Solar cell panel busbar laser stitch welding defect detection method and system

Abstract

The present application relates to photovoltaic material processing detection technical field, specifically disclose a kind of solar cell panel busbar laser build-up welding defect detection method, comprising: after the pretreatment of busbar laser build-up welding image sample is labeled with defect category, to obtain training set;Extract each covariance matrix in training set to constitute Lie algebra training point set;Lie algebra space point in Lie algebra training point set is clustered in Lie algebra space, and the center point of each category Lie algebra space point is obtained;The Euclidean distance between the Lie algebra test point of the busbar laser build-up welding image to be detected respectively to each category Lie algebra center point is calculated, and the category of the shortest Euclidean distance Lie algebra center point belongs is judged as the category of the busbar laser build-up welding image to be detected.The present application further discloses a kind of solar cell panel busbar laser build-up welding defect detection system.The present application can improve the accuracy and learning efficiency of solar cell panel busbar laser build-up welding defect identification.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic material processing and testing technology, and more specifically, to a method and system for detecting defects in laser overlay welding of busbars in solar panels. Background Technology

[0002] Busbar welding is a crucial step in the production of solar panel modules, and its welding performance directly determines the efficiency of the solar panel. Laser welding, being a non-contact welding process, offers advantages such as high energy density, deep and wide welds, fast welding speed, good repeatability, minimal workpiece deformation, and a small heat-affected zone. Therefore, laser welding of busbars has broad application prospects in the global photovoltaic power generation industry. Laser welding of solar panel busbars involves welding the interconnecting strips of two adjacent cell strings together, effectively connecting them in parallel. Since both the busbars and interconnecting strips are made of tin-coated copper alloy, representing a lap-welding of the same alloy, their welding performance is a significant quality factor affecting cell lifespan. Therefore, the inspection of the forming quality of laser-laminated busbars in solar panels has important theoretical and practical value.

[0003] Currently, the main methods for inspecting the welding quality of solar panel busbars in China include manual visual inspection, non-destructive testing (NDT), and machine vision inspection. Manual visual inspection of welding defects is inefficient and prone to missing defects. NDT has low sensitivity and resolution for defects. Machine vision-based laser welding quality inspection, due to its real-time capability, high efficiency, and accuracy, has become an important technological means for intelligent manufacturing in the photovoltaic industry. Existing visual inspection algorithms, such as neural networks, wavelet analysis, and support vector machines, suffer from drawbacks such as high computational load, numerous parameters, and complex model structures. Their classification efficiency and accuracy are not ideal, and their application effects are not significant, failing to adequately meet the needs of enterprise production. Summary of the Invention

[0004] To address the shortcomings and deficiencies of existing technologies, this invention provides a method and system for detecting defects in laser-laminated busbars of solar panels. Utilizing multidimensional correlation features of images, a covariance feature matrix is ​​extracted to form a Lie group manifold. Then, leveraging the mapping relationship between Lie groups and Lie algebras, a Mean-Shift algorithm is established to cluster the image covariance matrix on the Lie group manifold. The cluster whose Lie algebra center point is closest to the weld image to be detected is identified as belonging to that cluster. This method, utilizing the geometric structure of the Lie group manifold of the image covariance matrix, can improve the accuracy of laser-laminated busbar defect identification and the efficiency of machine learning, making intelligent inspection on actual production lines possible.

[0005] As a first aspect of the present invention, a method for detecting defects in laser overlay welding of busbars in solar panels is provided, comprising:

[0006] Step S1: Acquire busbar laser overlay image samples and perform image preprocessing;

[0007] Step S2: Mark the preprocessed busbar laser lap welding image samples, select the busbar laser lap welding image samples with defects, and register the defect categories to obtain the training set;

[0008] Step S3: Extract the covariance matrix of each bus laser slab welding image sample in the training set to form a first Lie group manifold, and map the first Lie group manifold to the Lie algebra space to form a Lie algebra training point set;

[0009] Step S4: In the Lie algebra space, the Mean-Shift algorithm is used to cluster the Lie algebra space points of all busbar laser slab welding image samples in the Lie algebra training point set, and the center point of each category of Lie algebra space points is obtained.

