A Machine Vision-Based Method for Identifying Anomalies in the Hole Structure of Porous Copper Foil

By constructing a single-section skeleton point fitting quality quantification evaluation mechanism and an adaptive weighted three-dimensional Hough transform, the abnormalities of the porous copper foil hole structure are identified, solving the problem of inaccurate hole axis detection caused by defects in the inner wall of the hole, and realizing high-precision and robust hole shape and position anomaly identification.

CN123118634APending Publication Date: 2026-11-17SHAANXI WESTERN NEW MATERIAL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611614899.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing three-dimensional Hough transform methods, when detecting the hole structure of porous copper foil, suffer from surface defects such as burrs and pits on the inner wall of the hole, which prevent the equal-weighted voting mechanism from accurately identifying the hole axis, thus reducing the robustness and reliability of hole shape and position anomaly detection.

Method used

By constructing a quality quantification evaluation mechanism for single-section skeleton point fitting, outlier points in the contour are identified. The voting weights of the skeleton points are dynamically adjusted in combination with topological distribution characteristics. An adaptive weighted three-dimensional Hough transform is used to fit the central axis of the channel, thereby improving detection accuracy and robustness.

Benefits of technology

Under complex working conditions, it significantly improves the detection accuracy and robustness of channel axis tilt anomalies, suppresses the interference of low-quality skeleton points on axis fitting, and improves the accuracy of identifying anomalies in porous copper foil hole structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118634A_ABST
    Figure CN123118634A_ABST
Patent Text Reader

Abstract

This invention proposes a machine vision-based method for identifying anomalies in the perforation structure of porous copper foil, belonging to the field of image processing technology. To address the problem of some skeleton points deviating from the true central axis of the perforation due to surface defects such as burrs and pits on the inner wall of the perforation during copper foil inspection, this method includes: acquiring point cloud data of the porous copper foil and performing single-perforation segmentation to obtain independent three-dimensional point cloud data for each perforation; analyzing the radial distribution characteristics of the point cloud across the perforation cross-section to identify outliers; quantifying the fitting quality of each skeleton point in the perforation based on the topological distribution characteristics of the outliers in the depth direction of the perforation; dynamically adjusting the voting weights of each skeleton point in the three-dimensional Hough transform; fitting the central axis of the perforation using an adaptive weighted three-dimensional Hough transform; calculating the perforation verticality parameter based on the central axis of the perforation; and identifying and determining anomalies in the perforation axis tilt. This invention can stably improve the detection accuracy and robustness of perforation axis tilt anomalies under complex working conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method for identifying abnormal hole structures in porous copper foil based on machine vision. Background Technology

[0002] In the manufacturing process of lithium-ion batteries, porous copper foil serves as the negative electrode current collector, and the quality of its pore structure directly affects the battery's cycle life and rate performance. However, during the electrodeposition process of porous copper foil, factors such as uneven current density distribution, fluctuations in additive concentration, and electrolyte flow field disturbances can easily lead to morphological anomalies, specifically, the central axis of the pores deviates from the normal direction of the copper foil surface, forming an skewed pore structure. These anomalies severely weaken the bonding strength between the active material and the current collector, causing interfacial peeling during charging and discharging, and simultaneously triggering localized current density concentrations, accelerating battery failure.

[0003] Currently, the three-dimensional Hough transform method is commonly used to detect the geometric parameters of porous copper foil hole structures. Specifically, firstly, three-dimensional volume data of the hole is obtained through laser scanning. Then, the center coordinates of the hole are obtained from the cross-sectional image of each layer. The center coordinates of all depths together constitute the three-dimensional skeleton point set of the hole. Finally, using the voting mechanism of the three-dimensional Hough transform, an equal-weighted voting rule is applied to all skeleton points, and the number of votes for each unit in the parameter space is counted. The line corresponding to the unit with the highest number of votes is the central axis of the hole. Finally, based on the fitted hole axis, the key geometric parameters of the hole are quantitatively calculated, thereby realizing the identification of hole shape and position anomalies.

[0004] The core of existing three-dimensional Hough transform methods lies in applying equal-weighted voting to each skeleton point, determining the optimal fitted straight line for the channel axis through the peak response in the accumulator. However, in actual production inspection, surface defects such as burrs and pits inevitably exist on the inner wall of the channel. Furthermore, the uneven surface roughness causes some skeleton points to deviate from the true central axis of the channel. In this case, the equal-weighted voting mechanism assigns the same contribution weight to each skeleton point. False votes from outliers disperse the peak energy in the accumulator, diluting the response intensity of the true channel axis. Ultimately, the fitted axis direction deviates significantly, significantly reducing the robustness and reliability of channel anomaly detection.

