as the potential key region and divided into different priorities according to its activity level.

[0024] Optionally, the change capture sub-module includes:

[0025] The vector calculation unit is responsible for loading the pre-processed welded pipeline image and allocating an independent calculation unit for each pixel point.

[0026] The numerical generation unit is responsible for constructing a local neighborhood window centered on the current pixel point, analyzing the distribution characteristics of the gradient vectors in the neighborhood, quantifying the gradient change intensity of the current pixel point based on the local gradient distribution, and generating the gradient change intensity value.

[0027] The characterization formation unit is responsible for capturing the gradient changes of each pixel point in the welded pipeline image row by row and column by column starting from the upper left corner.

[0028] The region determination unit is responsible for comparing the gradient change intensity value of each pixel point with the dynamic saliency threshold to determine whether it belongs to the region with significant gradient changes.

[0029] Optionally, the characterization formation unit maps the gradient change intensity value of each pixel point to the global gradient change intensity map to form the global distribution characterization of the gradient changes.

[0030] Optionally, the region determination unit marks the pixel points determined to have significant gradient changes as the significant region.

[0031] Optionally, the feature integration module includes:

[0032] The first-scale output sub-module is responsible for locally refining the analysis of the potential key region, capturing the geometric contour of the tiny precipitates, identifying the boundary points of the precipitates by analyzing the spatial distribution of the gradient amplitude changes, and reconstructing their shape features through the geometric fitting algorithm.

[0033] The second-scale output sub-module is responsible for globally scanning the welded pipeline image to identify the overall morphology of the weld seam; combined with the evaluation of the macroscopic gradient activity, it identifies whether the precipitates are concentrated in specific regions of the weld seam or exhibit an abnormal spatial distribution pattern.

[0034] The correlation relationship establishment sub-module is responsible for establishing the correlation relationship between microscopic features and macroscopic morphology through multi-scale consistency analysis of gradient magnitude and direction.

[0035] A method for detecting precipitates in a welded pipeline based on image recognition provided by the present invention includes the following steps:

[0036] Preprocess the original welded pipeline image, dynamically adjust the denoising parameters, identify the weld seam edge and precipitate area by analyzing local gradient information, and enhance the contrast.

[0037] Extract multi-level feature information from the contrast-enhanced welded pipeline image, including local details and global structure; at the same time, capture features at different scales and automatically focus on key areas.

[0038] Make a decision based on the detection results of the key areas; dynamically adjust the detection strategy according to the detection results, and generate a detailed detection report at the same time.

[0039] The image preprocessing component of the present invention preprocesses the original welded pipeline image by dynamically adjusting the denoising parameters and analyzing local gradient information, effectively identifies the weld seam edge and precipitate area, and improves the clarity of the image through contrast enhancement. The feature extraction and fusion component captures features at different scales and automatically focuses on key areas. The decision-making and feedback optimization component generates a detailed detection report including the type, distribution, size, and potential risk level of the precipitates.

[0040] Other features and advantages of the present invention will be described in the subsequent description, and, in part, will be obvious from the description or learned through the implementation of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written description and the drawings.

[0041] The technical solutions of the present invention will be further described in detail below with reference to the drawings and embodiments. Description of the Drawings

[0042] The drawings are used to provide a further understanding of the present invention and constitute a part of the description. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0043] Figure 1 It is a block diagram of a detection system for precipitates in a welded pipeline based on image recognition in Embodiment 1 of the present invention;

[0044] Figure 2 It is the block diagram of the image preprocessing component in Embodiment 2 of the present invention;

[0045] Figure 3 It is the block diagram of the feature extraction and fusion component in Embodiment 3 of the present invention;

[0046] Figure 4 It is the block diagram of the region screening module in Embodiment 4 of the present invention;

[0047] Figure 5 It is the block diagram of the change capture sub-module in Embodiment 5 of the present invention;

[0048] Figure 6 It is the block diagram of the feature integration module in Embodiment 6 of the present invention;

[0049] Figure 7 It is the block diagram of the first-scale output sub-module in Embodiment 7 of the present invention;

[0050] Figure 8 It is the block diagram of the second-scale output sub-module in Embodiment 8 of the present invention;

[0051] Figure 9 It is the block diagram of the decision-making and feedback optimization component in Embodiment 9 of the present invention;

[0052] Figure 10 It is the flowchart of the welding pipeline precipitate detection method based on image recognition in Embodiment 10 of the present invention. Specific embodiments

[0053] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not used to limit the present invention.

[0054] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms "a", "the" and "said" used in the embodiments of the present application are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0055] When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0056] Embodiment 1: As Figure 1 shown, an image recognition-based welding pipeline precipitate detection system is provided in an embodiment of the present invention, including:

[0057] An image preprocessing component, responsible for preprocessing the original welding pipeline image, dynamically adjusting the denoising parameters, identifying the weld edge and precipitate area by analyzing local gradient information, and enhancing the contrast;

[0058] A feature extraction and fusion component, responsible for extracting multi-level feature information from the contrast-enhanced welding pipeline image, including local details (such as tiny precipitates) and global structures (such as the overall shape of the weld); at the same time, capturing features at different scales and automatically focusing on key areas;

[0059] A decision-making and feedback optimization component, responsible for making decisions based on the detection results of the key areas; dynamically adjusting the detection strategy according to the detection results, and at the same time generating a detailed detection report, including the type, distribution, size and potential risk level of the precipitates.

[0060] The working principle and beneficial effects of the above technical solution are as follows: The image preprocessing component in this embodiment preprocesses the original welded pipeline image, dynamically adjusts the denoising parameters, identifies the weld edges and precipitate regions by analyzing the local gradient information, and enhances the contrast; the feature extraction and fusion component extracts multi-level feature information from the contrast-enhanced welded pipeline image, including local details (such as tiny precipitates) and global structures (such as the overall shape of the weld); at the same time, it captures features at different scales and automatically focuses on the key areas; the decision-making and feedback optimization component makes decisions based on the detection results of the key areas; dynamically adjusts the detection strategy according to the detection results, and at the same time generates a detailed detection report, including the type, distribution, size and potential risk level of the precipitates. The image preprocessing component of the above solution preprocesses the original welded pipeline image by dynamically adjusting the denoising parameters and analyzing the local gradient information, effectively identifies the weld edges and precipitate regions, and improves the clarity of the image through contrast enhancement. Significance: Preprocessing is a very important step in image recognition, which can improve the accuracy and efficiency of subsequent feature extraction; by identifying the weld and precipitate regions, it lays a solid foundation for feature extraction and detection decision-making. The feature extraction and fusion component captures features at different scales and automatically focuses on the key areas. Significance: Feature extraction is the core link of image analysis; by extracting multi-level features, it can comprehensively understand the state of the welded pipeline, identify tiny precipitates and the overall shape of the weld, which is crucial for accurately detecting weld precipitates. The decision-making and feedback optimization component generates a detailed detection report containing the type, distribution, size and potential risk level of the precipitates. Significance: It not only completes the detection decision of the precipitates in the welded pipeline, but also optimizes the detection process by dynamically adjusting the detection strategy, improving the adaptability and accuracy of the detection system. At the same time, the generated detailed detection report provides an important basis for subsequent repair and maintenance work.

