A PCB copper foil peeling detection method and system based on image analysis

Abstract

The present application relates to the technical field of PCB detection, in particular to a PCB copper foil peeling detection method and system based on image analysis. The present application comprehensively evaluates the interlayer bonding state of each detection area, generates an interlayer bonding state evaluation index of copper foil-substrate, thereby realizing early identification and accurate determination of hidden peeling, which can not only detect overt copper foil peeling, but also identify copper foil in the hidden peeling stage, thereby warning potential failure in advance during the manufacturing and use of PCB, effectively improving the quality and reliability of PCB, through multi-angle, all-around optical signal acquisition and data analysis, combined with the generation and optimization of local interface topology, the detection process is more accurate and reliable, significantly reducing the failure rate of PCB in subsequent assembly and use, improving production efficiency, and reducing cost loss caused by failure repair.

Description

Technical Field

[0001] This invention relates to the field of PCB board inspection technology, and in particular to a method and system for detecting copper foil detachment on PCB boards based on image analysis. Background Technology

[0002] PCB copper foil detachment refers to the partial or complete separation of the copper foil layer from the substrate on the printed circuit board. As a key conductive material on the PCB, copper foil plays a role in signal transmission and power distribution. Copper foil detachment can lead to circuit failure, signal loss, or short circuits, and in severe cases, it may cause the entire PCB to malfunction. Therefore, timely detection of copper foil detachment is crucial.

[0003] Currently, existing technologies primarily rely on automated optical inspection to detect copper foil detachment on PCBs. This involves comparing and analyzing surface images of the PCB with standard PCB design images to identify abnormal features such as missing, broken, or separated copper foil areas from the substrate. However, the physical process of copper foil detachment can be divided into two stages: latent and overt. The latent stage refers to interlayer separation of the copper foil from the substrate (caused by thermal stress, adhesive aging, soldering impact, etc.), but the copper foil surface remains intact, without cracks or detachment, only internal voids (interlayer gaps). The overt stage, on the other hand, involves further expansion of interlayer separation, with the copper foil surface cracking, lifting, or completely detaching, forming visible surface defects. Existing automated optical inspection technologies can only detect detachment in the overt stage, completely failing to identify latent interlayer detachment because its surface visual characteristics are consistent with normal copper foil (no abnormalities in grayscale, edges, or area). However, this latent interlayer detachment is a direct hidden danger for subsequent complete copper foil detachment and circuit breakage, leading to PCB failure during subsequent assembly and use. Summary of the Invention

[0004] The main objective of this invention is to provide an image analysis-based method for detecting copper foil detachment on PCB boards, aiming to solve the technical problems in the prior art.

[0005] This invention proposes an image analysis-based method for detecting copper foil detachment on PCB boards, comprising: Acquire multi-dimensional optical signal data and initial structural parameters of the copper foil-substrate interface for each area to be tested on the PCB board; Based on each of the multi-dimensional optical signal data and initial structural parameters, obtain the corresponding associated feature pairs, and construct a weighted kernel matrix based on the associated feature pairs to generate the corresponding high-dimensional associated feature set; Based on each of the high-dimensional correlation feature sets, the interlayer gap features of the corresponding copper foil-substrate are extracted, and the optical response data of the corresponding area to be detected under multiple detection perspectives are acquired in real time during the extraction process. Based on each interlayer gap feature and multiple optical response data, an interlayer bonding state evaluation index for the corresponding copper foil-substrate is generated, and multi-frame cross-sectional optical images of the PCB board under each interlayer bonding state evaluation index are obtained. The interface region pixels of each of the cross-sectional optical images are extracted to construct the corresponding initial cross-sectional contour, and multiple interface concave and convex feature points of each of the initial cross-sectional contours are extracted to draw the corresponding local interface topology map. Obtain the corresponding topology noise point and interface deviation value based on each local interface topology map, and construct the corresponding local optimized interface model based on each topology noise point and interface deviation value; Obtain the corresponding gap parameter dataset based on each of the interlayer bonding state evaluation indicators, and generate a copper foil detachment detection model based on multiple gap parameter datasets and a local optimization interface model. Acquire the target optical signal data and target structural parameters of the target detection area of ​​the PCB board, and input the target optical signal data and target structural parameters into the copper foil detachment detection model to obtain interlayer bonding status evaluation information; The copper foil detachment result is generated based on the interlayer bonding status assessment information.

[0006] This application also provides a PCB board copper foil detachment detection system based on image analysis, including: The acquisition module is used to acquire multi-dimensional optical signal data and initial structural parameters of the copper foil-substrate interface for each area to be tested on the PCB board. The first construction module is used to obtain corresponding associated feature pairs based on each of the multi-dimensional optical signal data and initial structural parameters, and to construct a weighted kernel matrix based on the associated feature pairs to generate a corresponding high-dimensional associated feature set; The first extraction module is used to extract the interlayer gap features of the corresponding copper foil-substrate according to each of the high-dimensional correlation feature sets, and to acquire the optical response data of the corresponding area to be detected under multiple detection perspectives in real time during the extraction process; The first generation module is used to generate an interlayer bonding state evaluation index for the corresponding copper foil-substrate based on each interlayer gap feature and multiple optical response data, and to obtain multi-frame cross-sectional optical images of the PCB board under each interlayer bonding state evaluation index. The second extraction module is used to extract the interface region pixels of each of the cross-sectional optical images to construct the corresponding initial cross-sectional contour, and to extract multiple interface concave and convex feature points of each of the initial cross-sectional contours to draw the corresponding local interface topology map. The second construction module is used to obtain the corresponding topology noise point and interface deviation value according to each local interface topology map, and to construct the corresponding local optimized interface model according to each topology noise point and interface deviation value. The second generation module is used to obtain the corresponding gap parameter dataset according to each of the interlayer bonding state evaluation indicators, and to generate a copper foil detachment detection model according to multiple gap parameter datasets and a local optimization interface model. The input module is used to acquire the target optical signal data and target structural parameters of the target detection area of ​​the PCB board, and input the target optical signal data and target structural parameters into the copper foil detachment detection model to obtain interlayer bonding status evaluation information. The third generation module is used to generate copper foil detachment results based on the interlayer bonding state evaluation information.

[0007] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described image analysis-based PCB copper foil detachment detection method.

[0008] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described image analysis-based PCB copper foil detachment detection method.

[0009] The beneficial effects of this invention are as follows: By comprehensively evaluating the interlayer bonding state of each area to be tested, this invention generates an evaluation index for the interlayer bonding state of copper foil and substrate, thereby achieving early identification and accurate judgment of latent detachment. It can not only detect visible copper foil detachment, but also identify copper foil in the latent detachment stage, thus providing early warning of potential faults during the manufacturing and use of PCB boards, effectively improving the quality and reliability of PCB boards. Through multi-angle and all-round optical signal acquisition and data analysis, combined with the generation and optimization of local interface topology maps, the detection process is made more accurate and reliable, significantly reducing the failure rate of PCB boards in subsequent assembly and use, improving production efficiency, and reducing cost losses caused by fault rework. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention.