[0005] Therefore, researching a high-precision intelligent identification method for the shape and position anomalies of porous copper foil channels has important engineering application value for ensuring battery manufacturing quality and improving yield. Summary of the Invention

[0006] To address the problems existing in the prior art, the purpose of this invention is to propose a machine vision-based method for identifying anomalies in the hole structure of porous copper foil. The core of this method lies in constructing a quantitative evaluation mechanism for the fitting quality of single-section skeleton points. First, outlier points in the contour are identified by analyzing the radial distribution characteristics of the point cloud of the hole cross section. Then, the fitting quality of the current skeleton point is quantitatively evaluated by combining the topological distribution characteristics of the outlier points. Based on this, the voting weights of each skeleton point in the three-dimensional Hough transform are dynamically adjusted to achieve optimal adaptation between the voting mechanism and the actual processing quality of the hole. This results in a stable improvement in the detection accuracy and robustness of hole axis tilt anomalies under complex working conditions.

[0007] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows: A machine vision-based method for identifying anomalies in the hole structure of porous copper foil includes the following steps: Point cloud data of porous copper foil is collected and segmented into individual holes to obtain independent three-dimensional point cloud data for each hole. Based on the independent 3D point cloud data of each channel, the macroscopic geometric features of the channel are used for rapid initial screening to identify suspicious and abnormal channels. The radial distribution features of the point cloud of each suspicious and abnormal channel at each depth section are analyzed to obtain the contour outliers caused by the inner wall defects. Based on the proportion of the topological distribution characteristics of the outlier points in the corresponding cross section, the fitting quality of each skeleton point in the channel is quantified, and the voting weight of each skeleton point in the three-dimensional Hough transform is dynamically adjusted. An adaptive weighted three-dimensional Hough transform is used to fit the central axis of the channel, and the verticality parameter of the channel is calculated based on the central axis to identify and determine the tilt anomaly of the channel axis.

[0008] As a further preferred option of the above scheme, independent three-dimensional point cloud data for each channel is obtained, including: The original point cloud of each channel formed by single-hole segmentation is subjected to statistical filtering, and the local copper foil reference plane around each channel is fitted by the random sampling consensus algorithm. The point cloud coordinate system of each channel is locally normalized so that the local copper foil reference plane coincides with the reference horizontal coordinate plane.

[0009] A further preferred approach is to perform rapid initial screening using the macroscopic geometric features of the channels to identify suspicious or abnormal channels, including: Equally spaced layered sampling was performed along the axial direction of the copper foil channel to obtain several single-section point cloud slices parallel to the reference horizontal coordinate plane; For each single-section point cloud slice, skeleton points are extracted to construct a three-dimensional skeleton point set for the channel; Macroscopic geometric feature values ​​for pore displacement assessment are constructed based on a 3D skeleton point set, and suspicious abnormal pores are screened out using these macroscopic geometric feature values; the expression for the macroscopic geometric feature values ​​is as follows: ; In the formula: tanh[] is the hyperbolic tangent function, m m represents the total number of cross-sectional samples for a single channel, and m-2 represents the number of intermediate skeleton points after excluding the skeleton points at the top and bottom surface orifices. intermediate skeleton point To the reference axis Euclidean distance, Indicates the reference axis of the upper and lower openings of the duct. axial direction of the ideal perpendicular channel The included angle, with values ​​ranging from [0, ... ], sin It is a sine function.

[0010] Based on the above technical solution, the further steps of extracting skeleton points for each single-section point cloud slice and constructing the three-dimensional skeleton point set of the channel include: For each single-section point cloud slice, the hole contour point cloud of the current section is circularly fitted in the reference horizontal coordinate plane to obtain the hole skeleton points of the current section. The skeleton points of each single-section point cloud slice along the axial direction of the channel are extracted, and the skeleton points of all sections constitute the three-dimensional skeleton point set of the channel.

[0011] As a preferred implementation method, the radial distribution characteristics of the point cloud at each depth section of each suspected anomalous channel are analyzed to obtain the contour outliers caused by the inner wall defects, specifically including: For the current cross-section of the duct contour point cloud set, the convex polygon boundary of the contour point cloud is extracted, and a dual reference system of inscribed and circumscribed circles is constructed based on the convex polygon boundary; the dual reference system of inscribed and circumscribed circles includes solving the minimum circumscribed circle using the rotating caliper algorithm and solving the maximum inscribed circle using the maximum empty circle algorithm. Based on the dual datum system of inscribed and circumscribed circles, the radial distribution characteristic values ​​of single-section contour points are established. Set an outlier detection threshold, and determine the outliers in the contour based on the comparison between the radial distribution feature value and the outlier detection threshold.