[0061] In summary, the three components in this embodiment together constitute a complete detection system to ensure the accurate and efficient detection of precipitates in welded pipelines; it can significantly improve the safety detection level of welded pipelines, prevent potential safety risks, and ensure the smooth progress of industrial production.

[0062] Embodiment 2: As Figure 2 shown, based on Embodiment 1, the image preprocessing component provided by the embodiment of the present invention includes:

[0063] An adaptive denoising module, which is responsible for intelligently adjusting the denoising parameters according to the local features of the original welded pipeline image; by real-time monitoring the local gradient changes of the original welded pipeline image, it identifies which regions contain weld edge and precipitate information;

[0064] Gradient analysis module, responsible for adopting multi-scale gradient analysis during the recognition process to capture the details of the original welded pipeline image at different resolution levels, identify the weld edge and precipitate area; after identifying the key area, adjust the contrast according to the local features and overall brightness distribution of the original welded pipeline image;

[0065] Parameter feedback optimization module, responsible for setting up a dynamic parameter feedback optimization mechanism, collecting effect feedback in real time, and dynamically adjusting the denoising and contrast enhancement parameters according to the feedback results.

[0066] Among them, the formula of the adaptive denoising module:

[0067]

[0068] In the formula, D′(x, y) represents the gray value of the denoised welded pipeline image at the position (x, y); \(I(x,y)\) represents the gray value of the original welded pipeline image at the position (x, y); μ i (x, y) represents the mean value of the i-th local area; λ i (x, y) represents the adaptive denoising parameter of the i-th local area; \(w_i(x,y)\) represents the weight of the i-th local area; γ i (x, y) represents the non-linear adjustment parameter of the i-th local area, used to enhance the denoising effect; N represents the total number of local areas; e represents the base of the natural logarithm; |·| represents the absolute value operation;

[0069] The formula of the contrast adjustment module:

[0070]

[0071] In the formula, I′(x, y) represents the gray value of the welded pipeline image with adjusted contrast at the position (x, y); I(x, y) represents the gray value of the original welded pipeline image at the position (x, y);

[0072] μ(x, y) represents the mean value of the local area; ξ(x, y) represents the standard deviation of the local area;

[0073] α(x, y) represents the local brightness adjustment parameter; β(x, y) represents the local contrast adjustment parameter;

[0074] σ(x, y) represents the local brightness offset, used to enhance details; ∈(x, y) represents a small constant of the local area, to avoid division by zero error;

[0075] The formula of the parameter feedback optimization module:

[0076]

[0077] In the formula, λi+1 (x, y) represents the updated denoising parameter; λ i (x, y) represents the current denoising parameter; L(λ i (x, y)) represents the loss function defined according to the denoising effect; R(λ i (x, y)) represents the regularization term used to prevent overfitting; represents the gradient of the loss function with respect to the denoising parameter; represents the gradient of the regularization term with respect to the denoising parameter; η i (x, y) represents the dynamic learning rate, adjusted according to local features; κ i (x, y) represents the weight coefficient of the regularization term; The output of the adaptive denoising module is used as the input of the gradient analysis module to extract the gradient information of the welded pipeline image; The output of the gradient analysis module is used to guide the contrast adjustment module to identify key areas and dynamically adjust the contrast; The parameter feedback optimization module dynamically adjusts the denoising parameters according to the feedback of the denoising and contrast adjustment effects to optimize the denoising effect; The parameters of all modules are dynamically updated through the parameter feedback optimization module to form a closed-loop optimization system. The above formula considers local features, multi-scale gradient analysis, dynamic parameter optimization, and regularization constraints, and can meet the requirements of complex welded pipeline image preprocessing.

[0078] The working principle and beneficial effects of the above technical solution are as follows: The adaptive denoising module in this embodiment intelligently adjusts the denoising parameters according to the local features of the original welded pipeline image; by real-time monitoring the local gradient changes of the original welded pipeline image, it identifies which areas contain weld edge and precipitate information; the gradient analysis module (whose main task is to capture the details of the original welded pipeline image through multi-scale gradient analysis during the image preprocessing stage, identify the weld edge and precipitate areas, and analyze the local gradient changes based on the image information at different resolutions, so as to provide a basis for subsequent denoising and contrast adjustment) adopts multi-scale gradient analysis during the identification process to capture the details of the original welded pipeline image at different resolution levels and identify the weld edge and precipitate areas; after identifying the key areas, it adjusts the contrast according to the local features and overall brightness distribution of the original welded pipeline image; the parameter feedback optimization module sets a dynamic parameter feedback optimization mechanism, collects the effect feedback in real time, and dynamically adjusts the denoising and contrast enhancement parameters according to the feedback results. The adaptive denoising module of the above solution can intelligently adjust the denoising parameters according to the local features of the welded pipeline image, thereby more effectively reducing the noise in the image while retaining the key information of the weld edge and precipitate. Significance: It ensures the accuracy and reliability of image analysis, reduces misjudgment caused by noise, and improves the monitoring level of welding quality. The gradient analysis module can capture the details of the weld edge and precipitate at different resolutions through multi-scale gradient analysis, and effectively identify the key areas. Significance: For the quality inspection of welded pipelines, detail identification is crucial; it ensures that even in a complex welding environment, the weld and precipitate can be accurately identified, providing an accurate information basis for processing. The parameter feedback optimization module collects the effect feedback of image processing in real time and dynamically adjusts the denoising and contrast enhancement parameters according to the feedback, which can ensure that the processing result is always in the optimal state. Significance: The dynamic parameter adjustment mechanism makes image processing more flexible and adaptive, and can adjust the parameters according to the characteristics and changes of the actual image, improving the efficiency and effect of processing.

[0079] In summary, this embodiment improves the accuracy and reliability of the welded pipeline image analysis, makes the welding quality monitoring more precise, thereby enhancing the overall quality and efficiency of the welding process; through intelligent preprocessing, the labor intensity and error rate of manual inspection can be greatly reduced, providing strong technical support for intelligent manufacturing and automated welding.

[0080] Embodiment 3: As Figure 3 shown, on the basis of Embodiment 1, the feature extraction and fusion component provided by the embodiment of the present invention includes:

[0081] The region screening module is responsible for the preliminary analysis of the preprocessed welded pipeline image, identifying regions with significant gradient changes, which are marked as potential key regions; setting the priority of the key regions according to the activity of the gradient change;

[0082] The feature integration module is responsible for extracting feature information layer by layer from the welded pipeline image marked with potential key regions, capturing the shape, texture and edge features of tiny precipitates at the microscopic level; at the macroscopic level, identifying the overall morphology of the weld and its spatial relationship with the precipitates; integrating feature information at different scales;

[0083] The automatic evaluation module is responsible for dynamically adjusting the priority of the analysis region, automatically evaluating the potential importance of each key region based on the results of gradient perception and multi-scale feature extraction; preferentially locking the key regions containing significant precipitates or abnormal structures.