[0012] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of this application.

[0013] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0015] like Figure 1 As shown, this application provides a method for detecting copper foil detachment on PCB boards based on image analysis, including: S1. Obtain multi-dimensional optical signal data and initial structural parameters of the copper foil-substrate interface for each area to be tested on the PCB board; S2. Obtain the corresponding associated feature pairs based on each of the multi-dimensional optical signal data and initial structural parameters, and construct a weighted kernel matrix based on the associated feature pairs to generate the corresponding high-dimensional associated feature set; S3. Extract the interlayer gap features of the corresponding copper foil-substrate according to each of the high-dimensional correlation feature sets, and obtain the optical response data of the corresponding area to be detected under multiple detection perspectives in real time during the extraction process; S4. Generate an interlayer bonding state evaluation index for the corresponding copper foil-substrate based on each interlayer gap feature and multiple optical response data, and obtain multi-frame cross-sectional optical images of the PCB board under each interlayer bonding state evaluation index. S5. Extract the interface region pixels of each of the cross-sectional optical images to construct the corresponding initial cross-sectional contour, and extract multiple interface concave and convex feature points of each of the initial cross-sectional contours to draw the corresponding local interface topology map. S6. Obtain the corresponding topology noise point and interface deviation value according to each local interface topology map, and construct the corresponding local optimized interface model according to each topology noise point and interface deviation value; S7. Obtain the corresponding gap parameter dataset according to each of the interlayer bonding state evaluation indicators, and generate a copper foil detachment detection model according to multiple gap parameter datasets and a local optimization interface model. S8. Obtain the target optical signal data and target structural parameters of the target detection area of ​​the PCB board, and input the target optical signal data and target structural parameters into the copper foil detachment detection model to obtain interlayer bonding status evaluation information; S9. Generate copper foil detachment results based on the interlayer bonding status evaluation information.

[0016] As described in steps S1-S9 above, the multi-dimensional optical signal data includes interference light signals, spectral reflection signals, and polarization state signals, which are acquired by a laser interferometer, a spectral analyzer, and a polarization imaging device, respectively. The initial structural parameters include copper foil thickness, substrate thickness, and initial spacing between the copper foil and the substrate, all acquired by an ultrasonic thickness gauge. The step of generating the copper foil detachment result is to determine whether the interlayer bonding state evaluation result is greater than the preset result; if the interlayer bonding state evaluation result is greater than the preset result, the target detection area of ​​the PCB board is determined to be the copper foil detachment area; if the interlayer bonding state evaluation result is not greater than the preset result, the target detection area of ​​the PCB board is determined to be the normal copper foil area.

[0017] Among them, topological noise points refer to noise points in the topology map that deviate from the fitted curve of adjacent feature points by more than a preset deviation threshold, as identified by statistical filtering algorithms. By removing topological noise points and calculating the deviation between the actual interface contour and the ideal interface (a flat contour of perfect copper foil-substrate bonding) in the topology map, the corrected feature points are smoothly fitted using basic spline curves, thereby generating a locally optimized interface model that includes the three-dimensional coordinates of the gap and the interface curvature.

[0018] It should be noted that this invention, by acquiring multi-dimensional optical signal data, can comprehensively capture the optical response of the PCB board surface and interior, avoiding the limitations of traditional methods that rely solely on a single perspective or two-dimensional image analysis. The initial structural parameters of the copper foil-substrate interface provide a foundation for subsequent feature extraction and model generation, ensuring high accuracy and reliability throughout the detection process. The combination of multi-dimensional data and initial structural parameters enables the system to detect latent delamination. Extracting interlayer gap features helps detect early signs of latent delamination, even if these delaminations do not show obvious physical changes on the surface. Acquiring optical response data from multiple detection perspectives improves detection accuracy. Interlayer bonding state evaluation indicators can accurately assess the bonding state between the copper foil and the substrate in each area to be inspected. Multi-frame cross-sectional optical images can display the dynamic changes in the interlayer bonding state from different time points and levels, helping to identify minute defects and anomalies. Quantitative analysis of the interlayer bonding state evaluation indicators can identify and analyze even more subtle changes, preventing potential copper foil delamination.

[0019] Local interface topology maps provide 3D information about the interface between the copper foil and the substrate, further revealing the microstructural features of the interface layer. Optimizing the local interface model aims to analyze the interlayer structure at a microscopic level through refined modeling, providing a more accurate interface model. This helps uncover deeper problems that cannot be identified by ordinary 2D image analysis. By generating local interface topology maps and optimized models, we can delve into the microstructural analysis of the material, detecting hidden delamination risks that conventional methods cannot identify. By collecting multiple gap parameter datasets, we can provide a more detailed analysis of the interlayer bonding state. After generating the copper foil delamination detection model, which is constructed based on an improved convolutional neural network using multiple gap parameter datasets and the locally optimized interface model, its core... The core technology learns the correlation between quantitative parameters of the gap between PCB copper foil and substrate layers (such as average gap width and gap growth rate) and three-dimensional topological features of the interface (such as gap three-dimensional coordinates and interface curvature). This enables precise differentiation between areas with normal surface but hidden internal detachment and areas with completely normal copper foil. It outputs a quantitative assessment result of the interlayer bonding state (detachment probability), rather than relying solely on qualitative judgment based on surface optical features. By inputting the target optical signal data and target structural parameters into a trained copper foil detachment detection model, it can automatically assess the risk of copper foil detachment in the target area, making the detection results more consistent and reliable. By comparing the preset results with the assessment results, it can accurately determine whether there is a copper foil detachment problem in the target area, reducing human error.

[0020] In one embodiment, step S2, which involves obtaining corresponding associated feature pairs based on each of the multi-dimensional optical signal data and initial structural parameters, and constructing a weighted kernel matrix based on the associated feature pairs to generate a corresponding high-dimensional associated feature set, includes: S21. Obtain the light intensity attenuation gradient, phase shift period and spectral absorption peak in the multi-dimensional optical signal data, as well as the copper foil thickness deviation, substrate density distribution and copper foil-substrate initial gap uniformity in the initial structural parameters. S22. Establish a one-to-one correspondence between the light intensity attenuation gradient and the substrate density distribution, the phase shift period and the initial gap uniformity of the copper foil-substrate, and the spectral absorption peak and the copper foil thickness deviation, respectively, to obtain corresponding correlation feature pairs. S23. Construct an initial feature matrix based on multiple associated feature pairs, and obtain the feature redundancy and inter-layer gap correlation degree based on the initial feature matrix; S24. Select multiple core related features based on the redundancy between features and the correlation between layers, and obtain the corresponding improved kernel function parameters and feature weight vector for each core related feature; S25. Construct a weighted kernel matrix based on the multiple improved kernel function parameters and feature weight vectors, and obtain multiple eigenvalues ​​and corresponding eigenvectors based on the weighted kernel matrix; S26. Obtain the variance contribution rate of each feature value, and take the feature vector corresponding to the variance contribution rate being greater than the preset contribution rate threshold as the principal component feature; S27. Arrange the multiple principal component features in descending order of variance contribution rate to form a high-dimensional correlation feature set.