[0010] Step S5: Acquire the laser slab image of the busbar to be tested, and extract the covariance matrix of the laser slab image of the busbar to be tested to form a second Lie group manifold. Map the second Lie group manifold to the Lie algebra space to form a Lie algebra test point set, wherein the Lie algebra test point set includes the Lie algebra test points of all the laser slab images of the busbar to be tested.

[0011] Step S6: Calculate the Euclidean distance between each Lie algebra test point in the Lie algebra test point set and the center point of each category of Lie algebra space point, and determine the category of the Lie algebra space point center point with the shortest Euclidean distance as the category of the laser overlay image of the busbar to be detected.

[0012] Further, in step S3, the step of extracting the covariance matrix of each bus laser shing image sample in the training set to form a first Lie group manifold, and mapping the first Lie group manifold to the Lie algebra space to form a Lie algebra training point set, further includes:

[0013] Let I be a busbar laser shing image sample with resolution M×N. Its color features are represented by R(x,y), G(x,y), and B(x,y), where x = 0, 1, ..., N-1, y = 0, 1, ..., M-1. Converting from RGB space to YCbCr space yields the grayscale feature Gray = 0.229R + 0.587G + 0.114B. Its gradient feature is the first derivative of the image, Gray. x (x,y),Gray x (x,y), its second derivative is Its spatial information is the distance between the pixel and the center of the entire image region, that is... Therefore, the associated 6-dimensional feature vector of the image is:

[0014]

[0015] Image covariance features C 6×6 It is the statistical information of equation (1), that is:

[0016]

[0017] Where μ is the mean vector of the feature vector f(x,y);

[0018] The covariance matrix is ​​a symmetric positive definite matrix that does not belong to Euclidean space but to a Lie group manifold M, with respect to any point C. i For any M, there exists a Lie algebra space S. C C is mapped through exponential mapping i Tangent vector V i ∈S C Mapped to a point C on a Lie group manifold M i Starting from a geodesic line of equal length and in the same direction, that is:

[0019] C′ i =exp Ci (V i ) = C i 1 / 2 exp(C i -1 / 2 V i C i -1 / 2 C i 1 / 2 (3)

[0020] There also exists an inverse mapping, namely a logarithmic mapping:

[0021]

[0022] Among them, V i That is, point C of the Lie group. i The corresponding Lie algebra space point.

[0023] Further, in step S4, the clustering of Lie algebra space points of all busbar laser shing image samples in the Lie algebra training point set using the Mean-Shift algorithm in the Lie algebra space also includes:

[0024] Suppose we have a set of Lie group data points C i , where C i ∈M, i=1,...,n, and the initial position of the iteration is C;

[0025] Step S41: First, store all Lie group data points C on the Lie group manifold M. iMultiplying C by its inverse matrix on the left transforms the initial position point C into the identity element E. Taking the logarithmic mapping then yields the Lie algebraic space point V. i The Lie algebra space point V corresponding to the identity element E;

[0026] Step S42: The Mean-Shift vector form of the Lie algebra space point V corresponding to the identity element E is:

[0027]

[0028] Where g(V) is the negative derivative of the kernel function K(V), h is the bandwidth, and the Mean-Shift vector always points in the direction of the maximum increase in probability density;

[0029] Step S43: Use the exponential mapping to map the Lie algebra space points back to the Lie group data points, and obtain the matrix transformation distance on the Lie group manifold as:

[0030] ΔE=exp(m h (V)) (6)

[0031] Step S44: Assign the new matrix to the original matrix, and move the initial position point C along the Mean-Shift vector to obtain the updated point on the Lie group manifold:

[0032] C′=C+Cexp(m h (V)) (7)

[0033] Step S45: Repeat steps S41, S42, S43, and S44 until ||m h (V)||≤ε, where ε is a set threshold, the loop ends, and the convergence value C is obtained. * .

