Visual recognition and positioning system for condenser tube weld edge

Abstract

The present application relates to the technical field of condenser tube welding quality detection, and discloses a visual identification and positioning system for the edge of a condenser tube welding spot, which comprises an image acquisition module, an edge feature module, a correlation analysis module, a shape optimization module, a reference positioning module and a decision output module.The image acquisition module acquires multi-angle gray-scale images and sets a scanning interval; the edge feature module divides contour nodes and analyzes gradient features to generate a vector; the correlation analysis module extracts key indicators, establishes positioning rules and acquires correction parameters; the shape optimization module identifies spatial correlation patterns, calibrates contour indicators and calculates an offset; the reference positioning module derives an optimal positioning threshold and generates a deviation sequence; and the decision output module integrates the deviation sequence to form a positioning scheme.This system realizes high-precision identification and positioning of welding spot edge features through multi-module cooperation, improves the automation level and precision of condenser tube welding quality detection, and is suitable for welding quality detection scenarios in automated production lines.

Description

Technical Field

[0001] This invention relates to the field of condenser tube welding quality inspection technology, specifically a visual recognition and positioning system for the edge of condenser tube weld points. Background Technology

[0002] In the refrigeration equipment manufacturing industry, condenser tubes, as key heat exchange components, directly affect the performance and reliability of the equipment due to their welding quality. Accurate identification and positioning of weld joint edges is a core aspect of weld quality assessment. Traditional manual inspection methods suffer from low efficiency, high subjectivity, and susceptibility to human error, making it difficult to meet the high-precision, high-efficiency inspection requirements of automated production lines. With the development of industrial automation and machine vision technology, image recognition-based inspection methods are gradually becoming mainstream. However, existing technologies still face many challenges in handling the complex morphological features of condenser tube weld joint edges.

[0003] On the one hand, the geometric morphology of the weld joint edges of condenser tubes is diverse. Different welding process parameters can lead to complex features such as curvature changes, contour polarity differences, and uneven morphological density at the weld joint edges. Traditional fixed-threshold edge detection algorithms are difficult to adapt to the dynamic recognition requirements of multi-morphological features. For example, when facing weld joint edges with large curvature changes, detection methods based on a single gradient threshold are prone to edge omissions or false detections. On the other hand, existing visual positioning systems lack effective analysis of the spatial association patterns of contour nodes when processing multi-angle image data, making it difficult to establish accurate positioning judgment rules, resulting in insufficient positioning accuracy. For example, in the process of multi-view image stitching, the lack of dynamic calibration of the spatial offset trend of contour nodes can easily generate cumulative errors, affecting the accuracy of the final positioning result.

[0004] Furthermore, existing systems fail to fully consider the dynamic characteristics of contour nodes when processing gradient fluctuation data. Their differentiated processing mechanisms for strong contour nodes and steady-state nodes are inadequate, resulting in insufficient representation capabilities of edge feature vectors. For example, for steady-state nodes, simply ignoring the gradient superposition information of their neighboring nodes may lead to the loss of edge features, affecting the accuracy of subsequent correlation analysis. Simultaneously, in the derivation of positioning thresholds, existing methods are mostly based on fixed standard template matching, lacking adaptive adjustment to the dynamic characteristics of the current contour offset, making it difficult to achieve rapid matching of the optimal positioning threshold in complex welding environments.

[0005] Existing condenser tube weld edge detection technologies have significant shortcomings in handling complex morphological features, multi-view data correlation analysis, and dynamic feature adaptive adjustment. There is an urgent need for a visual recognition and positioning system that can achieve multi-angle image data fusion, dynamic calibration of contour features, and accurate derivation of positioning thresholds to meet the actual needs of efficient detection of condenser tube welding quality on automated production lines. Summary of the Invention

[0006] The purpose of this invention is to provide a visual recognition and positioning system for the edge of the condenser tube weld joint, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a visual recognition and positioning system for the edge of a condenser tube weld joint, the system comprising:

[0008] The image acquisition module is used to acquire grayscale image data of the weld seam area of ​​the condenser tube from multiple angles and to set the contour scanning interval corresponding to the weld point positioning.

[0009] The edge feature module is used to divide the contour scanning range into multiple contour nodes, perform gradient feature parsing on the image data of each contour node, and generate the edge feature vector corresponding to the contour node.

[0010] The correlation analysis module is used to extract key contour indicators from edge feature vectors, establish positioning judgment rules associated with contour nodes, and obtain the geometric correction parameters corresponding to the rules.

[0011] The morphology optimization module is used to identify spatial correlation patterns in geometric correction parameters, dynamically calibrate key contour indicators based on correlation patterns, and calculate the contour offset of each contour node under different calibration strategies.

[0012] The baseline positioning module is used to derive the optimal positioning threshold based on the contour offset and generate a contour deviation sequence by comparing the current contour feature value with the optimal positioning threshold.

[0013] The decision output module is used to analyze the contour deviation sequence and integrate the contour deviation sequence into a weld point positioning decision scheme based on the spatial offset trend of the contour nodes.

[0014] Preferably, the implementation of the edge feature module includes: constructing a weld point feature library corresponding to the contour nodes, wherein the weld point feature library contains image data and geometric parameter vectors of gradient feature mapping;

[0015] Perform similar contour matching on the geometric parameter vectors, and divide the geometric parameter vectors into contour clusters based on the matching results; extract the morphological focal points of the image data from the contour clusters, and set the focal points as contour nodes.

[0016] Preferably, the contour clustering group that divides the geometric parameter vectors further includes:

[0017] Based on the spatial properties and gradient intensity in the geometric parameter vector, the curvature distribution, contour polarity, and morphological density parameters are extracted, and edge feature labels are generated based on the above parameters.

[0018] Edge feature labels are associated with geometric parameter vectors. By calculating the spatial similarity between feature labels, parameter vectors with similarity higher than a preset contour threshold are selected to form contour clusters.

[0019] Preferably, the implementation methods for generating the edge feature vectors corresponding to the contour nodes include:

[0020] For each contour node, based on the spatial position of the node within the contour scanning range, the gradient fluctuation data of the node within a preset window is obtained, and the gradient fluctuation coefficient of the node is calculated.

[0021] When the gradient fluctuation coefficient exceeds the first gradient threshold, the node is marked as a strong contour node, and its image data is extracted to form an edge feature vector; when the gradient fluctuation coefficient is lower than the first gradient threshold, the node is marked as a steady-state node, and the image data of the adjacent nodes of the node are gradient superimposed, and the superimposed data is reconstructed into an edge feature vector.