[0084] The working principle and beneficial effects of the above technical solution are as follows: The region screening module of this embodiment conducts a preliminary analysis on the preprocessed welding pipeline image to identify regions with significant gradient changes, which are marked as potential key regions; the priority of the key regions is set according to the activity of the gradient change (conduct a preliminary analysis on the preprocessed welding pipeline image to identify regions with significant gradient changes and mark them as potential key regions; set priorities for these regions according to the activity of the gradient change so that subsequent feature extraction and analysis can focus on the most important regions. The output of the gradient analysis module (such as the recognition results of the weld edge and precipitate region) can be used as the input of the region screening module to help lock in the key regions more efficiently; the priority setting of the region screening module can be fed back to the gradient analysis module to dynamically adjust the preprocessing parameters and further improve the image quality); the feature integration module extracts feature information layer by layer from the welding pipeline image marked with potential key regions, capturing the shape, texture, and edge features of tiny precipitates at the micro level; at the macro level, identify the overall shape of the weld and its spatial relationship with the precipitates; integrate feature information at different scales; the automatic evaluation module dynamically adjusts the priority of the analysis region, and based on the results of gradient perception and multi-scale feature extraction, automatically evaluates the potential importance of each key region; preferentially lock in the key regions containing significant precipitates or abnormal structures. The region screening module of the above solution conducts a preliminary analysis on the preprocessed welding pipeline image to identify regions with significant gradient changes, which are usually related to welding defects, precipitates, or other abnormal structures; according to the activity of the gradient change, set priorities for potential key regions to ensure that subsequent analysis can focus on the regions most likely to have problems. Significance: It can greatly reduce the computational workload of subsequent processing and avoid unnecessary analysis of irrelevant regions; by preferentially processing regions with significant gradient changes, the module can quickly lock in potential problem regions and provide an accurate starting point for feature extraction and analysis. The feature integration module can comprehensively and multi-dimensionally describe the state of the welding pipeline by integrating feature information at different scales. Significance: It can provide multi-level and multi-scale feature information to help the analysis comprehensively understand the state of the welding pipeline; feature capture at the micro level can reveal subtle defects or abnormalities, while feature recognition at the macro level can help analysts understand the health status of the overall structure; by integrating information, the module provides a solid foundation for subsequent evaluation and decision-making. The automatic evaluation module can automatically evaluate the potential importance of each key region, preferentially lock in the key regions containing significant precipitates or abnormal structures; it can ensure that analysis resources are concentrated in the regions that most need attention. Significance: It can achieve intelligent resource allocation, ensuring the efficiency and accuracy of the analysis process; by dynamically adjusting the priority, the module can respond in real time to changes in the analysis results, ensuring that under limited computational resources, the most critical regions are preferentially processed; it not only improves the efficiency of the analysis but also enhances the adaptive ability of the system.

[0085] In summary, each module of the feature extraction and fusion component in this embodiment works in coordination to jointly achieve the precise analysis and evaluation of the welded pipeline image; the region screening module provides a precise starting point for the analysis, the feature integration module provides multi-level and multi-dimensional feature information, and the automatic evaluation module ensures the efficient allocation of analysis resources; jointly constituting an efficient and intelligent welded pipeline image analysis system, providing strong support for the control of welding quality and defect detection.

[0086] Example 4: As Figure 4 shown, based on Example 3, the region screening module provided by the embodiment of the present invention includes:

[0087] A threshold preset sub-module, responsible for setting a dynamic saliency threshold for identifying the gradient change of the welded pipeline image; at the same time, the dynamic saliency threshold is adaptively adjusted according to the overall gradient distribution and local activity of the welded pipeline image.

[0088] A change capture sub-module, responsible for globally scanning the preprocessed welded pipeline image through gradient perception to capture the gradient change of each pixel point in the image; comparing with the dynamic saliency threshold to confirm whether it belongs to a region with significant gradient change.

[0089] A region discrimination sub-module, responsible for evaluating the gradient activity of each region. The gradient activity is based on the change frequency and local consistency of the gradient amplitude and gradient direction to distinguish active regions and stable regions; the active regions are marked as potential key regions and divided into different priorities according to their activity levels.

[0090] The working principle and beneficial effects of the above technical solution are as follows: The threshold preset sub-module of this embodiment sets a dynamic saliency threshold for identifying the gradient changes in the welded pipeline image; at the same time, the dynamic saliency threshold is adaptively adjusted according to the overall gradient distribution and local activity of the welded pipeline image; the change capture sub-module performs a global scan on the preprocessed welded pipeline image through gradient perception to capture the gradient changes of each pixel point in the image; by comparing with the dynamic saliency threshold, it is confirmed whether it belongs to the region with significant gradient changes; the region discrimination sub-module evaluates the gradient activity of each region, and the gradient activity is based on the change frequency and local consistency of the gradient amplitude and gradient direction to distinguish the active region and the stable region; the active region is marked as a potential key region and divided into different priorities according to its activity level. The threshold preset sub-module of the above solution dynamically adjusts the saliency threshold according to the overall gradient distribution and local activity of the welded pipeline image, avoiding over-detection or missed-detection problems caused by a fixed threshold; by analyzing the overall gradient distribution and local activity of the image, it ensures that the threshold can adapt to the characteristics of different images and working condition changes. The achieved significance: It improves the flexibility and adaptability of threshold setting, ensuring that regions with significant gradient changes can be effectively identified under different image qualities; by dynamically adjusting the threshold, it reduces misjudgment and missed judgment, and improves the accuracy of region screening. The change capture sub-module performs a global scan on the preprocessed welded pipeline image to capture the gradient changes of each pixel point, ensuring that no potential key region is missed; by comparing the captured gradient changes with the dynamic saliency threshold, it screens out the regions with significant gradient changes, providing basic data for region discrimination. The achieved significance: It realizes comprehensive and non-missing capture of gradient changes, ensuring the recognition coverage rate of key regions; by comparing with the dynamic saliency threshold, it screens out the truly significant regions and reduces the interference of invalid regions. The region discrimination sub-module evaluates the gradient activity of each region based on the change frequency and local consistency of the gradient amplitude and gradient direction, ensuring the scientificity and accuracy of the evaluation results; it marks the active region as a potential key region and divides it into different priorities according to the activity level, providing a clear key region for detection. The achieved significance: By evaluating the gradient activity, it accurately distinguishes the active region and the stable region, avoiding ineffective detection of the stable region; by dividing the priorities, it optimizes the allocation of detection resources and improves the detection efficiency.

[0091] In summary, in this embodiment, through dynamic significance threshold setting, gradient change capture, and gradient activity evaluation, the accurate identification and priority division of key regions are ensured; through priority division, detection resources are concentrated on high-activity regions, improving detection efficiency and resource utilization; dynamic threshold setting and gradient activity evaluation enable the method to adapt to different image characteristics and working conditions changes, with wide applicability; the region screening module provides a clear target region for detection, supporting an intelligent and automated detection process. Through the collaborative effect of the above sub-modules, the region screening module not only realizes the accurate screening of key regions in the welded pipeline image, but also provides an efficient and scientific basis for region division for detection, promoting the intelligent and precise development of the welded pipeline detection technology.