[0012] In some preferred embodiments, the fitting quality of each skeleton point in the channel is quantified, and the voting weights of each skeleton point in the three-dimensional Hough transform are dynamically adjusted, including: Based on contour outliers, an adaptive voting weight is constructed for single-section skeleton points; the calculation formula for the adaptive voting weight of the single-section skeleton points is as follows: ; In the formula: For adaptive voting weights; The basic voting weights for the current cross-section skeleton points. For the first i The number of all contour points within the cross section of a given skeleton point. This represents the number of outliers in the current cross-sectional profile. Adaptive voting weights are used to dynamically adjust the voting weights of each skeleton point in the three-dimensional Hough transform.

[0013] As a further preferred option, the specific process of fitting the central axis of the channel using adaptive weighted three-dimensional Hough transform includes: Each skeleton point votes on all parameter space units corresponding to the candidate channel axes passing through it, and the voting weight is positively correlated with the fitting reliability of the skeleton point. After accumulating the votes for all skeleton points, the straight line corresponding to the unit with the highest number of votes in the parameter space is determined as the final fitted channel center axis.

[0014] A further preferred approach is to identify and determine abnormal tilting of the duct axis, including: Using the line containing the normal vector of the copper foil reference plane as the reference direction for the ideal vertical channel, calculate the angle between the channel's central axis and the reference direction, which is then used as the channel's tilt angle. Based on the preset verticality judgment standard, the tilt angle of each channel is judged. For holes identified as having abnormal axis tilt, mark their positions on the porous copper foil and output the corresponding tilt angle value.

[0015] On the other hand, the present invention also discloses a machine vision-based system for identifying abnormal hole structures in porous copper foil. This system is used to implement the steps of the method described above, and includes: The data acquisition module is used to collect point cloud data of the porous copper foil; The segmentation module is used to segment the point cloud data of the porous copper foil into individual holes to obtain independent three-dimensional point cloud data for each hole. The contour outlier identification module, based on the independent 3D point cloud data of each channel, performs rapid initial screening through the macroscopic geometric features of the channel to identify suspicious and abnormal channels. It analyzes the radial distribution characteristics of the point cloud of each suspicious and abnormal channel at each depth section to obtain contour outliers caused by internal wall defects. The voting weight dynamic adjustment module quantifies the fitting quality of each skeleton point in the channel based on the proportion of the topological distribution characteristics of the contour outliers in the corresponding cross section, and dynamically adjusts the voting weight of each skeleton point in the three-dimensional Hough transform. The fitting module uses an adaptive weighted three-dimensional Hough transform to fit the central axis of the channel. The axis tilt anomaly identification module calculates the verticality parameters of the duct based on the center axis of the duct, and identifies and judges the axis tilt anomalies of the duct.

[0016] Compared with the prior art, the present invention can produce the following beneficial effects: This invention rapidly screens out suspicious abnormal channels that may exhibit significant axial bending or overall tilting trends by utilizing the macroscopic three-dimensional geometric features of the channel. For each suspicious abnormal channel, a dual-benchmark system of inscribed circles is constructed, integrating the overall roundness features of the cross-section with the local radial deviation features of individual points to accurately identify contour outliers caused by inner wall defects. Finally, based on the topological distribution characteristics of the outliers in the channel's depth direction, the overall fitting quality of each skeleton point in the channel is dynamically evaluated, and the voting weights of the skeleton points in the three-dimensional Hough transform are dynamically adjusted according to the overall fitting quality. This suppresses the interference of low-quality skeleton points on the axis fitting, thereby improving the overall axis fitting accuracy and robustness. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required in the embodiments will be briefly introduced below.

[0018] Figure 1 This is a flowchart of a machine vision-based method for identifying abnormal hole structures in porous copper foil according to the present invention. Figure 2 This is a schematic diagram of single-section sampling and extraction along the axial direction of the channel according to the present invention; Figure 3 This is a schematic diagram illustrating the construction of an inner and outer circumscribed circle dual-reference system for each single-section point cloud slice of a suspected abnormal channel according to the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in 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, and not all embodiments. For those skilled in the art, other embodiments can be obtained without creative effort.

[0020] Example 1: Reference Figures 1-3 This invention provides a machine vision-based method for identifying abnormal hole structures in porous copper foil, comprising the following steps: S101: Collect point cloud data of porous copper foil and perform single-hole segmentation to obtain independent three-dimensional point cloud data for each hole.