[0022] Preferably, the implementation of the correlation analysis module includes:

[0023] The main contour proportion, secondary contour proportion, and morphological fluctuation parameters are separated from the edge feature vector, and the positioning judgment rules of the contour nodes are generated based on the above parameters.

[0024] If the number of contour nodes covered by the current positioning judgment rule is less than the preset positioning threshold, then the edge feature vectors of adjacent contour nodes are traversed, and geometric indicators not included in the judgment rules of adjacent nodes are added to the current rule.

[0025] Preferably, the implementation of the morphology optimization module includes: obtaining the spatial factor of the morphology switching frequency and the fluctuation factor of the gradient jump amplitude in the spatial correlation mode;

[0026] Construct a focusing weight matrix associated with the spatial factor and the volatility factor, and determine the profile offset under different calibration strategies based on the probability distribution of each element in the matrix.

[0027] Preferably, constructing the focused weight matrix further includes:

[0028] Identify the morphological periodic characteristics of the spatial factor. If the current periodic characteristics completely match the preset contour period, then set the spatial factor as the starting index of the focusing weight matrix.

[0029] Calculate the correlation matching degree between spatial factors and volatility factors, and generate intermediate and terminal indices of the focusing weight matrix in descending order of matching degree;

[0030] The path backtracking is performed on the terminating index. When the matching degree of the terminating index is lower than the preset matching threshold, it is output as the final distribution of the focusing weight matrix.

[0031] Preferably, the implementation method for calculating the contour offset of each contour node under different calibration strategies includes:

[0032] The mean of the spatial factors and the range of the volatility factors for each terminating index in the statistical focus weight matrix are calculated, and the global variance of all index factors is calculated.

[0033] The spatial offset coefficient is obtained by subtracting the spatial factor mean of a single terminating index from the spatial factor mean of its adjacent indices and dividing by the global variance. At the same time, the ratio of the fluctuation factor range to the global variance is calculated, and the two are weighted and summed to obtain the profile offset of the index.

[0034] Preferably, the methods for deriving the optimal positioning threshold include:

[0035] Extract the morphological pattern from the standard weld point database that is closest to the current contour offset, calculate the cosine similarity between the two in spatial distribution, and use it as the first dynamic reference value.

[0036] The difference between the current contour offset and the standard shape pattern in the number of contour peaks is statistically analyzed, and the difference is used as a second dynamic reference value.

[0037] Based on a linear combination of the first dynamic reference value and the second dynamic reference value, the optimal positioning threshold in the preset positioning threshold table is matched.

[0038] Preferably, the decision output module is implemented by dividing the main offset interval and the secondary offset interval according to the spatial offset direction of each node in the contour deviation sequence;

[0039] The convergence rate of the contour deviation in the main offset interval and the diffusion rate of the contour deviation in the secondary offset interval are extracted, and the two are fused and calculated according to the spatial weight of the contour nodes to generate the adjustment parameters of the weld point positioning decision scheme.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] The condenser tube weld joint edge visual recognition and positioning system provided by this invention constructs a complete technical chain from image acquisition to positioning decision through the collaborative work of multiple modules. It effectively solves the problems of inaccurate recognition of complex morphological features, insufficient positioning accuracy, and poor dynamic adaptability in the prior art, and significantly improves the automation level and detection accuracy of condenser tube weld joint edge detection, specifically in the following aspects:

[0042] In the image acquisition and edge feature analysis stages, the image acquisition module obtains multi-angle grayscale image data and sets contour scanning intervals, providing a multi-dimensional data foundation for subsequent high-precision analysis. The edge feature module achieves refined segmentation of contour nodes by constructing a weld point feature library, clustering geometric parameter vectors, and setting morphological focal points. Differential processing mechanisms for different types of contour nodes (strong contour nodes and stable nodes) effectively improve the representational capability of edge feature vectors. For example, for stable nodes, the feature vector is reconstructed by superimposing gradients from adjacent nodes, avoiding feature loss caused by missing data from a single node, making edge feature analysis more comprehensive and accurate.

[0043] The correlation analysis module generates positioning judgment rules by separating the proportion of the main contour, the proportion of the secondary contour, and morphological fluctuation parameters. It then dynamically expands the rule coverage based on the number of contour nodes, enhancing the system's adaptability to complex contour features. This dynamic rule expansion mechanism effectively addresses the diversity of weld point edge features under different welding processes, ensuring the comprehensiveness and accuracy of the positioning judgment rules and avoiding positioning deviations caused by insufficient rule coverage.

[0044] The morphology optimization module identifies morphology switching frequency and gradient jump amplitude parameters in spatial correlation patterns, constructs a focusing weight matrix, and dynamically calibrates key contour indicators, achieving accurate calculation of contour offsets. The construction process of the focusing weight matrix fully considers the correlation and matching degree between spatial factors and fluctuation factors. Through morphological periodic feature matching and dynamic index generation mechanisms, the scientific validity of contour offset calculations under different calibration strategies is ensured. For example, the method of weighted summation based on the ratio of the spatial offset coefficient to the range of the fluctuation factor, using global variance, can effectively suppress noise interference and improve the stability and accuracy of contour offset calculations.

[0045] The baseline positioning module extracts morphological patterns from a standard weld point database and generates dynamic reference values ​​by combining cosine similarity and differences in the number of contour peaks, thus achieving adaptive derivation of the optimal positioning threshold. This threshold matching method based on dynamic reference values ​​can adapt to changes in the current contour offset in real time, avoiding the limitations of traditional fixed template matching methods and significantly improving the matching accuracy of the positioning threshold and the dynamic adaptability of the system.

[0046] The decision output module intelligently integrates weld point positioning decision schemes by dividing the main offset interval into secondary offset intervals, fusing convergence rate and diffusion rate, and combining spatial weights to generate adjustment parameters. This multi-dimensional analysis method based on spatial offset trends can comprehensively reflect the overall offset characteristics of the weld point edge, providing accurate quantitative basis for welding process adjustment and facilitating closed-loop control of welding quality.

[0047] This system, through collaborative innovation across multiple modules, constructs a complete technical closed loop encompassing image acquisition, feature analysis, rule establishment, dynamic calibration, and positioning decision-making, achieving high-precision identification and positioning of the edge features of condenser tube weld points. This system not only effectively addresses the challenges of feature diversity in complex welding environments but also enhances its robustness across multiple scenarios through a dynamic adaptive mechanism. It provides an advanced technical solution for the efficient inspection of condenser tube welding quality on automated production lines, demonstrating significant engineering application value and economic benefits. Attached Figure Description

[0048] Figure 1 This is a schematic diagram illustrating the working principle of the visual recognition and positioning system for the edge of the condenser tube weld point described in this invention.