[0034] Furthermore, in step S4, obtaining the center point of each category of Lie algebra space point also includes:

[0035] If the convergence value C of the Lie algebraic space points of multiple busbar laser cascade image samples * If the points are the same or the distance is less than the set threshold ε, then the Lie algebra space points of these busbar laser overlay image samples are clustered into the same class;

[0036] By calculating the average value of the same type of Lie algebra space points, the Lie algebra center points of these busbar laser slab welding image samples are obtained, thus completing the covariance matrix clustering process on the Lie group manifold.

[0037] Furthermore, step S6 also includes:

[0038]

[0039] Among them, i *This indicates the defect category of the laser overlay image of the busbar to be inspected. V represents the Lie algebra center of the i-th defect category. x Let c represent the Lie algebraic space point of the laser overlay image of the busbar to be inspected, c be the total number of defect categories, and arg be the logical NOT sign function.

[0040] As a second aspect of the present invention, a laser overlay defect detection system for busbars of solar panels is provided, comprising:

[0041] The acquisition module is used to acquire image samples of busbar laser overlay welding and perform image preprocessing;

[0042] The labeling module is used to label the preprocessed busbar laser lap welding image samples, select the busbar laser lap welding image samples with defects, and register the defect categories to obtain the training set;

[0043] The first constitutive module is used to extract the covariance matrix of each bus laser slab welding image sample in the training set to form a first Lie group manifold, and to map the first Lie group manifold to the Lie algebra space to form a Lie algebra training point set.

[0044] The clustering module is used to cluster the Lie algebra space points of all busbar laser slab welding image samples in the Lie algebra training point set using the Mean-Shift algorithm, and obtain the center point of each category of Lie algebra space points;

[0045] The second component module is used to acquire the laser slab image of the busbar to be tested, extract the covariance matrix of the laser slab image of the busbar to be tested to form a second Lie group manifold, and map the second Lie group manifold to the Lie algebra space to form a Lie algebra test point set, wherein the Lie algebra test point set includes the Lie algebra test points of all the laser slab images of the busbar to be tested.

[0046] The calculation module is used to calculate the Euclidean distance between each Lie algebra test point in the Lie algebra test point set and the center point of each category of Lie algebra space point, and to determine the category of the Lie algebra space point center point with the shortest Euclidean distance as the category of the laser overlay image of the busbar to be detected.

[0047] The present invention provides a method and system for detecting defects in laser-laminated busbars of solar panels, which has the following advantages: The method employs a Lie group-based manifold-based approach. Firstly, it uses an image covariance matrix to fuse image feature information, resulting in more features, fewer dimensions, and stronger robustness. Secondly, it utilizes the correspondence between Lie groups and Lie algebras to map the covariance matrix to a set of Lie algebra points, thus solving the problem of non-Euclidean distance measurement of the covariance matrix. Thirdly, the MeanShift algorithm on the Lie group manifold is a density-iterative algorithm, belonging to unsupervised learning classification methods. It can cluster sampling points of any dimension and distribution without pre-defining the shape or number of categories and is insensitive to noisy data. Finally, the accuracy of category recognition for the image under test is calculated. Compared with traditional methods such as the image covariance matrix Euclidean distance measurement algorithm, the recognition rate of this invention is improved by 3%. The algorithm of this invention has good interpretability and understandability, fewer model parameters, and superior computational performance. Therefore, this invention features high efficiency and good accuracy. Attached Figure Description

[0048] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but do not constitute a limitation thereof.

[0049] Figure 1 The flowchart illustrates the laser overlay defect detection method for busbars in solar panels provided by this invention.

[0050] Figure 2 The diagram showing the correspondence between Lie group manifolds and Lie algebra spaces provided by this invention. Detailed Implementation

[0051] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a laser overlay defect detection method and system for solar panel busbars proposed according to the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0052] This embodiment provides a method for detecting defects in laser overlay welding of busbars in solar panels, such as... Figure 1 As shown, the method for detecting defects in laser overlay welding of busbars in solar panels includes:

[0053] Step S1: Acquire busbar laser overlay image samples and perform image preprocessing;

[0054] Specifically, a laser welding platform for solar panel busbars was built, and camera data was collected from the weld joints under different physical properties, geometric dimensions and process parameters of the materials to obtain a large amount of image sample data, which was then preprocessed.