[0092] Embodiment 5: As Figure 5 shown, on the basis of Embodiment 4, the change capture sub-module provided by the embodiment of the present invention includes:

[0093] A vector calculation unit, which is responsible for loading the preprocessed welded pipeline image and allocating an independent calculation unit for each pixel point; for each pixel point in the welded pipeline image, calculating the gradient components in the horizontal and vertical directions to form a gradient vector; the gradient vector includes amplitude information and direction information;

[0094] Among them, the horizontal direction gradient component formula:

[0095]

[0096] The vertical direction gradient component formula:

[0097]

[0098] The gradient vector amplitude formula:

[0099]

[0100] The gradient vector direction formula:

[0101]

[0102] In the formula, G x (x, y) represents the horizontal direction gradient component at the position (x, y);

[0103] I(x + i, y + j) represents the original image gray value at the position (x + i, y + j);

[0104] I(x - i, y - j) represents the original image gray value at the position (x - i, y -);

[0105] ω ij(x, y) represents the local weight coefficient at the position (x, y), which is used to adjust the contribution of neighboring pixels; ||(i, j)|| 2 represents the Euclidean distance between the neighboring pixel (i, j) and the central pixel (x, y);

[0106] represents the adaptive scale parameter in the horizontal direction, which controls the width of the Gaussian kernel; M and N represent the width and height of the neighborhood window; exp(·) represents the exponential function, which is used to weight pixels with a greater distance; G y (x, y) represents the vertical direction gradient component at the position (x, y);

[0107] represents the adaptive scale parameter in the vertical direction, which controls the width of the Gaussian kernel;

[0108] G(x, y) represents the gradient vector magnitude at the position (x, y); ξ(x, y) represents the mixed gradient weight coefficient, which is used to enhance the accuracy of the gradient magnitude; θ(x, y) represents the gradient vector direction at the position (x, y); η(x, y) represents the direction correction coefficient, which is used to smooth the calculation of the gradient direction;

[0109] The numerical generation unit is responsible for constructing a local neighborhood window centered on the current pixel point, analyzing the distribution characteristics of the gradient vectors in the neighborhood, including the fluctuation range of the gradient magnitude and the consistency of the gradient direction; based on the local gradient distribution, quantifying the gradient change intensity of the current pixel point and generating a gradient change intensity value;

[0110] The characterization formation unit is responsible for starting from the upper left corner of the welded pipeline image, capturing the gradient change for each pixel point in the order of row by row and column by column; mapping the gradient change intensity value of each pixel point to the global gradient change intensity map to form a global distribution characterization of the gradient change; dynamically adjusting the significance threshold according to the distribution characteristics of the global gradient change intensity map to adapt to the overall gradient distribution and local activity of the image;

[0111] The region determination unit is responsible for comparing the gradient change intensity value of each pixel point with the dynamic significance threshold to determine whether it belongs to the region with significant gradient change; marking the pixel points determined to have significant gradient change as the significant region, and optimizing the boundary of the significant region to eliminate isolated points and noise interference.

[0112] The working principle and beneficial effects of the above technical solution are as follows: The vector calculation unit of this embodiment loads the preprocessed welded pipeline image and assigns an independent calculation unit to each pixel point; for each pixel point in the welded pipeline image, calculates the gradient components in the horizontal and vertical directions to form a gradient vector; the gradient vector contains amplitude information and direction information; the numerical generation unit constructs a local neighborhood window centered on the current pixel point and analyzes the distribution characteristics of the gradient vectors in the neighborhood, including the fluctuation range of the gradient amplitude and the consistency of the gradient direction; based on the local gradient distribution, quantifies the gradient change intensity of the current pixel point and generates a gradient change intensity value; the characterization formation unit starts from the upper left corner of the welded pipeline image and captures the gradient changes of each pixel point in the order of row by row and column by column; maps the gradient change intensity value of each pixel point to the global gradient change intensity map to form a global distribution characterization of the gradient change; according to the distribution characteristics of the global gradient change intensity map, dynamically adjusts the significance threshold to adapt to the overall gradient distribution and local activity of the image; the region determination unit compares the gradient change intensity value of each pixel point with the dynamic significance threshold to determine whether it belongs to the region with significant gradient changes; marks the pixel points determined to have significant gradient changes as the significant region and optimizes the boundary of the significant region to eliminate isolated points and noise interference. The vector calculation unit of the above solution calculates the gradient components in the horizontal and vertical directions for each pixel point to form a gradient vector containing amplitude and direction information, ensuring a comprehensive characterization of the gradient change; by assigning an independent calculation unit to each pixel point, high-efficiency parallel calculation is achieved, improving the speed and accuracy of gradient capture. The significance achieved: provides basic data support for gradient change analysis, ensuring the integrity and accuracy of gradient information; through parallel calculation, significantly improves the processing efficiency and meets the real-time requirements. The numerical generation unit quantifies the gradient change intensity of the current pixel point based on the local gradient distribution and generates a gradient change intensity value, providing a basis for the significance determination of the gradient change. The significance achieved: captures the subtle gradient changes in the image through local gradient analysis, avoiding missing potential key regions; the quantification of the gradient change intensity provides a scientific basis for subsequent determination, ensuring the accuracy and reliability of the determination. The characterization formation unit starts from the upper left corner of the image and captures the gradient changes of each pixel point in the order of row by row and column by column, ensuring global coverage; maps the gradient change intensity value of each pixel point to the global gradient change intensity map to form a global distribution characterization of the gradient change; according to the distribution characteristics of the global gradient change intensity map, dynamically adjusts the significance threshold to adapt to the overall gradient distribution and local activity of the image. The significance achieved: ensures the comprehensive capture of gradient changes through global gradient scanning, avoiding missing any potential key regions; the dynamic threshold adjustment makes the significance determination more adaptable and can cope with the characteristics and changes of different images.The region determination unit compares the gradient change intensity value of each pixel with the dynamic significance threshold to determine whether it belongs to the region with significant gradient change; marks the pixels determined to have significant gradient change as the significant region to ensure the integrity and continuity of the significant region; optimizes the boundary of the significant region to eliminate isolated points and noise interference and ensure the clarity and accuracy of the region. Significance achieved: Through significance determination, the key regions in the image are accurately identified, providing the target region for subsequent analysis and processing; the boundary optimization process improves the clarity of the significant region, reduces noise interference, and improves the reliability of the results.

[0113] In summary, through the collaborative work of each unit in this embodiment, a comprehensive capture and accurate determination of the gradient changes in the welded pipeline image are achieved, ensuring that key regions are not missed; through parallel computing and global scanning, the efficiency of gradient capture is significantly improved to meet the real-time requirements; the adjustment of the dynamic significance threshold enables the system to adapt to the characteristics and changes of different images, enhancing the robustness and versatility of the system; through boundary optimization processing, the clarity and accuracy of the significant region are improved, providing high-quality basic data for subsequent analysis and processing. The change capture sub-module provides a scientific basis and technical support for the analysis and processing of the welded pipeline image, with important practical value and innovative significance.

[0114] Embodiment 6: As Figure 6 shown, based on Embodiment 3, the feature integration module provided by the embodiment of the present invention includes:

[0115] The first-scale output sub-module is responsible for performing local refinement analysis on potential key regions to capture the geometric contours of tiny precipitates; identifying the boundary points of the precipitates by analyzing the spatial distribution of the gradient magnitude change and reconstructing their shape features through a geometric fitting algorithm; identifying the regularity and change patterns of the texture through the consistency analysis of the local gradient direction, and quantifying the roughness, directionality, and periodicity of the texture in combination with the fluctuation frequency of the gradient magnitude; extracting the boundary information between the precipitates and the surrounding regions by identifying the mutation points of the gradient direction; at the same time, distinguishing the sharpness of the edge and the smoothness of the transition region in combination with the local peak distribution of the gradient magnitude.