[0049] Figure 2 Design diagram for the implementation of the edge feature module;

[0050] Figure 3 Design diagrams generated for edge feature vectors;

[0051] Figure 4 This is a design diagram illustrating the implementation of the correlation analysis module. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figure 1 The present invention relates to a visual recognition and positioning system for the edge of a condenser tube weld joint.

[0054] The system includes an image acquisition module, an edge feature module, a correlation analysis module, a morphology optimization module, a baseline positioning module, and a decision output module. These modules work together to achieve visual recognition and positioning of the weld points on the condenser tubes. The specific steps are as follows:

[0055] Image acquisition module: Acquires grayscale image data of the weld seam area of ​​the condenser tube from multiple angles using equipment such as industrial cameras. Based on the weld point positioning requirements, a contour scanning interval corresponding to the weld point position is set in the image. This interval limits the area range of subsequent edge feature analysis to ensure that the image part related to the weld point is focused.

[0056] Edge Feature Module: Within the contour scanning range, multiple contour nodes are divided, and gradient feature analysis is performed on the image data of each node. Gradient feature analysis identifies the intensity and direction of edges by calculating the gradient values ​​and directions of image pixels, generating an edge feature vector corresponding to each contour node. This vector contains edge information at the node and is used for subsequent localization analysis.

[0057] The correlation analysis module extracts key contour indicators, such as contour length, curvature, and direction, from edge feature vectors and establishes positioning judgment rules associated with each contour node. These rules define the conditions under which a node belongs to the edge of a weld point. Simultaneously, it obtains the geometric correction parameters corresponding to the rules to correct contour deviations caused by image acquisition or calculation errors.

[0058] The morphology optimization module identifies spatial correlation patterns in geometric correction parameters, such as the spatial distribution and mutual influence of parameters. Based on these correlation patterns, it dynamically calibrates key contour indicators, improving the accuracy of contour features by adjusting indicator weights or thresholds. Simultaneously, it calculates the contour offset of each contour node under different calibration strategies, quantifying the changes in contour position before and after calibration.

[0059] The baseline positioning module derives the optimal positioning threshold based on the contour offset, which serves as the baseline for determining whether the contour belongs to the weld point edge. By comparing the current contour feature value with the optimal positioning threshold, a contour deviation sequence is generated, recording the difference between the feature value and the threshold at each node, reflecting the degree of deviation between the contour and the weld point edge.

[0060] Decision output module: Analyzes the contour deviation sequence and analyzes the spatial offset trend of contour nodes, such as the overall offset direction and local offset magnitude. The contour deviation sequence is integrated into a weld point positioning decision scheme, which includes the specific location information and positioning reliability of the weld point, providing accurate positioning guidance for subsequent welding processes.

[0061] Example 1: See Figure 2 This embodiment involves the specific implementation of the edge feature module, including the construction of the solder joint feature library, similar contour matching of geometric parameter vectors, contour clustering, and morphological focal point extraction. The specific implementation method is as follows:

[0062] During system operation, the edge feature module first initiates the weld joint feature library construction process. This module uses historical grayscale image data of the condenser pipe weld area acquired by the image acquisition module to annotate the weld joint edge regions in the images, extract gradient features from each annotated region, and map these gradient features into geometric parameter vectors. These geometric parameter vectors contain multi-dimensional parameters such as contour position coordinates, contour length, contour curvature, and gradient direction corresponding to the image data. These parameters are calculated using image processing algorithms. For example, the contour position coordinates are obtained by statistically analyzing the average value after determining the contour pixel coordinates using an edge detection algorithm (such as the Canny operator). The contour curvature is calculated by fitting a quadratic curve to the coordinates of adjacent contour points, and the gradient direction is determined by calculating the gradient vector angle of the pixel using the Sobel operator.

[0063] When constructing the weld point feature library, the system associates and stores the labeled image data with the corresponding geometric parameter vectors to form a structured feature library. Each entry in the feature library corresponds to a typical weld point edge feature sample, containing the original image data and its mapped geometric parameter vector, providing a data foundation for subsequent similar contour matching.

[0064] After constructing the weld point feature library, the edge feature module performs similar contour matching on the geometric parameter vectors. During the matching process, the system uses a similarity calculation algorithm (such as the Euclidean distance algorithm or the cosine similarity algorithm) to calculate the similarity value between the current geometric parameter vector to be processed and the existing vectors in the feature library. Taking the Euclidean distance algorithm as an example, for two n-dimensional vectors A (a1,a2,...,an) and B (b1,b2,...,bn), the Euclidean distance is calculated as √[(a1-b1)²+(a2-b2)²+...+(an-bn)²]. The smaller the distance value, the higher the vector similarity. The system groups geometric parameter vectors with similarity values ​​higher than a preset matching threshold into the same group, forming preliminary contour clustering candidate groups.

[0065] When dividing the geometric parameter vectors into contour clusters, the system further extracts spatial attributes and gradient intensity parameters from the vectors. Spatial attributes include the coordinate distribution range of the contour in the image, the contour orientation (e.g., horizontal, vertical, tilt), etc. Gradient intensity represents the edge sharpness through the magnitude of the pixel gradient. Based on spatial attributes and gradient intensity, the system calculates curvature distribution, contour polarity, and morphological density parameters: curvature distribution reflects the curvature change at each point of the contour, obtained by calculating the contour curvature point by point and statistically analyzing the distribution frequency; contour polarity is defined as the direction of the contour edge (e.g., the transition direction from light to dark or from dark to light), determined by comparing the difference in grayscale values ​​of pixels on both sides of the contour; morphological density parameter is the number of contour line segments per unit image area, calculated by statistically analyzing the ratio of the number of contour pixels to the area of ​​the corresponding image region.

[0066] The system generates edge feature labels for each geometric parameter vector based on the aforementioned parameters, such as "high curvature-vertical polarity-low density edge" and "low curvature-horizontal polarity-high density edge". The labels are multi-dimensional, with each dimension corresponding to a key parameter's feature description, facilitating subsequent feature matching and clustering analysis. After generating the labels, the system associates and stores the edge feature labels with the geometric parameter vectors and filters cluster members by calculating the spatial similarity between the feature labels. Spatial similarity calculation considers the matching degree of each dimension of the labels. For example, for the "curvature distribution" dimension, an interval matching method is used: if the curvature distributions of two vectors both belong to the "high curvature" interval (above a preset curvature threshold), the similarity for that dimension is counted as 1; otherwise, it is counted as 0. For the "contour polarity" dimension, if the polarities are the same, it is counted as 1; otherwise, it is counted as 0. For the "morphological density parameter" dimension, the normalized absolute value of the difference is used for calculation; the smaller the difference, the higher the similarity. The system weights and sums the similarities of each dimension to obtain the comprehensive spatial similarity value between feature labels. When this value is higher than the preset contour threshold, the corresponding geometric parameter vector is selected to enter the same contour cluster group.