[0021] In this step, a high-resolution confocal laser scanning microscope is used to scan the porous copper foil sample. The high-resolution confocal laser scanning microscope has a lateral resolution better than 0.1 μm and a longitudinal resolution better than 0.5 μm, enabling it to clearly capture the fine three-dimensional structure of micron-level pores. Each pore is scanned individually with high-resolution 3D scanning, and the scanning range automatically adapts to the actual size of the pore, obtaining independent 3D point cloud data for each pore. Specifically, during scanning, the laser power and scanning speed are dynamically adjusted based on the point cloud signal feedback. The laser power is increased and the scanning speed is decreased in real time according to the signal intensity and density of the current cross-section point cloud. If the signal is weak, the laser power is increased and the scanning speed is decreased; if the signal is too strong, the laser power is decreased and the scanning speed is increased, ensuring that the data is neither overexposed nor underexposed.

[0022] Furthermore, statistical filtering is performed on the raw point cloud collected separately for each channel to remove outliers caused by sensor noise and surface dust. A random sampling consensus algorithm is used to fit the local copper foil reference plane around the channel. Specifically, a plane is initialized by randomly selecting three points from the raw point cloud through iteration. Then, the distance from all points to the plane is calculated. The inner and outer points are distinguished based on the distance from the points to the plane. The plane with the most inner points is selected as the estimate of the local copper foil reference plane through multiple iterations.

[0023] The point cloud coordinate system of the processed channel is locally normalized so that the local copper foil reference plane coincides with the reference horizontal coordinate plane. This reference horizontal coordinate plane is... Figure 2 The XOY coordinate plane is shown in the figure. The above treatment of a single channel eliminates the influence of the overall warping of the copper foil on the calculation of the verticality of the individual channel, and the accuracy is much higher than that of overall normalization.

[0024] S102: Based on the independent three-dimensional point cloud data of each channel, the channel macroscopic geometric features are used for rapid initial screening to identify suspicious abnormal channels. The radial distribution features of the point cloud of each suspicious abnormal channel at each depth section are analyzed to obtain the contour outliers caused by the inner wall defects. Traditional three-dimensional Hough transform uses equal-weighted voting for all skeleton points. When there are processing defects such as burrs or pits on the inner wall of the channel or data acquisition noise, these outlier skeleton points that deviate from the true axis will generate a large number of invalid votes, which seriously interferes with the peak positioning in the parameter space.

[0025] Therefore, the core improvement of this embodiment lies in first quickly screening out suspicious abnormal channels that may have shape and position anomalies through macroscopic geometric features, and then deeply analyzing the radial distribution features of the point cloud on the depth section of each suspicious abnormal channel to accurately identify contour outliers caused by inner wall defects, as follows: (I) For all the channels to be detected, the macroscopic geometric features of the channels are used for rapid initial screening to identify normal channels and suspicious abnormal channels. Normal channels are directly detected using conventional methods. Suspected abnormal channels that may have obvious axial bending or overall tilting trends are subjected to subsequent refined weight analysis and axial fitting. This greatly improves the overall detection efficiency while ensuring detection accuracy, and avoids high-computation refined processing of all channels.

[0026] like Figure 2 As shown, for each channel to be inspected, along the axial direction of the copper foil channel (i.e., Figure 2 (In the Z-axis direction) equal-interval layered sampling is performed from the upper surface aperture to the lower surface aperture of the copper foil to obtain... m A plane parallel to the reference horizontal coordinate plane (i.e.) Figure 2 A single-section point cloud slice in the XOY coordinate plane, denoted as ,in This is the cross-section of the upper surface orifice. The lower surface orifice section is used. The spacing ΔZ between adjacent sections along the axial direction of the channel is adaptively set according to the total depth of the channel and the resolution of the device. In this embodiment, ΔZ=1μm to ensure that the sampling section can completely cover the full depth of the channel and accurately capture the axial change characteristics of the channel axis. In practical applications, it can be finely adjusted and optimized according to the detection accuracy.

[0027] For each single-section point cloud slice ( i =1,2, , m In the XOY coordinate plane, the least squares circle fitting algorithm is used to perform circular fitting on the point cloud of the duct profile within the cross-section, and the coordinates of the fitted circle center of the cross-section are obtained. The center of this circle is the skeleton point of the channel at the current cross-section. Along the depth direction of the channel, a skeleton point of the corresponding cross-section is extracted every ΔZ interval. The skeleton points of all cross-sections together constitute the three-dimensional skeleton point set of the channel. .

[0028] (II) Based on the three-dimensional skeleton point set, construct the macroscopic geometric feature value for duct offset assessment, and use the macroscopic geometric feature value to screen out suspicious abnormal ducts.