[0116] The second-scale output sub-module is responsible for globally scanning the welded pipeline image to identify the overall shape of the weld; determining the center line and boundary region of the weld through the global distribution analysis of the gradient magnitude; evaluating the straightness, curvature, and consistency with other structures of the weld in combination with the continuity analysis of the gradient direction; quantifying the distribution density, position offset, and distance from the weld boundary of the precipitates in the weld through the spatial distribution and direction change of the gradient magnitude, and identifying whether the precipitates are concentrated in specific regions of the weld or show an abnormal spatial distribution pattern in combination with the evaluation of the macroscopic gradient activity.

[0117] The correlation relationship establishment sub-module is responsible for establishing the correlation relationship between microscopic features and macroscopic morphology through multi-scale consistency analysis of gradient magnitude and direction; identifying whether the distribution of microscopic precipitates affects the overall gradient distribution of the weld, or whether the macroscopic weld morphology has a constraining effect on the characteristics of microscopic precipitates.

[0118] The working principle and beneficial effects of the above technical solution are as follows: The first-scale output sub-module of this embodiment conducts a local refinement analysis on potential key regions to capture the geometric contours of tiny precipitates; it uses spatial distribution analysis based on the change in gradient magnitude to identify the boundary points of the precipitates, and reconstructs their shape features through a geometric fitting algorithm; through the consistency analysis of the local gradient direction, it identifies the regularity and change patterns of the texture, and combines the fluctuation frequency of the gradient magnitude to quantify the roughness, directionality, and periodicity of the texture; by identifying the mutation points of the gradient direction, it extracts the boundary information between the precipitates and the surrounding regions; at the same time, it combines the local peak distribution of the gradient magnitude to distinguish the sharpness of the edge and the smoothness of the transition region; The second-scale output sub-module conducts a global scan of the welded pipeline image to identify the overall shape of the weld; through the global distribution analysis of the gradient magnitude, it determines the center line and boundary region of the weld; by combining the continuity analysis of the gradient direction, it evaluates the straightness, curvature, and consistency with other structures of the weld; through the spatial distribution and direction change of the gradient magnitude, it quantifies the distribution density, position offset, and distance from the weld boundary of the precipitates in the weld, and combines the evaluation of the macroscopic gradient activity to identify whether the precipitates are concentrated in specific regions of the weld or show an abnormal spatial distribution pattern; The correlation relationship establishment sub-module establishes the correlation relationship between the microscopic features and the macroscopic morphology through the multi-scale consistency analysis of the gradient magnitude and direction; it identifies whether the distribution of microscopic precipitates affects the overall gradient distribution of the weld, or whether the macroscopic weld morphology has a constraint effect on the characteristics of microscopic precipitates. The first-scale output sub-module of the above solution conducts a high-resolution scan of potential key regions to capture the geometric contours of tiny precipitates, ensuring the accurate extraction of microscopic features; through the spatial distribution analysis of the change in gradient magnitude and the geometric fitting algorithm, it reconstructs the shape features of the precipitates, providing a quantitative description of the geometric morphology of the precipitates; it uses the consistency analysis of the local gradient direction and the fluctuation frequency of the gradient magnitude to quantify the roughness, directionality, and periodicity of the texture, comprehensively describing the surface characteristics of the precipitates; by identifying the mutation points of the gradient direction and the local peak distribution of the gradient magnitude, it accurately extracts the edge information of the precipitates and distinguishes the sharpness of the edge and the smoothness of the transition region. The significance achieved: The first-scale output sub-module realizes the comprehensive capture of the shape, texture, and edge features of tiny precipitates, providing high-quality microscopic data for macroscopic analysis; by quantifying the geometric and texture features of the precipitates, it can effectively identify abnormal structures at the microscopic level, providing an important basis for the quality assessment of welded pipelines.The second-scale output sub-module globally scans the welded pipeline image to identify the overall shape of the weld seam, including the center line and the boundary area; through the global distribution analysis of the gradient magnitude and the continuity analysis of the gradient direction, it evaluates the straightness, curvature, and consistency with other structures of the weld seam; uses the spatial distribution and direction change of the gradient magnitude to quantify the distribution density, position offset, and distance from the weld seam boundary of the precipitates; combines the evaluation of the macroscopic gradient activity to identify whether the precipitates are concentrated in specific areas of the weld seam or show an abnormal spatial distribution pattern. Significance achieved: The second-scale output sub-module realizes the accurate identification of the overall shape of the weld seam, providing a macroscopic perspective for the structural integrity assessment of the welded pipeline; by quantifying the spatial relationship between the precipitates and the weld seam, it can identify whether the distribution of the precipitates affects the overall performance of the weld seam, providing support in the spatial dimension for anomaly detection. The correlation relationship establishment sub-module establishes the correlation relationship between the microscopic features and the macroscopic shape through the multi-scale consistency analysis of the gradient magnitude and direction; identifies whether the distribution of the microscopic precipitates affects the overall gradient distribution of the weld seam, such as whether the precipitates cause gradient anomalies in local areas of the weld seam; evaluates whether the macroscopic weld seam shape has a constraining effect on the characteristics of the microscopic precipitates, such as whether the curvature of the weld seam affects the distribution pattern of the precipitates. Significance achieved: The correlation relationship establishment sub-module realizes the dynamic integration of the microscopic features and the macroscopic shape, ensuring the comprehensiveness and consistency of the analysis results; by establishing the correlation relationship between the microscopic and macroscopic, it can accurately locate the source of the anomaly, such as determining whether the anomaly is caused by microscopic precipitates or by the macroscopic weld seam shape; improves the comprehensive analysis ability of the feature integration module, providing more comprehensive technical support for the quality assessment and anomaly detection of the welded pipeline.

[0119] In summary, through the first-scale and second-scale output sub-modules in this embodiment, hierarchical feature extraction from the microscopic to the macroscopic is realized, ensuring the comprehensiveness and accuracy of the analysis results; combined with the multi-dimensional analysis of the gradient magnitude and direction, it can capture features in multiple aspects such as shape, texture, edge, and spatial relationship, avoiding the limitations of single-feature analysis; through the dynamic integration of microscopic and macroscopic features, abnormal structures in the welded pipeline can be more accurately identified, providing a reliable basis for quality assessment and process optimization; through hierarchical and multi-dimensional feature extraction and integration, the innovation of the technical solution is reflected, providing new technical ideas and methods for the field of welded pipeline image analysis. In short, through the collaborative work of each sub-module, the feature integration module realizes the comprehensive capture and integration of key features in the welded pipeline image, providing strong technical support for welding quality assessment and anomaly detection.

[0120] Embodiment 7: As Figure 7 shown, based on Embodiment 6, the first-scale output sub-module provided by the embodiment of the present invention includes:

[0121] The boundary point confirmation unit is responsible for analyzing the spatial distribution of the gradient magnitude in the welded pipeline image, identifying the regions with significant changes in the gradient magnitude, which correspond to the boundaries of the precipitates or the regions of texture changes; by setting the threshold of the gradient magnitude, identifying the mutation points of the gradient magnitude, which are the boundary points of the precipitates;

[0122] The boundary point clustering unit is responsible for clustering the identified boundary points to form several boundary point sets, and each set corresponds to the contour of a precipitate;

[0123] The parameter calculation unit is responsible for selecting the fitting curve model of a straight line, a circle, an ellipse or a polygon according to the shape characteristics of the precipitate; using the coordinate information of the boundary points, estimating the parameters of the fitting curve by the least square method; according to the parameters of the fitting curve, reconstructing the geometric contour of the precipitate and calculating its shape characteristic parameters (such as area, perimeter, major axis, minor axis, etc.).