[0067] After contour clusters are formed, the system extracts morphological focal points from the image data of each cluster. The extraction of morphological focal points is based on the statistical characteristics of the geometric parameter vectors within the clusters. The specific process is as follows: First, the mean of the geometric parameters of all vectors within the cluster is calculated to obtain the central vector of the cluster, which represents the typical features of that type of contour. Then, in the image data corresponding to the cluster, the contour region that is closest to the features of the central vector is searched, and the geometric center or the point with the most significant feature of this region is set as the morphological focal point. For example, if the mean contour curvature of the central vector is high, then a region in the image with a curvature value close to this mean and consistent contour polarity is searched, and the centroid of this region is taken as the morphological focal point.

[0068] Once the morphological focal point is determined, the system sets it as a contour node within the contour scanning range. The distribution of contour nodes needs to cover the main feature areas of the solder joint edge. For different contour clusters, the number and position of the morphological focal points are dynamically adjusted according to the feature distribution of the clusters. For example, for clusters with large feature differences, multiple morphological focal points may be set to capture different edge features; for clusters with relatively consistent features, only one morphological focal point can be set as a representative.

[0069] Throughout the implementation process, the edge feature module effectively manages historical solder joint edge features by constructing and maintaining a solder joint feature library; it classifies massive geometric parameter vectors according to feature similarity through similar contour matching and contour clustering, reducing the complexity of subsequent feature analysis; and it ensures that contour nodes accurately reflect the key features of various types of solder joint edges through the extraction of morphological focal points and the setting of contour nodes, providing a solid foundation for subsequent gradient feature parsing, edge feature vector generation, and the establishment of positioning and determination rules.

[0070] This embodiment achieves systematic analysis and node division of the edge features of condenser tube weld joints through a multi-step, multi-dimensional feature processing flow. This ensures that the edge feature module can efficiently and accurately extract key information from the weld joint edges, laying the foundation for improving the accuracy of the entire visual recognition and positioning system. During implementation, each step is closely integrated. From data acquisition and feature extraction to cluster analysis and node setting, all are based on the fundamental principles of image processing and pattern recognition. Automated processing is achieved through parameter calculation and logical judgment, requiring no manual intervention and improving the system's stability and reliability.

[0071] Example 2: See Figure 3 and Figure 4 This embodiment relates to the specific implementation of the module for generating edge feature vectors corresponding to contour nodes and performing correlation analysis, including gradient fluctuation coefficient calculation, node type labeling, edge feature vector generation rules, and the establishment and expansion of positioning determination rules. The specific implementation method is as follows:

[0072] When generating the edge feature vectors corresponding to contour nodes, the system first determines the spatial coordinates of each contour node in the image within the contour scanning range. A preset window (e.g., a 3×3 pixel or 5×5 pixel rectangular window) is set centered on this node; the size of the window can be dynamically adjusted according to the image resolution and the detail level of the solder joint edges. The system acquires the grayscale data of all pixels within this window and calculates the gradient value and gradient direction of each pixel using a gradient calculation algorithm (such as the Sobel operator or the Prewitt operator), forming gradient fluctuation data within the window.

[0073] The gradient fluctuation coefficient is calculated based on the statistical characteristics of gradient values ​​within a window. The system first calculates the mean and variance of the gradient values ​​for all pixels within the window. The mean reflects the overall strength of the gradient, while the variance reflects the dispersion of the gradient values. The gradient fluctuation coefficient is defined as the ratio of the variance to the mean (or other composite statistics based on both), used to quantify the severity of gradient changes within the window. The range of this coefficient is pre-defined based on the differences in image noise levels and solder joint edge characteristics, for example, between 0 and 1, or mapped to a specific interval through normalization.

[0074] When the gradient fluctuation coefficient exceeds the first gradient threshold, the system determines the contour node to be a strong contour node. Strong contour nodes are characterized by significant changes in pixel grayscale values ​​within their region and clearly discernible edge features, such as sharp corners of solder joints or nodes at high-contrast boundaries. For such nodes, the system directly extracts image data (such as grayscale matrices or gradient vectors) within a preset window to form an edge feature vector. The dimension of the edge feature vector is related to the window size and feature type. For example, for a 3×3 pixel window, if the gradient value and direction of each pixel are extracted, the vector dimension is 3×3×2=18 dimensions, containing detailed gradient distribution information for that region.

[0075] When the gradient fluctuation coefficient is below the first gradient threshold, the node is marked as a steady-state node. Steady-state nodes are typically located in flat areas at the solder joint edges or in areas with high image noise; their own gradient changes are not significant, making it difficult to directly reflect edge features. In this case, the system analyzes the image data of the node's neighboring nodes (such as four or eight neighboring nodes). Through gradient superposition, the gradient values ​​of the steady-state node and its neighboring nodes are accumulated (e.g., pixel-by-pixel addition or weighted summation) to enhance the edge signal in that region. The superimposed gradient data is then normalized (e.g., subtracting the mean or dividing by the standard deviation) to reconstruct an edge feature vector. By fusing information from neighboring nodes, the edge features of the steady-state node are supplemented and strengthened, avoiding feature loss due to weak signals from a single node.

[0076] The processing flow of the correlation analysis module begins with the deconstruction of the edge feature vector. The system separates the main contour proportion, secondary contour proportion, and morphological fluctuation parameters from the edge feature vector. The main contour proportion represents the proportion of the dominant edge component in the vector, calculated by identifying the dominant gradient direction or curvature feature in the vector; the secondary contour proportion is the proportion of the minor edge component, reflecting the complexity or interference factors of the edge; the morphological fluctuation parameter is obtained by statistically analyzing the frequency of changes in gradient direction or curvature in the vector, characterizing the stability of the edge morphology.

[0077] Based on the above parameters, the system generates positioning rules for each contour node. These rules are expressed as logical expressions, such as "main contour percentage > 60% and morphological fluctuation parameter < preset threshold" or "secondary contour percentage < 20% and gradient fluctuation coefficient > second gradient threshold." These rules are set through machine learning algorithms (such as decision tree training) or human experience, aiming to distinguish between edge nodes and non-edge nodes of the solder joint. The thresholds in the rules are pre-calibrated based on the statistical characteristics of historical solder joint data; for example, the threshold for the main contour percentage is determined to be 60% by analyzing the edge feature distribution of a standard solder joint image.