[0021] As described in steps S21-S27 above, the feature value sequence of each row of associated feature pairs is obtained according to the initial feature matrix, and the mutual information entropy of any two rows of feature value sequences is calculated by the K-nearest neighbor estimation method to obtain the feature redundancy. The correlation coefficient between each row of feature value sequence and the preset interlayer gap standard dataset is calculated by the Pearson correlation coefficient formula to obtain the interlayer gap correlation degree.

[0022] By removing associated feature pairs with redundancy exceeding a preset redundancy threshold from the initial feature matrix and retaining associated feature pairs with interlayer gap correlation exceeding a preset correlation threshold, the core associated features can be obtained. Physical property parameters are then obtained based on the physical properties of the copper foil-substrate interface corresponding to the core associated features. Improved kernel function parameters are determined based on the numerical range of the physical property parameters and the core associated features. The improved kernel function parameters, based on the traditional Gaussian kernel function (basic kernel function), incorporate the physical properties of the copper foil-substrate interface and are used to define the improved kernel function's shape and control the feature mapping effect. The traditional Gaussian kernel function only controls the smoothness of the feature space through a single bandwidth parameter, failing to reflect the influence of interface physical properties on feature correlation. The improved kernel function parameters, by fusing physical property factors, enable the kernel function to more accurately capture the correlation between optical-structural features and interlayer gaps. Its core function is to enhance the sensitivity of the high-dimensional associated feature set to implicit interlayer delamination, including the improved bandwidth parameter and physical property parameters. The physical attribute adjustment coefficient is determined by the ratio of the interlayer gap correlation degree of each core associated feature to the sum of the interlayer gap correlation degrees of all core associated features. The corresponding physical weight correction coefficient is obtained based on the physical properties of the copper foil-substrate interface (such as differences in elastic modulus and thermal conductivity ratio) for each core associated feature. These physical properties directly affect the interlayer bonding state of the copper foil-substrate and are strongly correlated with the interlayer gap feature. These physical properties are the physical root cause of implicit interlayer delamination. For example, excessive differences in elastic modulus between the copper foil and the substrate can lead to uneven stress at the interface, easily causing interlayer gaps; abnormal thermal conductivity ratios can lead to differences in interlayer thermal stress during temperature changes, accelerating gap widening. The correction weight of each core associated feature is then obtained by multiplying the basic weight and the physical weight correction coefficient. The correction weights of all core associated features are arranged sequentially according to their sorting order to form a feature weight vector.

[0023] A set of sample points is obtained based on the feature value sequence of the core associated features. A basic kernel function (such as an improved Gaussian kernel function) is constructed based on the improved kernel function parameters. The kernel function value of any two sample points in the sample point set is calculated using the basic kernel function to form a basic kernel matrix. Each modified weight in the feature weight vector is multiplied by all elements in the corresponding row of the basic kernel matrix to obtain the elements of the weighted kernel matrix. The elements of the weighted kernel matrix are arranged in the row and column order of the original basic kernel matrix to construct the weighted kernel matrix. The weighted kernel matrix is ​​then decomposed using the Jacobi iteration method to obtain ordered eigenvalues ​​and corresponding eigenvectors. The proportion of each eigenvalue to the sum of all eigenvalues ​​is used as the variance contribution rate corresponding to that eigenvalue.

[0024] This invention acquires multi-dimensional optical signal data and initial structural parameters, covering both optical and physical structural characteristics, thus enabling more accurate detection of latent detachment signs. By establishing a one-to-one correspondence between different optical signals and structural parameters, it can discover the potential relationship between latent and overt detachment, thereby improving the comprehensiveness of detection. By analyzing feature redundancy, unnecessary repetitive information can be identified and removed. The calculation of interlayer gap correlation can reveal the impact of interlayer structural changes on copper foil detachment, thus providing additional indications of latent detachment. By screening multiple core correlated features, it helps reduce data volume and improve model interpretability. By obtaining improved kernel function parameters and feature weight vectors, the features can be optimized. By applying appropriate weights during model construction, the model's sensitivity to the latent detachment stage can be enhanced. Constructing a weighted kernel matrix can further improve the model's discriminative ability. Mapping data from the original space to a high-dimensional space can better reveal potential interlayer detachment problems. Calculating the variance contribution rate of eigenvalues ​​can effectively screen out principal component features that contribute significantly to the overall system changes, allowing the model to focus on the key features that best reflect latent detachment, thus avoiding interference from redundant features. Selecting eigenvectors with large variance contribution rates helps improve the system's reliability and operability. By sorting multiple principal component features from high to low variance contribution rates, it can be ensured that the model pays more attention to those features that have the greatest impact on copper foil detachment when making predictions.

[0025] In one embodiment, step S3, which involves extracting the interlayer gap features of the corresponding copper foil-substrate based on each of the high-dimensional correlation feature sets, includes: S31. Obtain the correlation coefficient and feature variation coefficient of each principal component feature in the high-dimensional correlation feature set, and obtain the feature-gap determination value of the corresponding principal component feature based on the product of each correlation coefficient and feature variation coefficient. S32. Extract the principal component features corresponding to the gap judgment values ​​that are greater than the preset judgment values ​​from the high-dimensional correlation feature set, and form a gap-sensitive feature group; S33. Obtain the feature-gap mapping model parameters and feature value fluctuation range based on the gap sensitive feature group, and construct a preliminary gap feature extraction model based on the feature-gap mapping model parameters and feature value fluctuation range; S34. Based on the gap features, a preliminary extraction model is used to obtain initial gap parameters and parameter deviation values. Based on the parameter deviation values, the initial gap parameters are corrected to obtain corrected initial gap parameters. S35. The initial gap parameters are optimized using the least squares method to obtain the interlayer gap characteristics of the copper foil-substrate.

[0026] As described in steps S31-S35 above, the correlation coefficient is a quantitative indicator of the degree of correlation between the principal component features in the high-dimensional correlation feature set and the physical quantities of the gap between the copper foil and the substrate (such as gap width and gap distribution density) through the Pearson correlation coefficient. The feature variation coefficient is an indicator that measures the degree of dispersion of the principal component feature values ​​in the high-dimensional correlation feature set. It is calculated by the ratio of the standard deviation of the feature values ​​of each principal component feature to the mean of the feature values. The feature-gap judgment value is a quantitative indicator that combines the fluctuation characteristics of the principal component features themselves with the degree of correlation between the principal component features and the gap between the layers. It is used to objectively measure the effectiveness of each principal component feature in capturing the gap between the copper foil and the substrate, and avoid the bias of single-dimensional judgment.