[0029] The above surface orifice skeleton points With the lower surface orifice skeleton point Construct a reference axis straight line for the duct, using the endpoints as reference points. The straight line The theoretical connection between the upper and lower openings of the channel serves as the benchmark for macroscopic offset evaluation. Simultaneously, a macroscopic geometric eigenvalue M for channel offset evaluation is constructed using a 3D skeleton point set, expressed as follows: ; In the formula: tanh[] is the hyperbolic tangent function, used to normalize the calculation results. m This represents the total number of cross-sectional samples taken from a single channel. m -2 represents the number of intermediate skeleton points after excluding the skeleton points at the top and bottom surface orifices. intermediate skeleton point To the reference axis Euclidean distance, Indicates the reference axis of the upper and lower openings of the duct. axial direction of the ideal perpendicular channel The angle between the reference plane normal vector of the copper foil and the line containing the reference plane normal vector is [0, ...]. ], sin It is a sine function used to normalize the calculated angle.

[0030] In this formula, the first term The macroscopic offset of all intermediate skeleton points in the duct relative to the line connecting the upper and lower orifice references was quantified: the closer the value is to 1, the greater the overall offset of the intermediate skeleton points, and the higher the probability of an abnormal axial tilt in the duct; the closer the value is to 0, the more likely the intermediate skeleton points are distributed in an approximately straight line along the reference line, and the higher the probability that the duct is a normal duct; the second term The macroscopic tilt of the channel was quantified: the smaller the value, the more the reference axis coincides with the ideal vertical direction, and the channel has no overall tilt; the larger the value, the more serious the overall deviation of the channel from the vertical direction.

[0031] The macroscopic geometric feature value M integrates two dimensions of macroscopic geometric characteristics: the degree of curvature of the internal axis of the channel (radial offset) and the degree of overall tilt (axial deviation). The closer the value is to 1, the higher the probability that the channel has both severe central axis curvature and overall tilt, indicating an abnormal channel. The closer the value is to 0, the higher the probability that the channel axis is both close to a straight line and basically perpendicular to the copper foil surface, indicating a normal channel. This feature value provides accurate quantitative basis for rapid initial screening through the synergistic characterization of two dimensions.

[0032] The classification threshold T1 is set according to the requirements of detection efficiency and accuracy. In this embodiment, T1=0.3. In actual application, it can be adjusted according to the detection cycle and the strictness of quality control. The classification threshold T1 is used to judge normal channels and suspicious abnormal channels: when M≤T1, the channel is judged as a normal channel, and the conventional three-dimensional Hough transform is directly used to complete the axis fitting and detection; when M>T1, the channel is judged as a suspicious abnormal channel, and the subsequent single-section fine feature analysis and weight adaptive allocation process is entered.

[0033] (III) Analyze the radial distribution characteristics of the point cloud at each depth section for each suspicious abnormal channel to obtain the contour outliers caused by the inner wall defects.

[0034] Traditional single-section channel profile anomaly detection often uses least-squares circle fitting residuals as the criterion, which cannot distinguish between overall profile roundness deviations and discrete defects such as local burrs, pits, and protrusions, and lacks sensitivity to minute local defects. In this embodiment, by constructing a dual-benchmark system of inscribed and circumscribed circles, the overall roundness characteristics of the cross-section and the local radial deviation characteristics of a single point are integrated to accurately locate local outliers, providing a reliable data foundation for subsequent refined weight allocation and fitting of the channel axis.

[0035] like Figure 3 As shown, for the point cloud set of the duct contour of the current cross section, the convex polygon boundary of the contour point cloud is extracted by the convex hull algorithm. Then, the rotating caliper algorithm is used to rotate the tangent of the convex hull to find the minimum radius circle that can completely enclose all vertices of the convex hull, which is taken as the minimum circumcircle of the cross section. , radius is The maximum empty circle algorithm is used to find the circle with the largest radius that does not contain any contour points within the closed region enclosed by the contour point cloud. This circle is then used as the maximum inscribed circle of the cross section. , radius is Construct a dual-reference system of inner and outer circumscribed circles.

[0036] Based on the dual datum system of inscribed and circumscribed circles, establish the radial distribution characteristic value of single-section contour points. Its expression is as follows: ; In the formula: This is the difference between the minimum circumradius and the maximum incircle radius of the current single cross-section. Indicates the current section number j A contour point cloud to the inscribed circle radial distance, This indicates the distance from all contour points of the cross section to the inscribed circle. The average radial distance.