[0078] If the number of contour nodes covered by the current positioning rule is less than a preset positioning threshold (e.g., 70% of the total number of nodes), the coverage of the system's judgment rule is insufficient and needs to be expanded. In this case, the system traverses the edge feature vectors of adjacent contour nodes, extracting geometric indicators from these vectors that are not included in the current rule, such as contour angle (the angle between adjacent contour segments), local contrast (the grayscale contrast of the region where the node is located), and contour continuity (the degree of continuity of edge pixels). These indicators are calculated using image processing algorithms. For example, the contour angle is calculated by the vector dot product to determine the directional angle between adjacent contour segments, and the local contrast is measured by the difference between the maximum and minimum grayscale values ​​in the node's neighborhood.

[0079] The system incorporates the extracted geometric indicators into the current positioning judgment rule, forming an expanded rule. For example, if the original rule is "main contour percentage > 60%", and the edge feature vectors of adjacent nodes generally contain the feature of "contour angle < 45°" which is not included in the original rule, then the new rule expands to "main contour percentage > 60% or contour angle < 45°". The expanded rule needs to re-verify the number of covered nodes until a preset positioning threshold is met or the maximum number of expansions is reached. Through this iterative expansion mechanism, the positioning judgment rule can adapt to changes in edge features of different solder joint shapes, improving the robustness of the system.

[0080] Throughout the implementation process, the mechanism for generating edge feature vectors differentiates between strong contour nodes and steady-state nodes, enabling differentiated processing of regions with varying feature intensities: strong contour nodes directly extract features to preserve details, while steady-state nodes enhance feature expression through neighborhood information fusion, ensuring that the edge feature vectors comprehensively reflect the true state of the solder joint edges. The correlation analysis module establishes a logical mapping from edge feature vectors to location determination through parameter separation, rule generation, and dynamic expansion. This allows the system to adaptively adjust determination conditions based on real-time acquired image data, improving the accuracy and reliability of solder joint edge recognition.

[0081] Each step in this embodiment is based on the fundamental principles of image processing and statistical analysis methods, achieving automated processing through operations such as gradient calculation, parameter extraction, and logical judgment. From the calculation of the gradient fluctuation coefficient to the expansion of the positioning determination rules, each step has clear physical meaning and algorithmic support, ensuring that the system can operate stably under different working conditions and providing reliable feature inputs for subsequent geometric correction, morphological optimization, and benchmark positioning.

[0082] Example 3: This example involves the specific implementation of the morphology optimization module, including the acquisition of spatial factors and fluctuation factors, the construction of the focusing weight matrix, and the calculation of the contour offset. The specific implementation method is as follows:

[0083] The morphology optimization module first extracts key features of the spatial correlation pattern from the geometric correction parameters, namely the spatial factor of morphology switching frequency and the fluctuation factor of gradient jump amplitude. The spatial factor reflects the frequency of change of contour morphology in spatial distribution, such as the number of changes in contour curvature or direction per unit length, and is calculated by statistically analyzing the feature differences between adjacent contour nodes within the contour scanning interval. The fluctuation factor characterizes the degree of abrupt change in gradient intensity, such as the maximum difference in gradient values ​​between adjacent pixels, and is obtained by analyzing gradient data in edge feature vectors.

[0084] When constructing the focusing weight matrix, the system identifies the morphological periodic characteristics of the spatial factors. Morphological periodicity refers to the periodic fluctuation pattern exhibited by the spatial factors as their contour position changes, such as the regular undulations at the edge of a weld joint caused by the welding process. The system analyzes the sequence data of the spatial factors using Fourier transform or sliding window statistical methods to extract the period length corresponding to their dominant frequency components. If the current periodic characteristic perfectly matches the preset contour period (such as the period value of a typical weld joint edge in historical data) (i.e., the period length error is within the preset range), then the spatial factor is set as the starting index of the focusing weight matrix, serving as the benchmark for constructing the matrix dimensions.

[0085] The system calculates the correlation between the spatial factor and the volatility factor. The correlation degree is measured using a mutual information algorithm or the Pearson correlation coefficient to assess their linear correlation, expressed by the following formula:

[0086]

[0087] in, For correlation matching degree, Let be the spatial factor value of the i-th contour node. The mean of the spatial factors. For the first The volatility factor value of each contour node. This represents the mean of the volatility factor. Total number of contour nodes. Association matching degree. The range of values ​​is The larger the absolute value, the stronger the correlation between the two factors. The system generates intermediate and terminal indices of the focusing weight matrix in descending order of correlation matching degree, ensuring that the order of the matrix indices reflects the strength of the synergistic effect between the factors.

[0088] After index generation is complete, the system performs path backtracking on the terminating index. During path backtracking, the system checks whether the correlation matching degree corresponding to the terminating index is lower than a preset matching threshold (e.g., 0.3). If it is lower than the threshold, it means that the factor combination corresponding to that index has a small impact on contour calibration, and its weight distribution in the matrix needs to be adjusted; if it is higher than the threshold, the original index order is retained. Finally, the system outputs the backtracked and verified index sequence as the final distribution of the focus weight matrix, where each element in the matrix... Representing the The spatial factor index and the first The weights of each volatility factor index are normalized (e.g., using the Softmax function) to ensure their sum is 1, as shown in the formula:

[0089]

[0090] in, Spatial factor index With volatility factor index The degree of correlation matching, This represents the total number of spatial factor indices. This represents the total number of volatility factor indices.

[0091] When calculating the contour offset of each contour node under different calibration strategies, the system first calculates the mean of the spatial factors and the range of the fluctuation factors corresponding to each termination index in the focus weight matrix. (Mean of spatial factors) The range of the volatility factor is the arithmetic mean of all spatial factor values ​​under this index, reflecting the typical characteristics of this factor range; This is the difference between the maximum and minimum values ​​of the volatility factor under this index, reflecting the fluctuation range of the factor. The system further calculates the global variance of all indexed factors. The global variance is calculated by combining all spatial factor and volatility factor data, using the following formula:

[0092]

[0093] in, For the first Individual factor data (spatial factors or volatility factors). The total mean of the factor data. This represents the total number of factor data points. Global variance characterizes the overall dispersion of the factor data and is used to standardize subsequent calculations.

[0094] For a single terminating index, the system calculates the difference between its spatial factor mean and the spatial factor mean of its adjacent indices. Divide the difference by the global variance Obtain spatial offset coefficient The formula is:

[0095] The spatial offset coefficient reflects the relative variation of the spatial factor between adjacent indices. Simultaneously, the system calculates the ratio of the volatility factor range to the global variance. The fluctuation offset coefficient is obtained. Finally, the spatial offset coefficients... With fluctuation offset coefficient According to preset weights and Weighted summation yields the contour offset for that index. The formula is:

[0096]

[0097] Among them, weight Pre-set the influence of spatial factor and fluctuation factor on solder joint positioning (e.g.) This indicates that spatial factors have a greater impact.