[0027] The steps for obtaining the feature-gap mapping model parameters and the feature value fluctuation range are as follows: Based on each principal component feature of the gap-sensitive feature group, multiple sets of feature value samples corresponding to known interlayer gap values ​​are collected to form a sample dataset; a nonlinear regression algorithm is used to fit the mapping relationship between the feature value and the interlayer gap value for each principal component feature to obtain the feature-gap mapping model; multiple mapping model parameters (including coefficients, intercepts, etc.) of the feature-gap mapping model are obtained; a significance test (such as an F-test) is performed on the mapping model parameters; significant parameters are retained as feature-gap mapping model parameters; and the feature value fluctuation range is obtained by multiplying the difference between the maximum and minimum values ​​of each sample dataset by a preset fluctuation coefficient (the preset fluctuation coefficient is set by the influence of the accuracy of the detection equipment (such as noise in optical signal acquisition) and the differences in PCB board processes (such as local deviations in copper foil thickness) when collecting feature value samples corresponding to known gap values ​​in the laboratory, in order to expand the coverage of abnormal fluctuations in actual detection).

[0028] The steps for constructing a preliminary gap feature extraction model are as follows: based on the feature-gap mapping model of each principal component feature in the gap-sensitive feature group, a preliminary extraction framework for multi-feature fusion is constructed. The input of the preliminary extraction framework is the feature value of the gap-sensitive feature group, and the output is the initial gap parameter. The input threshold is set according to the fluctuation range of the feature value. When a feature value exceeds the fluctuation range, an outlier handling mechanism is triggered (such as replacing the outlier value with the mean of the principal component feature). Weights are assigned to the feature-gap mapping model of each principal component feature according to the correlation coefficient, and then the preliminary gap feature extraction model is obtained by weighted summation.

[0029] The feature values ​​from the gap-sensitive feature group are input into the gap feature preliminary extraction model, which outputs initial gap parameters. Multiple sets of repeated feature values ​​from the same detection area are collected and input into the model to obtain corresponding initial gap parameters. The standard deviation of the multiple initial gap parameters is calculated as the parameter deviation value. Based on the magnitude of the parameter deviation value, three correction levels are determined: Level 1 correction, Level 2 correction, and Level 3 correction. For example, a parameter deviation value < 2 μm is Level 1 correction, 2 μm ≤ parameter deviation value < 5 μm is Level 2 correction, and a parameter deviation value ≥ 5 μm is Level 3 correction. For Level 1 correction, a moving average method is used to correct the current initial gap parameter by the mean of the initial parameters of the current and the two previous adjacent detection points. The initial parameters can effectively eliminate random noise interference from a single detection while retaining the core information of the initial parameters. For the secondary correction, by introducing external reference parameters (such as the copper foil thickness deviation obtained by an ultrasonic thickness gauge), a correction relationship between the reference parameters and the initial gap parameters is established to correct the deviation, avoiding the limitations of simply relying on internal data correction and improving the accuracy of moderately fluctuating parameters. For the tertiary correction, the gap-sensitive feature values ​​of the area to be detected are re-acquired and input into the model to obtain new initial gap parameters and parameter deviation values ​​until the new deviation value is less than the preset value, thus obtaining the corrected initial gap parameters. This ensures that the correction results fit the actual interface state and solves the problem of accurate correction under extreme deviations.

[0030] Based on the correction of the initial gap parameters, a least-squares optimization objective function is constructed. The objective function is to minimize the sum of squared residuals between the initial gap parameters and the optimized parameters at each detection point. Constraints are introduced, including that the optimized gap parameters must be within the physical gap range of the PCB board and that the rate of change of the optimized parameters at adjacent detection points is less than the preset rate of change (to avoid abrupt parameter changes). The objective function is solved using a constrained least-squares method (such as the Lagrange multiplier method). The optimized gap parameters at each detection point are obtained, and the residuals between the optimized parameters and the initial parameters after error correction are calculated. If the residual is less than the preset residual, the optimized parameters are used as the final interlayer gap features (including the average gap width, gap uniformity, and gap edge roughness). Otherwise, the constraints are readjusted, and least-squares optimization is performed again until the residual is not less than the preset residual.

[0031] This invention obtains the feature-gap determination value by acquiring the correlation coefficient and feature variation coefficient of each principal component feature and calculating their product. This is because interlayer gap-sensitive features must simultaneously satisfy the requirements of being able to distinguish different gap states through numerical differences (relying on the feature variation coefficient) and that the numerical changes originate from gap changes rather than other interferences (relying on the correlation coefficient). The feature-gap determination value formed by multiplying the two can quantitatively evaluate the practical value of the feature. Only features that simultaneously possess high volatility (high variation coefficient) and high physical correlation (correlation coefficient) can obtain a high determination value, thus being accurately selected as gap-sensitive features. This provides a reliable foundation for subsequent interlayer gap feature extraction, thereby effectively identifying copper foil interlayer detachment in the latent stage. Interlayer detachment in the latent stage does not show obvious abnormalities in surface visual features and is difficult to detect using traditional optical techniques alone. By selecting principal component features with feature-gap determination values ​​greater than a preset threshold, a gap-sensitive feature group is formed, ensuring that only features sensitive to gap changes are selected, optimizing the feature set of the data, and improving the accuracy and reliability of the model.

[0032] By obtaining feature-gap mapping model parameters based on gap-sensitive feature groups and combining the feature value fluctuation range, a preliminary gap feature extraction model is constructed. The complex relationship between features and interlayer gaps is transformed into quantifiable model parameters through the mapping relationship, thereby ensuring that the system can more accurately identify and predict the risk of interlayer delamination. By correcting the initial gap parameters, more accurate corrected initial gap parameters can be obtained. Through error correction technology, the deviations introduced by external factors (such as material inhomogeneity, changes in production process, etc.) in the model can be effectively removed, making the estimation of gap parameters more accurate. The corrected initial gap parameters are optimized by the least squares method to ensure that the optimal copper foil-substrate interlayer gap features are obtained, further improving the model fitting accuracy.

[0033] Current automated optical inspection technologies can only detect visible delamination but cannot identify latent delamination. This is because latent delamination refers to interlayer delamination that has not yet developed to the surface visible stage, and the visual characteristics of the copper foil surface are almost identical to those of normal copper foil. Automated optical inspection technologies mainly rely on surface optical features such as grayscale, edges, and area, and cannot perform multidimensional statistical analysis or principal component correlation analysis, thus failing to distinguish this type of latent delamination. Therefore, existing technologies cannot identify interlayer delamination in advance through features that are not obviously abnormal on the surface. This invention, however, obtains correlated feature pairs by combining multidimensional optical signals with initial structural parameters, and then generates high-dimensional correlated features. This method can capture minute optical or structural changes that traditional automated optical inspection technologies cannot identify. Through high-dimensional correlation feature analysis, it extracts minute optical signals and structural parameter changes to obtain interlayer gap features and capture latent detachment signals. This solves the problem of latent detachment that cannot be reflected by surface visual features. By establishing a feature-gap mapping model and extracting accurate gap parameters through correction and least squares optimization, the quantification accuracy of interlayer gaps is improved, providing reliable data for latent detachment assessment. By screening sensitive features, suppressing noise, and adapting to batch differences, the reliability and universality of latent detachment detection are enhanced.