[0124] The working principle and beneficial effects of the above technical solution are as follows: The boundary point confirmation unit in this embodiment analyzes the spatial distribution of the gradient magnitude in the welding pipeline image, identifies the regions where the gradient magnitude changes significantly, which correspond to the boundaries of the precipitates or the regions of texture changes; by setting a threshold for the gradient magnitude, the abrupt change points of the gradient magnitude are identified, which are the boundary points of the precipitates; the boundary point clustering unit clusters the identified boundary points to form several boundary point sets, and each set corresponds to the contour of a precipitate; the parameter calculation unit selects a fitting curve model of a straight line, a circle, an ellipse or a polygon according to the shape characteristics of the precipitate; uses the coordinate information of the boundary points to estimate the parameters of the fitting curve by the least squares method; according to the parameters of the fitting curve, reconstructs the geometric contour of the precipitate and calculates its shape characteristic parameters (such as area, perimeter, major axis, minor axis, etc.). The boundary point confirmation unit in the above solution analyzes the spatial distribution of the gradient magnitude in the welding pipeline image, identifies the regions where the gradient magnitude changes significantly, and these regions usually correspond to the boundaries of the precipitates or the regions of texture changes; by setting a threshold for the gradient magnitude, the abrupt change points of the gradient magnitude are accurately identified, and these points are the boundary points of the precipitates. Significance achieved: The boundary point confirmation unit can accurately capture the boundary information of the precipitates, avoiding boundary recognition errors caused by noise or blurring; the identified boundary points provide key inputs for subsequent clustering and geometric fitting, and are the basis for shape feature extraction. The boundary point clustering unit clusters the identified boundary points to form several boundary point sets, and each set corresponds to the contour of a precipitate; separates the boundary points of different precipitates through clustering algorithms (such as K-means or DBSCAN) to avoid cross-interference. Significance achieved: The boundary point clustering unit can process the boundary points of multiple precipitates simultaneously, realizing multi-objective analysis and improving the analysis efficiency; through clustering, it ensures the integrity of the contour information of each precipitate and avoids shape distortion caused by boundary point confusion. The parameter calculation unit selects a suitable fitting curve model (such as a straight line, a circle, an ellipse or a polygon) according to the shape characteristics of the precipitate; uses the coordinate information of the boundary points to estimate the parameters of the fitting curve by the least squares method; according to the parameters of the fitting curve, reconstructs the geometric contour of the precipitate and calculates its shape characteristic parameters (such as area, perimeter, major axis, minor axis, etc.). Significance achieved: The parameter calculation unit can accurately reconstruct the geometric contour of the precipitate, providing a reliable basis for shape feature extraction; by calculating the shape characteristic parameters, it realizes the quantitative analysis of the shape of the precipitate, providing data support for subsequent quality assessment and anomaly detection; the use of the fitting algorithm significantly improves the accuracy of shape feature extraction and avoids errors caused by noise or blurring in traditional methods.

[0125] In summary, the first-scale output sub-module of this embodiment realizes the automated analysis of the shape characteristics of precipitates through boundary point confirmation, clustering, and parameter calculation, significantly improving the analysis efficiency; through gradient magnitude analysis, clustering algorithms, and geometric fitting, it significantly improves the accuracy of extracting the shape characteristics of precipitates, avoiding errors caused by noise or blurring in traditional methods; the extracted shape characteristics of precipitates provide important inputs for subsequent micro-macro feature correlation analysis, such as determining whether the distribution of precipitates affects the overall performance of the weld; the quantified shape feature parameters provide a reliable basis for the quality assessment and anomaly detection of welded pipelines, such as determining whether the size and distribution density of precipitates exceed the allowable range. The first-scale output sub-module realizes the precise capture and quantitative analysis of the shape characteristics of precipitates through the collaborative work of three units: boundary point confirmation, clustering, and parameter calculation. It not only improves the analysis efficiency and accuracy but also provides important support for subsequent micro-macro feature correlation analysis, quality assessment, and anomaly detection, which is an indispensable technical link in the image analysis of welded pipelines.

[0126] Embodiment 8: As Figure 8 shown, based on Embodiment 6, the second-scale output sub-module provided by the embodiment of the present invention includes:

[0127] The first anomaly judgment unit is responsible for dividing the weld area into a high-gradient active area and a low-gradient active area based on the global distribution characteristics of the gradient magnitude, analyzing the gradient activity, and evaluating whether the distribution of precipitates presents an abnormal spatial distribution pattern; if the precipitates show a significant concentrated distribution characteristic in the high-gradient active area or the low-gradient active area, it indicates that its distribution pattern may be abnormal;

[0128] The second anomaly judgment unit is responsible for comprehensively evaluating whether the distribution presents an abnormal pattern by combining the distribution density and position offset of the precipitates; if the precipitates show a high-density distribution in a specific area of the weld and its position offset significantly deviates from the weld center line, it indicates that its distribution pattern may be abnormal;

[0129] The third anomaly judgment unit is responsible for determining whether the distribution of precipitates presents an abnormal pattern by combining the analysis results of gradient activity and direction mutation points. If the precipitates show a significant concentrated distribution in the high-gradient active area and the number of gradient direction mutation points at its boundary significantly increases, it indicates that its distribution pattern may be abnormal.

[0130] The working principle and beneficial effects of the above technical solution are as follows: The first anomaly judgment unit in this embodiment divides the weld area into a high-gradient active area and a low-gradient active area based on the global distribution characteristics of the gradient amplitude, analyzes the gradient activity, and evaluates whether the distribution of the precipitates shows an abnormal spatial distribution pattern; if the precipitates show a significant concentrated distribution characteristic in the high-gradient active area or the low-gradient active area, it indicates that its distribution pattern may be abnormal; the second anomaly judgment unit combines the distribution density and position offset of the precipitates to comprehensively evaluate whether the distribution shows an abnormal pattern; if the precipitates show a high-density distribution in a specific area of the weld and its position offset significantly deviates from the weld center line, it indicates that its distribution pattern may be abnormal; the third anomaly judgment unit combines the analysis results of the gradient activity and the direction mutation points to determine whether the distribution of the precipitates shows an abnormal pattern. If the precipitates show a significant concentrated distribution in the high-gradient active area and the number of gradient direction mutation points at its boundary increases significantly, it indicates that its distribution pattern may be abnormal. The first anomaly judgment unit of the above solution can more accurately identify the areas in the weld that may have defects through the division of gradient activity, avoiding missed inspections or false inspections; the analysis of gradient activity provides basic data support for subsequent anomaly judgment, reducing the need for manual intervention and improving the detection efficiency; through the evaluation of global distribution characteristics, it can more comprehensively reflect the weld quality and ensure the reliability of the detection results. The second anomaly judgment unit can effectively identify defects in local areas of the weld, such as pores and slag inclusions, by analyzing the distribution density and position offset; the identification of abnormal distribution patterns provides data support for the optimization of the welding process, helps to improve welding parameters, and enhances the welding quality; timely discovery of abnormal position offset can avoid potential safety hazards caused by insufficient weld strength. The third anomaly judgment unit can identify more complex defect types, such as cracks and lack of fusion, through the analysis of gradient direction and mutation points; the introduction of gradient direction mutation points provides a more refined dimension for anomaly judgment, further improving the detection accuracy; the analysis results provide a scientific basis for the in-depth evaluation of weld quality, helping to achieve more comprehensive quality control.