[0098] During implementation, the morphology optimization module decouples spatial and fluctuation factors, transforming complex geometric correction parameters into quantifiable factor combinations. It then uses a focused weight matrix to dynamically model the contribution of different factors. Periodic matching and correlation analysis during matrix construction ensure that the matrix structure matches the actual feature distribution of the weld point edges, while an index backtracking mechanism avoids interference from low-correlation factors. The calculation of contour offsets introduces global variance as a standardization factor, eliminating the influence of differences in factor dimensions and enabling cross-regional and cross-scale comparisons of offsets, thus improving the universality of the calibration strategy.

[0099] Each step in this embodiment closely revolves around the analysis and optimization of spatial correlation patterns and calibration strategies. From factor extraction and matrix construction to offset calculation, all steps are based on statistical analysis and linear algebra principles, using mathematical modeling to achieve dynamic calibration of contour features. The physical meaning of each character in the formula is clear, corresponding to feature statistics, correlation indicators, and weight parameters, ensuring that the calculation process is traceable and verifiable. The entire process requires no manual intervention, using automated algorithms to adaptively optimize the edge morphology of the weld points, providing high-precision contour offset data for the baseline positioning module, thereby ensuring the positioning accuracy of the entire visual recognition and positioning system.

[0100] Example 4: This example involves the specific implementation of deriving the optimal positioning threshold in the reference positioning module, including matching the standard solder joint database, calculating dynamic reference values, and the threshold matching process. A detailed explanation with specific examples follows:

[0101] During system operation, the reference positioning module first obtains the contour offset data of each contour node from the shape optimization module. The contour offset reflects the difference between the calibrated contour position and the ideal position, and is a key indicator for measuring the positioning accuracy of the weld point edge. Taking the weld point detection of a certain type of condenser tube as an example, assuming that there are 100 contour nodes in the contour scanning range of the currently acquired weld image, each node obtains a corresponding contour offset after shape optimization, forming a two-dimensional data sequence containing position coordinates and offset values.

[0102] After the reference positioning module is started, it accesses the standard weld point database. This database stores the contour feature data of various types of condenser tube weld points calibrated in the past. Each standard weld point record includes a contour offset sequence, a morphological pattern description (such as "straight edge", "arc edge", "multi-inflection edge", etc.), and the corresponding preset positioning threshold. The records in the database are stored in categories according to weld point type, specifications, welding process, etc., for easy and fast retrieval. For example, for weld points with "straight edges", the database stores the morphological patterns and thresholds corresponding to different offset ranges; for weld points with "arc edges", they are grouped and stored according to the radius of curvature.

[0103] When extracting the morphological pattern closest to the current contour offset, the system compares the current offset sequence with standard morphological patterns in the database using a pattern matching algorithm (such as Dynamic Time Warping (DTW) or Longest Common Subsequence (LCS)). Taking DTW as an example, assuming the current offset sequence is... A certain standard morphological pattern sequence is The algorithm finds the optimal matching path by calculating the time-warped distance between the two elements, minimizing the cumulative distance. The smaller the distance value, the more similar the morphological patterns of the two elements are.

[0104] In the example, if the current offset sequence, after matching, has the smallest distance value to the "straight edge - small offset" pattern in the standard solder joint database, then this pattern is determined to be the closest morphological pattern. The system extracts the contour offset data and spatial distribution characteristics of this pattern. For example, in the standard pattern, the mean offset of the contour nodes is 0.2 pixels, the standard deviation is 0.05 pixels, and the offset direction is concentrated in the positive X-axis direction, forming an approximately straight line distribution trend.

[0105] The system calculates the cosine similarity between the current contour offset and the standard morphological pattern in spatial distribution, using this as the first dynamic reference value. Cosine similarity measures the similarity of spatial distribution by comparing the directional differences between two vectors. The current offset sequence and the standard pattern sequence are then converted into vectors. and Calculate the ratio of the dot product of the two to the product of their moduli, with a range of values ​​of [value missing]. The closer the value is to 1, the more consistent the spatial distribution. In the example, if the calculated cosine similarity is 0.85, it means that the spatial distribution of the current offset is highly similar to the standard pattern.

[0106] The second dynamic reference value is the difference between the current contour offset and the standard morphological pattern in the number of contour peaks. Contour peaks refer to local maxima or minima in the offset sequence, reflecting the undulating characteristics of the contour. The system traverses the offset sequence, detecting the offset change trend of adjacent nodes. When the offset of a node is greater than that of its immediate neighbors, it is determined to be a local maximum (peak); if it is less, it is determined to be a local minimum (valley). For example, if the current offset sequence detects 3 peaks, while the standard pattern contains 2 peaks, the difference is 1, which is used as the second dynamic reference value.

[0107] Based on the first and second dynamic reference values, the system performs a linear combination to match a preset positioning threshold table. The preset positioning threshold table is a two-dimensional lookup table. The row dimension represents the range of the first dynamic reference value (e.g., 0.8-0.9, 0.9-1.0, etc.), and the column dimension represents the second dynamic reference value (e.g., 0, 1, 2, etc.). Each cell in the table stores the corresponding optimal positioning threshold. In the example, the first dynamic reference value 0.85 belongs to the "0.8-0.9" range, and the second dynamic reference value 1 corresponds to the column dimension. Looking up the table, the optimal positioning threshold for this combination is found to be 0.3 pixels.

[0108] The physical meaning of the positioning threshold is the critical offset that distinguishes whether a contour belongs to the edge of a solder joint. When the offset of the current contour node is less than or equal to the threshold, it is determined to be a valid solder joint edge; if it exceeds the threshold, it is determined to be a non-edge or an area that needs correction. For example, in the example, the offset of a contour node is 0.25 pixels, which is less than the threshold of 0.3 pixels, and it is determined to be a solder joint edge node; another node has an offset of 0.35 pixels, which exceeds the threshold, and it is determined to be a non-edge node, requiring further analysis to determine whether it is caused by welding defects or image noise.

[0109] In practical applications, the standard solder joint database needs to be updated regularly to incorporate feature data of new condenser tube models or solder joints produced using new processes. During the update process, the system records new solder joint contour offset sequences and morphological patterns into the database through manual annotation or automatic calibration, and adjusts the values ​​in the preset positioning threshold table based on the historical threshold usage effects. For example, when the actual positioning error of a certain type of solder joint consistently exceeds the theoretical error corresponding to the threshold, the system prompts engineers to adjust the threshold range for that type of solder joint to improve the adaptability of the threshold.