[0034] In one embodiment, step S4, which generates a corresponding evaluation index for the interlayer bonding state of the copper foil-substrate based on each interlayer gap characteristic and multiple optical response data, includes: S41. Obtain gap quantification parameters and gap risk coefficients based on the interlayer gap characteristics, wherein the gap quantification parameters include the average gap width, gap uniformity, and gap edge roughness; S42. Obtain optical characteristic parameters and optical consistency coefficients based on multiple optical response data, wherein the optical characteristic parameters include light intensity attenuation rate, phase shift, and polarization state change degree; S43. Establish corresponding mapping relationships between the average gap width and the light intensity attenuation rate, the gap uniformity and the phase shift, and the gap edge roughness and the polarization state change degree, respectively. S44. Use the cosine similarity algorithm to obtain the correlation degree of each mapping relationship, and obtain the average correlation degree based on multiple correlation degrees; S45. Calculate the interlayer bonding status evaluation index based on the average correlation coefficient, gap risk coefficient, and optical consistency coefficient, wherein the calculation formula for the interlayer bonding status evaluation index is: A = [α*B + β*(1-C)]*D; Where A represents the interlayer bonding status assessment index, α represents the dynamic weight of the gap risk coefficient, B represents the gap risk coefficient, β represents the dynamic weight of the optical consistency coefficient, C represents the optical consistency coefficient, and D represents the average correlation degree.

[0035] As described in steps S41-S45 above, the dynamic weights of the gap risk coefficient and the optical consistency coefficient are dynamically adjusted according to the gap risk coefficient. When the gap risk coefficient > a preset risk coefficient threshold, the weight of the gap risk coefficient is increased while the weight of the optical consistency coefficient is decreased, prioritizing the risk inherent in the gap itself. When the gap risk coefficient is not greater than the preset risk coefficient threshold, the weight of the gap risk coefficient is decreased while the weight of the optical consistency coefficient is increased. The gap risk coefficient is calculated by the ratio of the average gap width to the maximum allowable gap in the PCB board process (the larger the ratio, the higher the risk coefficient). The maximum allowable gap in the process is a critical safety value for interlayer gaps determined in PCB board manufacturing based on product design requirements, material characteristics, usage scenarios, and industry standards. It can be determined through process documents and design specifications or industry / national mandatory standards. The optical consistency coefficient is determined by the difference between the maximum standard deviation reference value and the standard deviation of the data samples of optical response data under multiple detection angles, and then by the ratio of this difference to the maximum standard deviation reference value. Where C represents the optical consistency coefficient, E represents the maximum standard deviation reference value, and F represents the data sample standard deviation. The maximum standard deviation reference value can be obtained through historical sample statistics.

[0036] It should be noted that this invention, by introducing quantified gap parameters, can more objectively and accurately characterize the bonding state between the copper foil and the substrate. The gap risk coefficient helps assess potential problems in interlayer bonding. By combining multiple optical characteristic parameters, it can effectively reduce misjudgments or incomplete judgments that may arise from relying solely on a single optical response. The introduction of the optical consistency coefficient can eliminate noise caused by fluctuations in optical response data, thereby improving the stability and accuracy of the evaluation results. By establishing a systematic mapping relationship between different interlayer gap parameters and optical characteristics, it can effectively combine these two different detection methods, making the detection results more logical and accurate. By mapping different characteristic parameters and optical responses, the interlayer bonding state can be comprehensively evaluated from multiple perspectives, especially for early warning of latent delamination processes. By combining the average correlation, gap risk coefficient, and optical consistency coefficient to calculate the interlayer bonding state evaluation index, the correlation between risk, optical characteristics, and mapping relationships can comprehensively reflect the bonding state between the copper foil and the substrate, thereby enabling the timely detection of potential latent problems during the production process and avoiding future failures.

[0037] In one embodiment, step S5, which involves extracting interface region pixels from each of the cross-sectional optical images to construct a corresponding initial cross-sectional contour and extracting multiple interface concave-convex feature points from each of the initial cross-sectional contours to draw a corresponding local interface topology map, includes: S51. Extract the interface region pixels of each of the cross-sectional optical images, and obtain the interface edge grayscale gradient of each of the interface region pixels. S52. The interface region pixels corresponding to the interface edge grayscale gradient that is greater than the preset gradient threshold are taken as edge pixels, and multiple edge pixels are fitted to construct an initial interface contour. S53. Extract all interface concave and convex feature points in the initial interface contour, and obtain the topological position data of each interface concave and convex feature point. S54. Connect the corresponding interface concave and convex feature points according to each of the topological location data to generate a local interface topology map.

[0038] S6. Obtain the corresponding topology noise point and interface deviation value according to each local interface topology map, and construct the corresponding local optimized interface model according to each topology noise point and interface deviation value; As described in steps S51-S54 and S6 above, the interface pixels are extracted from the cross-sectional optical image using a threshold segmentation algorithm (such as the Otsu's method), and the gray-level gradient of the interface edge can be calculated using the Sobel operator. The interface concave and convex feature points can be extracted from the initial interface contour using the Harris corner detection algorithm. The interface concave and convex feature points can be such as gap protrusions, depressions, etc. A two-dimensional coordinate system is established with the lower left corner of the cross-sectional optical image as the origin. By connecting the corresponding interface concave and convex feature points according to the topological position data, a local interface topology map containing gap distribution and interface flatness information can be formed.

[0039] This invention, by acquiring the grayscale gradient of the interface region, can provide more reliable basic data for subsequent edge recognition and contour fitting. By using a fitting method instead of simple edge detection, errors can be reduced, and an initial contour that more closely matches the actual interface shape can be constructed. This helps to reduce missed detections caused by irregular interfaces or minor defects, and provides a more accurate basis for subsequent feature point extraction and topology map construction. By drawing the connections between concave and convex feature points, the overall shape of the interface and the distribution of defects can be presented intuitively. The topology map can not only show the local interface shape, but also reveal the continuity of defects and potential expansion trends, providing a basis for optimizing subsequent interface repair schemes.

[0040] Existing technologies primarily target the visible detachment stage, where the copper foil exhibits obvious abnormal changes in appearance. However, they fail to address noise and instability in the topology diagram, potentially leading to detection errors or unstable results. This invention, through the construction of a locally optimized interface model, further analyzes the extracted local interface data, filters out topological noise points, and quantifies interface deviation values. This allows for the identification of copper foil detachment issues that are not readily apparent on the surface, providing early warning of potential circuit fault risks. By extracting multiple interface feature points, it can meticulously capture subtle changes in the detachment process—changes that might be overlooked in traditional grayscale analysis methods. The local optimization method automatically removes irrelevant topological noise points, ensuring model accuracy and avoiding false alarms and missed alarms. It can identify latent copper foil detachment problems in advance, and through refined topological data analysis and noise adaptive optimization, it significantly improves the accuracy and sensitivity of the detection technology.