[0037] It directly reflects the overall roundness deviation and radial non-uniformity of the cross-sectional profile: the larger the value, the more significant the difference between the maximum and minimum radial boundaries of the cross-sectional profile, the more serious the overall deviation of the profile from a perfect circle, and the possible overall defects such as ellipticization and eccentricity; the smaller the value, the closer the cross-sectional profile is to a perfect circle, and the better the overall roundness. The degree of local deviation of the point relative to the overall contour was quantified: the greater the deviation, the more the point deviates from the average radial position of the cross section, and the higher the probability that it is an outlier point such as a local burr or pit; the smaller the deviation, the better the consistency between the point and the overall contour, and the higher the probability that it is a normal contour point.

[0038] In this embodiment, the inscribed circle is used as the radial distance reference because the inscribed circle reflects the minimum diameter of the channel, and its radial distance change directly affects the flow performance of the electrolyte. If the circumscribed circle is used as the reference, its radial distance change mainly reflects the outward expansion deformation of the contour. The physical meaning and calculation logic of the two are completely consistent, and they can be flexibly switched according to the actual detection focus.

[0039] radial distribution eigenvalues This method integrates two dimensions: the overall roundness deviation of the cross-sectional profile and the degree of local radial deviation at a single point. A value closer to 1 indicates that the cross-section at that point exhibits both significant overall roundness anomalies and severe deviation from the overall profile, making it a higher probability of being an outlier. A value closer to 0 indicates good overall roundness of the cross-section at that point and that the point is highly consistent with the overall profile, making it a higher probability of being a normal profile point. Furthermore, an outlier determination threshold T2 is set. Based on the comparison between the radial distribution feature value and the outlier determination threshold, outliers are identified. This outlier identification mechanism effectively solves the problem that the traditional single-circle fitting residual method cannot distinguish between overall deformation and local anomalies, significantly improving the accuracy and robustness of outlier identification. In this embodiment, to balance defect detection rate and false positive rate, the outlier determination threshold T2 = 0.4. When T2 ≤ T2, the point is determined to be a normal contour point; when When the threshold value is greater than T2, the point is identified as an outlier in the contour. In practical applications, the outlier identification threshold can be adjusted based on detection sensitivity and data quality.

[0040] S103: Based on the topological distribution characteristics of the outlier points in the longitudinal direction of the channel, quantify the fitting quality of each skeleton point in the channel and dynamically adjust the voting weight of each skeleton point in the three-dimensional Hough transform. When a cross-section of a channel contains numerous outliers such as burrs and pits, the fitting result of the skeleton point (i.e., the center of the circle) at that cross-section is more likely to deviate from the true axis. Traditional 3D Hough transform, which assigns equal weight to all skeleton points, severely interferes with peak localization in the parameter space. Therefore, in this embodiment, the topological distribution characteristics of the outliers are used to evaluate the fitting quality of the current skeleton point, and the voting weight of the current skeleton point in the 3D Hough transform is adaptively adjusted accordingly. This suppresses the interference of low-quality skeleton points on axis fitting, improving fitting accuracy and robustness.

[0041] (I) Based on the contour outlier points, construct the adaptive voting weights for single-section skeleton points; According to the above The adaptive voting weights for the single-section skeleton points constructed from the T2-identified contour outliers are calculated using the following formula: ; In the formula: For adaptive voting weights; The basic voting weights for the current cross-section skeleton points (fitted circle centers) Prior experience settings based on the geometric detection of the duct cross-section; this embodiment takes =1, For the first i The number of all contour points within the cross section of a given skeleton point. This represents the number of outliers in the current cross-sectional profile. / The proportion reflects the overall contour quality of the cross section and the reliability of the fit between the skeleton points.

[0042] (II) Adaptive voting weights are used to dynamically adjust the voting weights of each skeleton point in the three-dimensional Hough transform. In the formula for calculating the adaptive voting weight of a single-section skeleton point, when the proportion of outliers in the profile of a certain section is higher, it indicates that the profile deformation of that section is more severe, and the skeleton point obtained based on that section is more likely to deviate from the true center of the channel (i.e., the fitting quality is lower). Therefore, its voting weight is higher. The voting weight is correspondingly reduced to weaken its interference with axis fitting; when the proportion of outliers in a certain cross-sectional profile approaches 0, the voting weight... This approach aligns with the equal-weighted voting benchmark of traditional 3D Hough transform, ensuring that the contributions of high-quality skeleton points remain unaffected. The weight adjustment mechanism achieves precise matching between voting weights and cross-sectional profile quality, with extremely low computational complexity, thus avoiding any additional computational burden.

[0043] S104: The channel center axis is fitted using an adaptive weighted three-dimensional Hough transform, and the channel verticality parameter is calculated based on the channel center axis to identify and determine the channel axis tilt anomaly.