[0110] The core logic of the baseline positioning module lies in providing a dynamic baseline for the current inspection task through the statistical patterns of historical data. Its implementation process encompasses data matching, feature comparison, and threshold mapping, each based on the geometric and statistical characteristics of the weld point edge. By transforming abstract contour offsets into quantifiable reference values ​​and correlating them with preset thresholds, the system achieves standardized judgment of weld point edge positioning accuracy, providing a clear basis for subsequent generation of contour deviation sequences and positioning decision schemes.

[0111] Taking the "straight edge - small offset" pattern matching example, the entire process demonstrates a complete link from data acquisition and pattern recognition to threshold decision-making: First, the current offset data is obtained through image acquisition and morphological optimization. Then, the most similar historical pattern is found using database matching. Next, the threshold is accurately mapped using dual reference values, ultimately achieving effective classification of contour nodes. This process does not rely on real-time experimental data; it can quickly generate positioning thresholds suitable for different working conditions simply by reusing and statistically analyzing historical features, thus improving the system's versatility and real-time performance.

[0112] Example 5: This example involves the specific implementation of the decision output module parsing the contour deviation sequence and generating a weld point positioning decision scheme. A detailed description with a specific example follows:

[0113] Assuming a condenser pipe weld area contains 200 contour nodes within its contour scanning range, the reference positioning module has generated a contour deviation value for each node (i.e., the difference between the current contour feature value and the optimal positioning threshold), forming a contour deviation sequence. Each element in this sequence corresponds to the spatial position of a node (such as the coordinates along the weld length direction) and a deviation value. For example, the deviation values ​​for nodes 1 to 200 are +0.1mm, +0.15mm, +0.2mm, ..., -0.05mm, respectively (positive values ​​indicate that the offset direction is the preset positive direction, and negative values ​​indicate the opposite direction).

[0114] The decision output module first divides the main offset interval and the secondary offset interval according to the spatial offset direction of each node in the contour deviation sequence. The spatial offset direction is determined by the sign of the contour deviation value and the trend of spatial position change. For example, if the deviation values ​​of the first 150 nodes are all positive and the offset direction is along the positive X-axis (the preset welding path direction), then this area is defined as the main offset interval; among the last 50 nodes, the first 30 deviation values ​​are negative (the offset direction is the negative X-axis direction), and the last 20 are positive. Because the negative offset accounts for a higher proportion, this is defined as the secondary offset interval. The main offset interval usually corresponds to the main offset trend of the weld point edge, while the secondary offset interval reflects local anomalies or transition areas.

[0115] After dividing the system into intervals, the system extracts the convergence rate of the contour deviation within the main offset interval. The convergence rate is calculated by analyzing the trend of deviation values ​​with spatial location. For example, within the main offset interval, if the deviation values ​​from node 1 to node 100 gradually decrease from +0.3mm to +0.1mm, it indicates that the deviation is converging. The convergence rate can be represented by the slope of a linear regression fit; the larger the absolute value of the slope, the faster the convergence speed. Similarly, if the deviation values ​​from node 151 to node 180 within the secondary offset interval gradually increase from -0.1mm to -0.25mm, it indicates that the deviation is spreading. The spreading rate is measured by the absolute value of the fitted slope.

[0116] The spatial weight of a contour node is set based on its distance from the center of the weld joint. Nodes closer to the weld joint have a greater impact on positioning accuracy and therefore a higher weight. For example, by establishing a coordinate system with the weld joint center as the origin and along the weld direction, the node spatial weight can be set to... (in The coordinates are in mm, so the nodes in the center region have higher weights. In the example, within the main offset interval, node 50 (coordinate 0 mm, near the center of the solder joint) has a weight of 1, and node 100 (coordinate +5 mm) has a weight of 0.17; within the secondary offset interval, node 150 (coordinate +7.5 mm) has a weight of 0.11, and node 180 (coordinate +9 mm) has a weight of 0.1.

[0117] The system fuses the convergence rate of the main offset interval and the diffusion rate of the secondary offset interval according to spatial weights. During the fusion process, the rate values ​​within each interval are first weighted and averaged, with the weight being the spatial weight percentage of the corresponding node. For example, the convergence rate of the main offset interval is -0.002 mm / node (the negative sign indicates convergence), the sum of the weights of each node is 5, and the weighted average is -0.002 × (each node weight / 5) cumulative value; the diffusion rate of the secondary offset interval is +0.003 mm / node, the sum of the weights is 2, and the weighted average is +0.003 × (each node weight / 2) cumulative value. Finally, the two weighted results are added together to obtain the core component of the adjustment parameter, which reflects the overall strength of the overall offset trend.

[0118] When generating a weld joint positioning decision scheme, the system combines the statistical characteristics of the adjustment parameters and the contour deviation sequence. For example, if the weighted convergence rate of the main offset interval is -0.0015 mm / node and the weighted diffusion rate of the secondary offset interval is +0.001 mm / node, the combined adjustment parameter is -0.0005 mm / node, indicating that the overall offset trend is slightly converging, but there is local diffusion in the secondary offset interval. In this case, the decision scheme may suggest that the welding equipment be finely adjusted by 0.5 mm in the opposite direction of the main offset (negative X-axis direction), and that the welding parameters (such as current and speed) in the secondary offset interval be locally optimized to compensate for the diffusion trend.

[0119] In another example, if the contour deviation sequence exhibits random fluctuations without a clear primary offset direction (e.g., alternating positive and negative deviations), the system uses cluster analysis to divide the fluctuation region into multiple sub-intervals. Each sub-interval determines its primary and secondary attributes based on the proportion of offset directions. For example, in the first 100 nodes, 60% are positive offsets and 40% are negative offsets, defined as the primary offset interval (positive direction); in the last 100 nodes, 55% are negative offsets and 45% are positive offsets, defined as the secondary offset interval (negative direction). The fluctuation amplitude of the primary interval (standard deviation of 0.08 mm) and the fluctuation amplitude of the secondary interval (standard deviation of 0.12 mm) are extracted, and combined with spatial weights, adjustment parameters are calculated to indicate that the equipment needs to improve the stability of the welding path or adjust the image acquisition angle to reduce noise interference.

[0120] The core of the decision output module lies in transforming quantified deviation data into executable process instructions. Its implementation process includes:

[0121] Spatial trend analysis: Identify the overall trend and local anomalies of the weld joint edge by offset direction and interval division;

[0122] Rate quantization: The rate of change of the offset trend is evaluated by calculating the convergence / diffusion rate;

[0123] Weighted fusion: By allocating spatial weights, the impact of key areas on positioning accuracy is highlighted;

[0124] Solution generation: Based on the comprehensive analysis results, output a decision solution that includes location adjustment, parameter optimization, and other aspects.