[0041] In one embodiment, step S7, which involves obtaining a corresponding gap parameter dataset based on each of the interlayer bonding state evaluation indices and generating a copper foil detachment detection model based on multiple gap parameter datasets and a local optimization interface model, includes: S71. Obtain the index level classification result and the corresponding raw gap data of the detection area according to each interlayer bonding state evaluation index, and filter the effective gap parameters in the raw gap data according to each index level classification result to obtain the corresponding gap parameter dataset. S72. Obtain a parameter feature vector based on each of the gap parameter datasets, wherein the parameter feature vector includes the mean gap width, the gap growth rate, and the gap stability; S73. Obtain the three-dimensional features of the interface based on the local optimization interface model, and construct a model training sample set based on the three-dimensional features of the interface and multiple parameter feature vectors, wherein the three-dimensional features of the interface include the three-dimensional coordinates of the gap, the curvature of the interface, and the flatness of the interface. S74. Divide the model training sample set into a training set, a validation set, and a test set, and train an improved convolutional neural network model based on the training set and the validation set. With the goal of minimizing the mean square error between the interlayer combined state evaluation index and the model prediction result, iteratively adjust the network parameters until the model recognition accuracy reaches the preset accuracy based on the test set, and obtain the copper foil detachment detection model.

[0042] As described in steps S71-S74 above, the step of screening valid gap parameters in the raw gap data involves obtaining the index level classification results (e.g., 0-30 for normal level, 31-60 for suspicious level, and 61-100 for risk level) and the raw gap data. The raw gap data refers to all quantitative data related to the gap between copper foil and substrate layers that are directly collected from the area to be inspected on the PCB board without screening or depth correction. This includes basic parameters such as gap width, gap distribution density, and gap edge roughness, as well as auxiliary data such as environmental parameters (e.g., detection temperature and humidity) and equipment parameters (e.g., optical signal sampling frequency) during the collection process. The data is screened according to the index level classification results. For example, when the raw gap data is the interlayer gap width, parameters with an interlayer gap width ≤ 3μm are retained for the normal level, parameters with an interlayer gap width ≤ 8μm are retained for the suspicious level, and parameters with an interlayer gap width > 8μm are retained for the risk level.

[0043] The parameter feature vector and the interface 3D features are weighted and fused to obtain a fused feature vector. When training the improved convolutional neural network model, the fused feature vector is used as input. Deep features are extracted through 3 convolutional layers and 2 fully connected layers are used to map the feature dimensions. The output layer uses the Sigmoid function to output the debonding probability. The copper foil debonding detection model includes a 3D convolutional layer, a parameter attention module, and a gap temporal fusion layer. The input layer adopts a dual-input structure of parameter feature vector and 3D interface feature map to adapt to the dual information of quantization parameters and spatial morphology of the gap between layers. The first input is the parameter feature vector, which is mapped to the corresponding feature map through the fully connected layer. The second input is the interface 3D feature map (generated by rendering the gap 3D coordinates, interface curvature, and interface flatness of the locally optimized interface model), which preserves the spatial distribution details of the gap (such as the position, shape, and depth of the gap). The two inputs are fused into a hybrid feature map through feature concatenation, which provides comprehensive information for subsequent feature extraction, which is different from the limitation of conventional convolutional neural networks that only have a single image input.

[0044] For the feature extraction layer, a 3D convolutional layer is used instead of a conventional 2D convolutional layer, and a custom gap-sensitive convolutional kernel is used. Two 3D convolutional layers are set, where the feature extraction of the convolutional layer is Y. k =σ(W k *X+b k ), k=1,2,3; where Y k Let W represent the output feature map of the k-th convolutional layer, σ represent the ReLU activation function, and W represent the output feature map of the k-th convolutional layer. k Let X represent the weight matrix of the k-th convolutional kernel, and let b represent the fused feature vector. k This represents the bias vector of the k-th layer, where k represents the number of convolutional layers, * represents the convolution operation, and the dimension mapping of the fully connected layer is Z1=σ(U1*Flatten(Y)). kZ2 = σ(U2*Z1+c2); where U1 and U2 represent the weight matrices of the first and second fully connected layers, Z1 and Z2 represent the outputs of the first and second fully connected layers, c1 and c2 represent the first and second bias vectors, and Flatten represents the feature map flattening operation.

[0045] For the attention module, a gap parameter attention mechanism is adopted. Addressing the characteristic that gap parameters dominate risk (e.g., the risk of detachment increases significantly when the gap growth rate is >5%), a parameter attention module is designed to strengthen the impact of key parameters on the detection results. Considering the dynamic changes in interlayer gaps (e.g., gaps may develop into explicit detachment over time), a gap temporal fusion layer is added to associate historical gap parameters within the same detection region. Historical parameters are mapped to corresponding temporal feature maps via fully connected layers and fused with the current feature map through temporal attention weighting (the weight of historical parameters decays over time, e.g., 0.3 for the first time, 0.1 for the second time). The fused feature map is then subjected to global average pooling to obtain the corresponding global feature vector, reflecting the comprehensive information of current spatial features and historical parameter changes. This addresses the shortcomings of conventional convolutional neural networks that only consider single-frame data and ignore dynamic gap risks.

[0046] For the output layer, to address the interlayer detachment risk classification requirement, the implicit detachment probability is used to determine whether implicit detachment exists. The output layer detachment probability is calculated as follows: Where P represents the output layer dropout probability, V represents the output layer weight vector, Z2 represents the fully connected layer output, and d represents the bias term.

[0047] It should be noted that this invention, by acquiring and classifying interlayer bonding state evaluation indicators, can categorize different interlayer detachment states, enabling precise identification and filtering of gap data in each state. After acquiring the raw gap data, feature vector extraction effectively transforms the data into an operable parameter form, allowing subsequent model training to focus on the most representative features. Through structured feature vector extraction, the entire data processing flow becomes more standardized and clear, providing a solid data foundation for subsequent optimization steps and model iteration. The locally optimized interface model, by extracting three-dimensional features, can more realistically recreate the spatial distribution during the copper foil detachment process. Features can provide a more accurate description of the detachment state in the spatial dimension. By combining three-dimensional interface features and multi-dimensional parameter feature vectors, a multi-dimensional and multi-angle training sample set can be constructed to comprehensively reflect the complex mechanism of copper foil detachment. By introducing a locally optimized interface model, the model's ability to perceive local details is further improved and combined with the overall detection target. By taking the minimization of the mean square error between the inter-layer combined state evaluation index and the model prediction results as the objective, the network parameters are iteratively adjusted to achieve the best balance between recognition accuracy and error rate. This ensures that the training process does not focus excessively on overfitting of a specific region, but rather comprehensively optimizes the overall performance of the model.