[0131] In summary, through multi-dimensional and multi-level anomaly judgment in this embodiment, the weld quality can be more comprehensively evaluated, reducing the occurrence of missed inspections and false inspections; the automated analysis ability provides technical support for the intelligent detection of welding quality, promoting the intelligent transformation of the industrial detection field; by timely discovering abnormal distribution patterns in the weld, potential safety accidents caused by welding defects can be effectively avoided, ensuring the safety and stability of industrial production.

[0132] Embodiment 9: As Figure 9 shown, on the basis of Embodiment 1, the decision-making and feedback optimization component provided by the embodiment of the present invention includes:

[0133] The risk assessment and grading module is responsible for conducting risk assessment on each key area based on the determination results of abnormal patterns and generating risk levels (such as low risk, medium risk, high risk).

[0134] The dynamic strategy optimization module is responsible for dynamically adjusting the detection strategy according to the risk assessment results.

[0135] The decision-making generation module is responsible for generating a detailed detection report, including the types of precipitates (such as pores, slag inclusions, cracks), distributions (such as concentrated distribution, dispersed distribution), sizes (such as maximum diameter, average diameter), and potential risk levels.

[0136] The working principle and beneficial effects of the above technical solution are as follows: The risk assessment and grading module in this embodiment conducts risk assessment on each key area based on the determination result of the abnormal pattern, and generates risk levels (such as low risk, medium risk, high risk); the dynamic strategy optimization module dynamically adjusts the detection strategy according to the risk assessment result; the decision-making generation module generates a detailed detection report, including the type of precipitates (such as pores, slag inclusions, cracks), distribution (such as concentrated distribution, dispersed distribution), size (such as maximum diameter, average diameter), and potential risk level. The risk assessment and grading module of the above solution accurately identifies the abnormal areas in the welded pipeline through multi-dimensional feature fusion (such as precipitate density, morphology, position, gradient activity, etc.), avoiding missed detection or false detection; converts the detection result into a quantifiable risk level (such as low risk, medium risk, high risk), providing a scientific basis for decision-making; based on real-time detection data and historical data, dynamically adjusts the criteria and thresholds for risk assessment to ensure the accuracy and adaptability of the assessment result. The achieved significance is as follows: It helps operators quickly identify high-risk areas in the welded pipeline, avoiding quality problems or safety hazards caused by missed detection or false detection; through the quantifiable risk level, it provides a clear direction for subsequent repair or process improvement, enhancing the quality control level of the welded pipeline; by dynamically adjusting the assessment criteria, it can continuously optimize its own detection ability and adapt to changes in different working conditions and environments. The dynamic strategy optimization module dynamically adjusts the detection strategy according to the risk assessment result. For example, it improves the detection accuracy for high-risk areas and reduces the detection frequency for low-risk areas, thereby enhancing the overall detection efficiency; through intelligent allocation of computing resources (such as image processing computing power, feature extraction dimensions, etc.), it ensures that the detection of key areas is given priority, avoiding resource waste; it feeds back the detection result to the system in real time for optimizing subsequent detection strategies, forming a closed-loop control. The achieved significance is as follows: By dynamically adjusting the strategy, it can significantly improve the detection efficiency while ensuring the detection accuracy, reducing the waste of time and resources; it can automatically adjust the detection strategy according to changes in different working conditions and environments to ensure the stability and reliability of the detection result; through real-time feedback and closed-loop optimization, the system can continuously learn and improve, gradually enhancing its own detection ability and intelligent level. The decision-making generation module integrates the detection results (such as precipitate type, distribution, size, risk level) into a structured detection report, providing comprehensive data support; through forms such as charts and heat maps, it intuitively displays the detection results, facilitating quick understanding and analysis by operators; based on the detection results, it generates targeted repair or process improvement suggestions, providing clear guidance for subsequent operations. The achieved significance is as follows: Through the structured and visualized detection report, operators can quickly grasp the quality status of the welded pipeline, shortening the decision-making time; the generated guiding suggestions are based on scientific data analysis, avoiding the subjectivity and errors of human judgment and enhancing the accuracy of decision-making; the detection report not only provides a basis for current quality control but also accumulates data support for long-term process optimization and quality improvement.

[0137] In summary, this embodiment can significantly improve the detection efficiency while ensuring the detection accuracy, meeting the high-efficiency requirements of industrial production; it can dynamically adjust the detection strategy and risk assessment criteria according to the real-time detection results and historical data, gradually improving its own intelligent level; through accurate risk assessment and detailed detection reports, it provides a scientific basis for the quality control of welded pipelines, helping enterprises reduce quality risks and production costs.

[0138] Embodiment 10: As Figure 10 shown, based on Embodiments 1 - 9, the method for detecting precipitates in welded pipelines based on image recognition provided by the embodiment of the present invention includes the following steps:

[0139] S100: Preprocess the original welded pipeline image, dynamically adjust the denoising parameters, identify the weld edge and precipitate area by analyzing local gradient information, and enhance the contrast.

[0140] S200: Extract multi-level feature information from the contrast-enhanced welded pipeline image, including local details (such as tiny precipitates) and global structures (such as the overall shape of the weld); at the same time, capture features at different scales and automatically focus on key areas.

[0141] S300: Make a decision based on the detection results of the key area; dynamically adjust the detection strategy according to the detection results, and at the same time generate a detailed detection report, including the type, distribution, size and potential risk level of the precipitates.

[0142] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the original welded pipeline image is first preprocessed, and the denoising parameters are dynamically adjusted. By analyzing the local gradient information, the weld edge and the precipitate area are identified, and the contrast is enhanced. Secondly, multi-level feature information is extracted from the contrast-enhanced welded pipeline image, including local details (such as minute precipitates) and global structures (such as the overall shape of the weld). At the same time, features at different scales are captured, and the key areas are automatically focused on. Finally, a decision is made based on the detection results of the key areas; the detection strategy is dynamically adjusted according to the detection results, and a detailed detection report is generated at the same time, including the type, distribution, size, and potential risk level of the precipitates. For the step S100 of image preprocessing and contrast enhancement in the above solution, the dynamic adjustment of the denoising parameters can adapt to the welded pipeline images with different image qualities, avoiding the loss of details caused by excessive denoising or the noise interference introduced by insufficient denoising; by analyzing the local gradient information, the weld edge and the precipitate area are accurately identified, laying a foundation for subsequent feature extraction; the contrast between the weld and the background, and between the precipitate and the weld is enhanced, making the minute precipitates and weld details clearer, facilitating subsequent feature extraction. The significance achieved is: improving the image quality and ensuring the accuracy of feature extraction; by dynamically adjusting the denoising parameters and contrast enhancement, adapting to the welded pipeline images under different working conditions, and improving the robustness and generality of the method. For the step S200 of multi-level feature extraction and key area focusing, at the microscopic level, the shape, texture, and edge features of minute precipitates are captured to ensure the accurate identification of details; at the macroscopic level, the overall shape of the weld and its spatial relationship with the precipitates are analyzed to provide detection information from a global perspective; through multi-scale analysis, the key information that may be missed at a single scale is avoided, ensuring the comprehensiveness of detection; based on the gradient information and the feature extraction results, the key areas containing significant precipitates or abnormal structures are automatically locked, improving the detection efficiency. The significance achieved is: realizing the comprehensive feature extraction from local to global, ensuring the accuracy and integrity of the detection results; by multi-scale analysis and key area focusing, significantly improving the detection efficiency and reducing the waste of computing resources. For the step S300 of decision-making and dynamic adjustment, according to the detection results of the key areas, the detection strategy is dynamically adjusted; for example, more refined analysis is performed on high-priority areas, and rapid screening is performed on low-priority areas to optimize the detection process; detailed information such as the type, distribution, size, and potential risk level of the precipitates is provided to support decision-making with data. The significance achieved is: by dynamically adjusting the detection strategy, realizing the intelligentization and high efficiency of the detection process, and adapting to the requirements of different scenarios; generating a detailed detection report, providing a scientific basis for the quality assessment, maintenance decision-making, and risk warning of the welded pipeline.