[0125] Taking a typical scenario where the edge of a condenser tube weld joint exhibits "the main offset interval slowly converges along the positive X-axis, while the secondary offset interval rapidly expands along the negative X-axis" as an example, the decision scheme generated by the system through the above process may include:

[0126] Position adjustment: Move the initial positioning point of the welding torch 0.3mm in the negative X-axis direction to compensate for the main offset trend;

[0127] Path correction: Add a 0.2mm reverse compensation path at the weld position corresponding to the secondary offset interval;

[0128] Parameter adjustment: Reduce welding speed by 10% in the diffusion zone and increase molten pool cooling time to reduce offset.

[0129] This embodiment achieves hierarchical processing from data to decision through multi-dimensional analysis of the contour deviation sequence. Each step closely revolves around the spatial characteristics of the weld point edge, ensuring that the decision scheme reflects both the overall positioning requirements and local feature differences through logic such as trend recognition, rate calculation, and weight allocation. Without relying on preset experimental data or empirical formulas, an adaptive positioning scheme can be generated solely through dynamic analysis of real-time deviation data, improving the system's responsiveness to complex welding scenarios.

[0130] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0131] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A visual recognition and positioning system for the edge of a condenser tube weld joint, characterized in that, include: The image acquisition module is used to acquire grayscale image data of the weld seam area of ​​the condenser tube from multiple angles and to set the contour scanning interval corresponding to the weld point positioning. The edge feature module is used to divide the contour scanning range into multiple contour nodes, perform gradient feature parsing on the image data of each contour node, and generate the edge feature vector corresponding to the contour node. The correlation analysis module is used to extract key contour indicators from edge feature vectors, establish positioning judgment rules associated with contour nodes, define the conditions for a node to belong to the edge of a weld point, and obtain the geometric correction parameters corresponding to the rules. The morphology optimization module is used to identify spatial correlation patterns in geometric correction parameters. These spatial correlation patterns represent the distribution patterns and mutual influence relationships of the parameters in space. Based on the correlation patterns, the module dynamically calibrates key contour indicators and calculates the contour offset of each contour node under different calibration strategies. The baseline positioning module is used to derive the optimal positioning threshold based on the contour offset and generate a contour deviation sequence by comparing the current contour feature value with the optimal positioning threshold. The decision output module is used to parse the contour deviation sequence and integrate the contour deviation sequence into a weld point localization decision scheme based on the spatial offset trend of the contour nodes; it constructs a weld point feature library corresponding to the contour nodes, which contains image data and geometric parameter vectors of gradient feature mapping. Perform similar contour matching on the geometric parameter vectors, and divide the geometric parameter vectors into contour clusters based on the matching results; extract the morphological focal points of the image data from the contour clusters, and set the focal points as contour nodes; Based on the spatial properties and gradient intensity in the geometric parameter vector, the curvature distribution, contour polarity, and morphological density parameters are extracted, and edge feature labels are generated based on the above parameters. Edge feature labels are associated with geometric parameter vectors. By calculating the spatial similarity between feature labels, parameter vectors with similarity higher than a preset contour threshold are selected to form contour clusters. The main contour proportion, secondary contour proportion, and morphological fluctuation parameters are separated from the edge feature vector, and the positioning judgment rules of the contour nodes are generated based on the above parameters. If the number of contour nodes covered by the current positioning judgment rule is less than the preset positioning threshold, then the edge feature vectors of adjacent contour nodes are traversed, and geometric indicators not included in the judgment rules of adjacent nodes are added to the current rule. Extract the morphological pattern from the standard weld point database that is closest to the current contour offset, calculate the cosine similarity between the two in spatial distribution, and use it as the first dynamic reference value. The difference between the current contour offset and the standard shape pattern in the number of contour peaks is statistically analyzed, and the difference is used as a second dynamic reference value. Based on the linear combination of the first dynamic reference value and the second dynamic reference value, the optimal positioning threshold in the preset positioning threshold table is matched. Based on the spatial offset direction of each node in the contour deviation sequence, the main offset interval and the secondary offset interval are divided. The main offset interval corresponds to the main offset trend of the weld point edge, while the secondary offset interval reflects local anomalies or transition areas. The convergence rate of the contour deviation in the main offset interval and the diffusion rate of the contour deviation in the secondary offset interval are extracted, and the two are fused and calculated according to the spatial weight of the contour nodes to generate the adjustment parameters of the weld point positioning decision scheme.

2. The visual recognition and positioning system for the edge of the condenser tube weld point according to claim 1, characterized in that, The methods for generating edge feature vectors corresponding to contour nodes include: For each contour node, based on the spatial position of the node within the contour scanning range, the gradient fluctuation data of the node within a preset window is obtained, and the gradient fluctuation coefficient of the node is calculated. When the gradient fluctuation coefficient exceeds the first gradient threshold, the node is marked as a strong contour node, and its image data is extracted to form an edge feature vector; when the gradient fluctuation coefficient is lower than the first gradient threshold, the node is marked as a steady-state node, and the image data of the adjacent nodes of the node are gradient superimposed, and the superimposed data is reconstructed into an edge feature vector.

3. The visual recognition and positioning system for the edge of the condenser tube weld point according to claim 1, characterized in that, The implementation of the morphology optimization module includes: obtaining the spatial factor of the morphology switching frequency and the fluctuation factor of the gradient jump amplitude in the spatial correlation mode; Construct a focusing weight matrix associated with the spatial factor and the volatility factor, and determine the profile offset under different calibration strategies based on the probability distribution of each element in the matrix.

4. The visual recognition and positioning system for the edge of the condenser tube weld point according to claim 3, characterized in that, Constructing the focused weight matrix also includes: Identify the morphological periodic characteristics of the spatial factor. If the current periodic characteristics completely match the preset contour period, then set the spatial factor as the starting index of the focusing weight matrix. Calculate the correlation matching degree between spatial factors and volatility factors, and generate intermediate and terminal indices of the focusing weight matrix in descending order of matching degree; The path backtracking is performed on the terminating index. When the matching degree of the terminating index is lower than the preset matching threshold, it is output as the final distribution of the focusing weight matrix.

5. The visual recognition and positioning system for the edge of the condenser tube weld point according to claim 4, characterized in that, The methods for calculating the contour offset of each contour node under different calibration strategies include: The mean of the spatial factors and the range of the volatility factors for each terminating index in the statistical focus weight matrix are calculated, and the global variance of all index factors is calculated. The spatial offset coefficient is obtained by subtracting the spatial factor mean of a single terminating index from the spatial factor mean of its adjacent indices and dividing by the global variance. At the same time, the ratio of the fluctuation factor range to the global variance is calculated, and the two are weighted and summed to obtain the profile offset of the index.

Images