[0044] For identified normal channels, a conventional three-dimensional Hough transform (equal weight) is directly used to fit the central axis. For identified suspicious abnormal channels, the following adaptive weighted three-dimensional Hough transform is used for intelligent identification and judgment: (I) In this embodiment, the channel center axis is fitted by adaptive weighted three-dimensional Hough transform. Based on the conventional three-dimensional Hough transform, this method upgrades the original equal-weight voting mechanism of all skeleton points to an adaptive weighted voting mechanism based on the fitting quality of skeleton points. Specifically, each skeleton point will vote for all parameter space units corresponding to the candidate channel axis passing through it. The voting weight is positively correlated with the fitting reliability of the skeleton point. That is, the higher the fitting quality of the skeleton point, the greater its contribution to the final axis determination. Conversely, the voting contribution of low-quality skeleton points affected by burrs, pits or acquisition noise will be significantly weakened.

[0045] (II) After the voting accumulation of all skeleton points is completed, the straight line corresponding to the unit with the highest vote in the parameter space is determined as the final fitted channel center axis.

[0046] (III) The straight line containing the normal vector of the copper foil reference plane As the reference direction for the ideal vertical channel, the angle between the fitted channel center axis and the reference direction is calculated. This angle is the channel tilt angle, which serves as the core parameter for quantifying channel verticality. Based on the preset channel verticality qualification judgment standard, the tilt angle of each channel is judged to distinguish between qualified channels and channels with abnormal axis tilt. For channels judged as having abnormal axis tilt, their positions on the porous copper foil are marked and the corresponding tilt angle value is output, thus completing the intelligent identification and data output of abnormal channel axis tilt.

[0047] The technical effect of the above embodiment 1 of the present invention is as follows: by analyzing the radial distribution characteristics of the point cloud of the channel section, outlier points in the contour are identified, and the fitting quality of the current skeleton points is quantitatively evaluated by combining the topological distribution characteristics of the outlier points in the contour. Based on this, the voting weight of each skeleton point in the three-dimensional Hough transform is dynamically adjusted to achieve the optimal adaptation between the voting mechanism and the actual processing quality of the channel. This can stably improve the detection accuracy and robustness of the channel axis tilt anomaly under complex working conditions.

[0048] Example 2: This invention also provides a machine vision-based system for identifying anomalies in the perforation structure of porous copper foil. The system implements the steps of the methods described above. The system includes a data acquisition module, a segmentation module, a contour outlier identification module, a dynamic voting weight adjustment module, a fitting module, and an axis tilt anomaly identification module, wherein: The data acquisition module is used to collect point cloud data of the porous copper foil; The segmentation module is used to segment the point cloud data of the porous copper foil into individual holes to obtain independent three-dimensional point cloud data for each hole. The contour outlier identification module, based on the independent 3D point cloud data of each channel, performs rapid initial screening through the macroscopic geometric features of the channel to identify suspicious and abnormal channels. It analyzes the radial distribution characteristics of the point cloud of each suspicious and abnormal channel at each depth section to obtain contour outliers caused by internal wall defects. The voting weight dynamic adjustment module quantifies the fitting quality of each skeleton point in the channel based on the proportion of the topological distribution characteristics of the contour outliers in the corresponding cross section, and dynamically adjusts the voting weight of each skeleton point in the three-dimensional Hough transform. The fitting module uses an adaptive weighted three-dimensional Hough transform to fit the central axis of the channel. The axis tilt anomaly identification module calculates the verticality parameters of the duct based on the center axis of the duct, and identifies and judges the axis tilt anomalies of the duct.

[0049] It should be understood that the above detailed description of the embodiments of the present invention provided in the accompanying drawings is only for further illustration of the present invention and is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. Based on the embodiments of the present invention, any non-essential improvements and adjustments made by those skilled in the art according to the above description of the present invention are within the scope of protection of the present invention.

Claims

1. A method for identifying abnormal hole structures in porous copper foil based on machine vision, characterized in that, Includes the following steps: Point cloud data of porous copper foil is collected and segmented into individual holes to obtain independent three-dimensional point cloud data for each hole. Based on the independent 3D point cloud data of each channel, the macroscopic geometric features of the channel are used for rapid initial screening to identify suspicious and abnormal channels. The radial distribution features of the point cloud of each suspicious and abnormal channel at each depth section are analyzed to obtain the contour outliers caused by the inner wall defects. Based on the proportion of the topological distribution characteristics of the outlier points in the corresponding cross section, the fitting quality of each skeleton point in the channel is quantified, and the voting weight of each skeleton point in the three-dimensional Hough transform is dynamically adjusted. An adaptive weighted three-dimensional Hough transform is used to fit the central axis of the channel, and the verticality parameter of the channel is calculated based on the central axis to identify and determine the tilt anomaly of the channel axis.