[0048] like Figure 2 As shown, this application also provides an image analysis-based PCB copper foil detachment detection system, comprising: The acquisition module is used to acquire multi-dimensional optical signal data and initial structural parameters of the copper foil-substrate interface for each area to be tested on the PCB board. The first construction module is used to obtain corresponding associated feature pairs based on each of the multi-dimensional optical signal data and initial structural parameters, and to construct a weighted kernel matrix based on the associated feature pairs to generate a corresponding high-dimensional associated feature set; The first extraction module is used to extract the interlayer gap features of the corresponding copper foil-substrate according to each of the high-dimensional correlation feature sets, and to acquire the optical response data of the corresponding area to be detected under multiple detection perspectives in real time during the extraction process; The first generation module is used to generate an interlayer bonding state evaluation index for the corresponding copper foil-substrate based on each interlayer gap feature and multiple optical response data, and to obtain multi-frame cross-sectional optical images of the PCB board under each interlayer bonding state evaluation index. The second extraction module is used to extract the interface region pixels of each of the cross-sectional optical images to construct the corresponding initial cross-sectional contour, and to extract multiple interface concave and convex feature points of each of the initial cross-sectional contours to draw the corresponding local interface topology map. The second construction module is used to obtain the corresponding topology noise point and interface deviation value according to each local interface topology map, and to construct the corresponding local optimized interface model according to each topology noise point and interface deviation value. The second generation module is used to obtain the corresponding gap parameter dataset according to each of the interlayer bonding state evaluation indicators, and to generate a copper foil detachment detection model according to multiple gap parameter datasets and a local optimization interface model. The input module is used to acquire the target optical signal data and target structural parameters of the target detection area of ​​the PCB board, and input the target optical signal data and target structural parameters into the copper foil detachment detection model to obtain interlayer bonding status evaluation information. The third generation module is used to generate copper foil detachment results based on the interlayer bonding state evaluation information.

[0049] In one embodiment, the second extraction module includes: An extraction unit is used to extract the interface region pixels of each of the cross-sectional optical images and obtain the interface edge grayscale gradient of each of the interface region pixels. The fitting construction unit is used to take the interface region pixels corresponding to the interface edge grayscale gradient that is greater than the preset gradient threshold as edge pixels, and to fit multiple edge pixels to construct an initial interface contour. The first acquisition unit is used to extract all interface concave and convex feature points in the initial interface contour and acquire the topological position data of each interface concave and convex feature point. The generation unit is used to draw and connect the corresponding interface concave and convex feature points according to each of the topological location data to generate a local interface topology map. The second acquisition unit is used to acquire topological noise points and interface deviation values ​​based on the local interface topology map, and to construct a local optimized interface model based on the topological noise points and interface deviation values.

[0050] It should be noted that each module and unit in the image analysis-based PCB copper foil detachment detection system corresponds one-to-one with the steps in the image analysis-based PCB copper foil detachment detection method.

[0051] like Figure 3 As shown, this application also provides a computer device, which can be a server, and its internal structure can be as follows: Figure 3As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores all data required for the process of the image analysis-based PCB copper foil detachment detection method. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the image analysis-based PCB copper foil detachment detection method.

[0052] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0053] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described image analysis-based methods for detecting copper foil detachment from PCB boards.

[0054] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0055] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0056] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for detecting copper foil detachment on PCB boards based on image analysis, characterized in that, include: Acquire multi-dimensional optical signal data and initial structural parameters of the copper foil-substrate interface for each area to be tested on the PCB board; Based on each of the multi-dimensional optical signal data and initial structural parameters, obtain the corresponding associated feature pairs, and construct a weighted kernel matrix based on the associated feature pairs to generate the corresponding high-dimensional associated feature set; Based on each of the high-dimensional correlation feature sets, the interlayer gap features of the corresponding copper foil-substrate are extracted, and the optical response data of the corresponding area to be detected under multiple detection perspectives are acquired in real time during the extraction process. Based on each interlayer gap feature and multiple optical response data, an interlayer bonding state evaluation index for the corresponding copper foil-substrate is generated, and multi-frame cross-sectional optical images of the PCB board under each interlayer bonding state evaluation index are obtained. The interface region pixels of each of the cross-sectional optical images are extracted to construct the corresponding initial cross-sectional contour, and multiple interface concave and convex feature points of each of the initial cross-sectional contours are extracted to draw the corresponding local interface topology map. Obtain the corresponding topology noise point and interface deviation value based on each local interface topology map, and construct the corresponding local optimized interface model based on each topology noise point and interface deviation value; Obtain the corresponding gap parameter dataset based on each of the interlayer bonding state evaluation indicators, and generate a copper foil detachment detection model based on multiple gap parameter datasets and a local optimization interface model. Acquire the target optical signal data and target structural parameters of the target detection area of ​​the PCB board, and input the target optical signal data and target structural parameters into the copper foil detachment detection model to obtain interlayer bonding status evaluation information; The copper foil detachment result is generated based on the interlayer bonding status assessment information.

2. The PCB board copper foil detachment detection method based on image analysis according to claim 1, characterized in that, The step of obtaining corresponding associated feature pairs based on each of the multi-dimensional optical signal data and initial structural parameters, and constructing a weighted kernel matrix based on the associated feature pairs to generate a corresponding high-dimensional associated feature set, includes: Acquire the light intensity attenuation gradient, phase shift period, and spectral absorption peak from the multi-dimensional optical signal data, as well as the copper foil thickness deviation, substrate density distribution, and copper foil-substrate initial gap uniformity from the initial structural parameters; A one-to-one correspondence was established between the light intensity attenuation gradient and the substrate density distribution, the phase shift period and the initial gap uniformity of the copper foil-substrate, and the spectral absorption peak and the copper foil thickness deviation, respectively, to obtain corresponding correlation feature pairs; An initial feature matrix is ​​constructed based on multiple pairs of associated features, and the redundancy between features and the interlayer gap correlation are obtained based on the initial feature matrix. Multiple core related features are selected based on the redundancy between features and the correlation between interlayer gaps, and corresponding improved kernel function parameters and feature weight vectors are obtained for each core related feature. A weighted kernel matrix is ​​constructed based on multiple improved kernel function parameters and feature weight vectors, and multiple eigenvalues ​​and corresponding eigenvectors are obtained based on the weighted kernel matrix; Obtain the variance contribution rate of each feature value, and take the feature vector corresponding to the variance contribution rate being greater than a preset contribution rate threshold as the principal component feature; The principal component features are arranged from high to low according to their variance contribution rate to form a high-dimensional correlation feature set.