[0143] In summary, through dynamic preprocessing, multi-level feature extraction, and intelligent decision-making, this embodiment significantly improves the accuracy and efficiency of detecting precipitates in welded pipelines. The dynamic adjustment mechanism enables the method to adapt to different working conditions and image qualities, with wide applicability. The generated detailed inspection reports provide reliable data support for the quality control and maintenance of welded pipelines, reducing potential risks. It provides an intelligent and automated solution for the field of welded pipeline inspection, promoting the progress of industrial inspection technology. Through the synergistic effect of the above steps, not only the accurate detection of precipitates in welded pipelines is achieved, but also an efficient and intelligent technical solution is provided for the field of industrial inspection.

[0144] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of equivalent technologies of the present invention, the present invention also intends to include these modifications and variations.

Claims

1. A welding pipeline precipitate detection system based on image recognition, characterized in that It includes the following steps: An image preprocessing component, which is responsible for preprocessing the original welded pipeline image, dynamically adjusting the denoising parameters, identifying the weld edges and precipitate regions by analyzing the local gradient information, and enhancing the contrast; A feature extraction and fusion component, which is responsible for extracting multi-level feature information from the contrast-enhanced welded pipeline image, including local details and global structure; at the same time, capturing features at different scales and automatically focusing on key regions; A decision-making and feedback optimization component, which is responsible for making decisions based on the detection results of key regions; dynamically adjusting the detection strategy according to the detection results, and generating a detailed detection report at the same time.

2. The image recognition-based welding pipeline precipitate detection system according to claim 1, wherein The image preprocessing component includes: An adaptive denoising module, which is responsible for intelligently adjusting the denoising parameters according to the local features of the original welded pipeline image; identifying which regions contain weld edge and precipitate information by real-time monitoring of the local gradient changes of the original welded pipeline image; A gradient analysis module, which is responsible for adopting multi-scale gradient analysis during the identification process, capturing the details of the original welded pipeline image at different resolution levels, and identifying the weld edge and precipitate regions; A parameter feedback optimization module, which is responsible for setting a dynamic parameter feedback optimization mechanism, collecting effect feedback in real time, and dynamically adjusting the denoising and contrast enhancement parameters according to the feedback results.

3. The image recognition-based welding pipeline precipitate detection system according to claim 1, wherein, The feature extraction and fusion component includes: A region screening module, which is responsible for preliminarily analyzing the preprocessed welded pipeline image, identifying regions with significant gradient changes, and marking them as potential key regions; A feature integration module, which is responsible for layer-by-layer extracting feature information from the welded pipeline image marked with potential key regions; An automatic evaluation module, which is responsible for dynamically adjusting the priority of the analysis regions, and automatically evaluating the potential importance of each key region based on the results of gradient perception and multi-scale feature extraction; Prioritize and lock the key regions containing significant precipitates or abnormal structures.

4. The image recognition-based welding pipeline precipitate detection system according to claim 3, wherein The region screening module sets the priority of the key regions according to the activity of the gradient change.

5. The image recognition-based welding pipeline precipitate detection system according to claim 3, wherein The region screening module includes: A threshold preset sub-module, which is responsible for setting a dynamic significance threshold for identifying the gradient changes of the welded pipeline image; at the same time, the dynamic significance threshold is adaptively adjusted according to the overall gradient distribution and local activity of the welded pipeline image; A change capture sub-module, which is responsible for globally scanning the preprocessed welded pipeline image through gradient perception, capturing the gradient changes of each pixel point in the image; comparing with the dynamic significance threshold to confirm whether it belongs to the region with significant gradient changes; A region discrimination sub-module, which is responsible for evaluating the gradient activity of each region. The gradient activity is based on the change frequency and local consistency of the gradient amplitude and gradient direction, and differentiates the active region and the stable region; the active region is marked as a potential key region and divided into different priorities according to its activity level.

6. The image recognition-based welding pipeline precipitate detection system according to claim 5, characterized in that The change capture sub-module includes: A vector calculation unit, which is responsible for loading the preprocessed welded pipeline image and allocating independent calculation units for each pixel point; A numerical generation unit, which is responsible for constructing a local neighborhood window centered on the current pixel point, analyzing the distribution characteristics of the gradient vectors in the neighborhood; quantifying the gradient change intensity of the current pixel point based on the local gradient distribution, and generating a gradient change intensity value; A characterization formation unit, responsible for starting from the upper left corner of the welded pipeline image and capturing the gradient change of each pixel point in the order of row by row and column by column; An area determination unit, responsible for comparing the gradient change intensity value of each pixel point with the dynamic saliency threshold to determine whether it belongs to the area with significant gradient change.

7. The image recognition-based welding pipeline precipitate detection system according to claim 6, wherein, The characterization formation unit maps the gradient change intensity value of each pixel point into the global gradient change intensity map to form a global distribution characterization of the gradient change.

8. The image recognition-based welding pipeline precipitate detection system according to claim 6, wherein The area determination unit marks the pixel points determined to have significant gradient change as significant areas.

9. The image recognition-based welding pipeline precipitate detection system according to claim 3, wherein A feature integration module, including: A first-scale output sub-module, responsible for performing local refinement analysis on potential key areas to capture the geometric contours of tiny precipitates; using spatial distribution analysis based on gradient magnitude change to identify the boundary points of precipitates, and reconstructing their shape features through geometric fitting algorithms; A second-scale output sub-module, responsible for globally scanning the welded pipeline image to identify the overall morphology of the weld; combining the evaluation of macroscopic gradient activity to identify whether the precipitates are concentrated in specific areas of the weld or show abnormal spatial distribution patterns; An association relationship establishment sub-module, responsible for establishing the association relationship between microscopic features and macroscopic morphology through multi-scale consistency analysis of gradient magnitude and direction.

10. A method for detecting welding pipeline precipitates based on image recognition, characterized in that, Including the following steps: Preprocess the original welded pipeline image, dynamically adjust the denoising parameters, identify the weld edge and precipitate area by analyzing local gradient information, and enhance the contrast; Extract multi-level feature information from the contrast-enhanced welded pipeline image, including local details and global structure; at the same time, capture features at different scales and automatically focus on key areas; Make decisions based on the detection results of key areas; dynamically adjust the detection strategy according to the detection results, and generate a detailed detection report at the same time.

Images