2. The method for identifying abnormal hole structures in porous copper foil based on machine vision according to claim 1, characterized in that, Obtain independent 3D point cloud data for each channel, including: The original point cloud of each channel formed by single-hole segmentation is subjected to statistical filtering, and the local copper foil reference plane around each channel is fitted by the random sampling consensus algorithm. The point cloud coordinate system of each channel is locally normalized so that the local copper foil reference plane coincides with the reference horizontal coordinate plane.

3. The method for identifying abnormal hole structures in porous copper foil based on machine vision according to claim 2, characterized in that, Rapid initial screening using macroscopic geometric features of the channels identifies suspicious and abnormal channels, including: Equally spaced layered sampling was performed along the axial direction of the copper foil channel to obtain several single-section point cloud slices parallel to the reference horizontal coordinate plane; For each single-section point cloud slice, skeleton points are extracted to construct a three-dimensional skeleton point set for the channel; Macroscopic geometric feature values ​​for pore displacement assessment are constructed based on a 3D skeleton point set, and suspicious abnormal pores are screened out using these macroscopic geometric feature values; the expression for the macroscopic geometric feature values ​​is as follows: ; In the formula: tanh[] is the hyperbolic tangent function, m m is the total number of cross-sectional samples for a single channel, and m-2 is the number of intermediate skeleton points after excluding the skeleton points at the top and bottom surface orifices. intermediate skeleton point To the reference axis Euclidean distance, Indicates the reference axis of the upper and lower openings of the duct. axial direction of the ideal perpendicular channel The included angle, with values ​​ranging from [0, ... ], sin It is a sine function.

4. The method for identifying abnormal hole structures in porous copper foil based on machine vision according to claim 3, characterized in that, For each single-section point cloud slice, skeleton points are extracted to construct a 3D skeleton point set for the channel, including: For each single-section point cloud slice, the hole contour point cloud of the current section is circularly fitted in the reference horizontal coordinate plane to obtain the hole skeleton points of the current section. The skeleton points of each single-section point cloud slice along the axial direction of the channel are extracted, and the skeleton points of all sections constitute the three-dimensional skeleton point set of the channel.

5. The method for identifying abnormal hole structures in porous copper foil based on machine vision according to claim 4, characterized in that, Analyze the radial distribution characteristics of the point cloud at each depth section for each suspected anomalous channel to obtain contour outliers caused by internal wall defects, specifically including: For the current cross-section of the duct contour point cloud set, the convex polygon boundary of the contour point cloud is extracted, and a dual reference system of inscribed and circumscribed circles is constructed based on the convex polygon boundary; the dual reference system of inscribed and circumscribed circles includes the minimum circumscribed circle solved by the rotating caliper algorithm and the maximum inscribed circle solved by the maximum empty circle algorithm. Based on the dual datum system of inscribed and circumscribed circles, the radial distribution characteristic values ​​of single-section contour points are established. Set an outlier detection threshold, and determine the outliers in the contour based on the comparison between the radial distribution feature value and the outlier detection threshold.

6. The method for identifying abnormal hole structures in porous copper foil based on machine vision according to claim 5, characterized in that, Quantify the fitting quality of each skeleton point in the channel and dynamically adjust the voting weights of each skeleton point in the three-dimensional Hough transform, including: Based on contour outliers, an adaptive voting weight is constructed for single-section skeleton points; the calculation formula for the adaptive voting weight of the single-section skeleton points is as follows: ; In the formula: For adaptive voting weights; The basic voting weights for the current cross-section skeleton points. For the first i The number of all contour points within the cross section of a given skeleton point. This represents the number of outliers in the current cross-sectional profile. Adaptive voting weights are used to dynamically adjust the voting weights of each skeleton point in the three-dimensional Hough transform.

7. The method for identifying abnormal hole structures in porous copper foil based on machine vision according to claim 6, characterized in that, The specific process of fitting the central axis of the channel using adaptive weighted three-dimensional Hough transform includes: Each skeleton point votes for all parameter space units corresponding to all candidate channel axes passing through it; After accumulating the votes for all skeleton points, the straight line corresponding to the unit with the highest number of votes in the parameter space is determined as the final fitted channel center axis.

8. The method for identifying abnormal hole structures in porous copper foil based on machine vision according to claim 7, characterized in that, Identifying and determining abnormal tilt of the duct axis, including: Using the line containing the normal vector of the copper foil reference plane as the reference direction for the ideal vertical channel, calculate the angle between the channel's central axis and the reference direction, which is then used as the channel's tilt angle. Based on the preset verticality judgment standard, the tilt angle of each channel is judged. For holes identified as having abnormal axis tilt, mark their positions on the porous copper foil and output the corresponding tilt angle value.