3. The PCB board copper foil detachment detection method based on image analysis according to claim 1, characterized in that, The step of extracting the interlayer gap features of the corresponding copper foil-substrate based on each of the high-dimensional correlation feature sets includes: Obtain the correlation coefficient and feature variation coefficient of each principal component feature in the high-dimensional correlation feature set, and obtain the feature-gap determination value of the corresponding principal component feature based on each correlation coefficient and feature variation coefficient; Extract the principal component features corresponding to gap judgment values ​​greater than preset judgment values ​​from the high-dimensional correlation feature set, and form a gap-sensitive feature group; Based on the gap-sensitive feature group, obtain the feature-gap mapping model parameters and feature value fluctuation range, and construct a preliminary gap feature extraction model based on the feature-gap mapping model parameters and feature value fluctuation range; The initial gap parameters and parameter deviation values ​​are obtained by preliminary extraction model based on the gap features, and the initial gap parameters are corrected based on the parameter deviation values ​​to obtain the corrected initial gap parameters. The interlayer gap characteristics of copper foil-substrate were obtained by optimizing the modified initial gap parameters using the least squares method.

4. The PCB board copper foil detachment detection method based on image analysis according to claim 1, characterized in that, The step of generating corresponding interlayer bonding state evaluation indicators for copper foil-substrate based on each interlayer gap characteristic and multiple optical response data includes: Based on the interlayer gap characteristics, gap quantification parameters and gap risk coefficients are obtained, wherein the gap quantification parameters include the average gap width, gap uniformity, and gap edge roughness; Optical characteristic parameters and optical consistency coefficients are obtained based on multiple optical response data, wherein the optical characteristic parameters include light intensity attenuation rate, phase shift, and polarization state change degree; Establish corresponding mapping relationships between the average gap width and the light intensity attenuation rate, the gap uniformity and the phase shift, and the gap edge roughness and the polarization state change degree, respectively. The cosine similarity algorithm is used to obtain the correlation degree of each mapping relationship, and the average correlation degree is obtained based on multiple correlation degrees. The interlayer bonding status evaluation index is calculated based on the average correlation coefficient, gap risk coefficient, and optical consistency coefficient. The calculation formula for the interlayer bonding status evaluation index is as follows: A = [α*B + β*(1-C)]*D; Where A represents the interlayer bonding status assessment index, α represents the dynamic weight of the gap risk coefficient, B represents the gap risk coefficient, β represents the dynamic weight of the optical consistency coefficient, C represents the optical consistency coefficient, and D represents the average correlation degree.

5. The PCB board copper foil detachment detection method based on image analysis according to claim 1, characterized in that, The steps of extracting interface region pixels from each of the cross-sectional optical images to construct a corresponding initial cross-sectional contour, and extracting multiple interface concave-convex feature points from each of the initial cross-sectional contours to draw a corresponding local interface topology map, include: Extract the interface region pixels of each of the cross-sectional optical images, and obtain the interface edge grayscale gradient of each interface region pixel; The pixels in the interface region corresponding to the gray-scale gradient of the interface edge that is greater than the preset gradient threshold are taken as edge pixels, and the initial interface contour is obtained by fitting multiple edge pixels. Extract all interface concave and convex feature points in the initial interface contour, and obtain the topological position data of each interface concave and convex feature point; Based on each of the topological location data, the corresponding interface concave and convex feature points are drawn and connected to generate a local interface topology map.

6. The PCB board copper foil detachment detection method based on image analysis according to claim 1, characterized in that, The step of obtaining the corresponding gap parameter dataset based on each of the interlayer bonding state evaluation indices, and generating a copper foil detachment detection model based on multiple gap parameter datasets and a local optimization interface model, includes: Based on each interlayer bonding state evaluation index, the index level classification result and the corresponding raw gap data of the detection area are obtained, and the effective gap parameters in the raw gap data are filtered according to each index level classification result to obtain the corresponding gap parameter dataset. Obtain parameter feature vectors based on each of the gap parameter datasets; The interface three-dimensional features are obtained based on the local optimization interface model, and a model training sample set is constructed based on the interface three-dimensional features and multiple parameter feature vectors. The model training sample set is divided into a training set, a validation set, and a test set. An improved convolutional neural network model is trained based on the training set and the validation set. The network parameters are iteratively adjusted with the goal of minimizing the mean square error between the interlayer combined state evaluation index and the model prediction result. The accuracy of the model recognition reaches the preset accuracy rate based on the test set, thus obtaining the copper foil detachment detection model.

7. A PCB board copper foil detachment detection system based on image analysis, characterized in that, include: The acquisition module is used to acquire multi-dimensional optical signal data and initial structural parameters of the copper foil-substrate interface for each area to be tested on the PCB board. The first construction module is used to obtain corresponding associated feature pairs based on each of the multi-dimensional optical signal data and initial structural parameters, and to construct a weighted kernel matrix based on the associated feature pairs to generate a corresponding high-dimensional associated feature set; The first extraction module is used to extract the interlayer gap features of the corresponding copper foil-substrate according to each of the high-dimensional correlation feature sets, and to acquire the optical response data of the corresponding area to be detected under multiple detection perspectives in real time during the extraction process; The first generation module is used to generate an interlayer bonding state evaluation index for the corresponding copper foil-substrate based on each interlayer gap feature and multiple optical response data, and to obtain multi-frame cross-sectional optical images of the PCB board under each interlayer bonding state evaluation index. The second extraction module is used to extract the interface region pixels of each of the cross-sectional optical images to construct the corresponding initial cross-sectional contour, and to extract multiple interface concave and convex feature points of each of the initial cross-sectional contours to draw the corresponding local interface topology map. The second construction module is used to obtain the corresponding topology noise point and interface deviation value according to each local interface topology map, and to construct the corresponding local optimized interface model according to each topology noise point and interface deviation value. The second generation module is used to obtain the corresponding gap parameter dataset according to each of the interlayer bonding state evaluation indicators, and to generate a copper foil detachment detection model according to multiple gap parameter datasets and a local optimization interface model. The input module is used to acquire the target optical signal data and target structural parameters of the target detection area of ​​the PCB board, and input the target optical signal data and target structural parameters into the copper foil detachment detection model to obtain interlayer bonding status evaluation information. The third generation module is used to generate copper foil detachment results based on the interlayer bonding state evaluation information.

8. The PCB copper foil detachment detection system based on image analysis according to claim 7, characterized in that, The second extraction module includes: An extraction unit is used to extract the interface region pixels of each of the cross-sectional optical images and obtain the interface edge grayscale gradient of each of the interface region pixels. The fitting construction unit is used to take the interface region pixels corresponding to the interface edge grayscale gradient that is greater than the preset gradient threshold as edge pixels, and to fit multiple edge pixels to construct an initial interface contour. The first acquisition unit is used to extract all interface concave and convex feature points in the initial interface contour and acquire the topological position data of each interface concave and convex feature point. The generation unit is used to draw and connect the corresponding interface concave and convex feature points according to each of the topological location data to generate a local interface topology map.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images