[0055] Preferably, the image preprocessing in step S1 refers to denoising and resizing the image. In this invention, all busbar laser overlay image samples are set to 256*256 (width*height, unit: pixels).

[0056] Step S2: Manually label the preprocessed busbar laser lap welding image samples, select the busbar laser lap welding image samples with defects, and register the defect categories to obtain the training set and achieve comprehensive division of the training set;

[0057] Preferably, in step S2, the welding defects detected by the present invention include four categories: insufficient tin, poor solder joint, missing solder joint, and partial solder joint. The rest are classified as qualified, and the training set is divided into a total of five categories.

[0058] Step S3: Extract the covariance matrix of each bus laser slab welding image sample in the training set to form a first Lie group manifold, and map the first Lie group manifold to the Lie algebra space to form a Lie algebra training point set;

[0059] Preferably, in step S3, the step of extracting the covariance matrix of each bus laser shing image sample in the training set to form a first Lie group manifold, and mapping the first Lie group manifold to the Lie algebra space to form a Lie algebra training point set, further includes:

[0060] Let I be a busbar laser shing image sample with resolution M×N. Its color features are denoted by R(x,y), G(x,y), and B(x,y), where x = 0, 1, ..., N-1, y = 0, 1, ..., M-1, for example, x = 0, 1, ..., 255, y = 0, 1, ..., 255. Converting from RGB space to YCbCr space yields the grayscale feature Gray = 0.229R + 0.587G + 0.114B. Its gradient feature is the first derivative of the image, Gray. x (x,y),Gray x (x,y), its second derivative is Its spatial information is the distance between the pixel and the center of the entire image region, that is... Therefore, the associated 6-dimensional feature vector of the image is:

[0061]

[0062] Image covariance features C 6×6 It is the statistical information of equation (1), that is:

[0063]

[0064] Where μ is the mean vector of the feature vector f(x,y);

[0065] The covariance matrix is ​​a symmetric positive definite matrix that does not belong to Euclidean space but to a Lie group manifold M, with respect to any point C. i For any M, there exists a Lie algebra space S. C C is mapped through exponential mapping i Place

[0066] i

[0067] Tangent vector V i ∈S C Mapped to a point C on a Lie group manifold M i Starting from a geodesic line of equal length and in the same direction, that is:

[0068]

[0069] There also exists an inverse mapping, namely a logarithmic mapping:

[0070]

[0071] Among them, V i That is, point C of the Lie group. i The corresponding Lie algebra space point.

[0072] Step S4: In the Lie algebra space, the Mean-Shift algorithm is used to cluster the Lie algebra space points of all busbar laser slab welding image samples in the Lie algebra training point set, and the center point of the Lie algebra space points of each defect category is obtained.

[0073] Preferably, Figure 2 The correspondence between Lie group manifolds and Lie algebra spaces is demonstrated. Step S4, which involves clustering the Lie algebra space points of all busbar laser overlay image samples in the Lie algebra training point set using the Mean-Shift algorithm, further includes:

[0074] Suppose we have a set of Lie group data points C i , where C i ∈M, i=1,...,n, and the initial position of the iteration is C;

[0075] Step S41: First, store all Lie group data points C on the Lie group manifold M. i Multiplying C by its inverse matrix on the left transforms the initial position point C into the identity element E. Taking the logarithmic mapping then yields the Lie algebraic space point V. i (i = 1, ..., n) and the Lie algebra space point V corresponding to the identity element E;

[0076] Step S42: The Mean-Shift vector form of the Lie algebra space point V corresponding to the identity element E is:

[0077]

[0078] Where g(V) is the negative derivative of the kernel function K(V), h is the bandwidth, and the Mean-Shift vector always points in the direction of maximum increase in probability density; in this invention, K(V) = exp(-||V|| 2 The bandwidth is set to h = 1.2.

[0079] Step S43: Use the exponential mapping to map the Lie algebra space points back to the Lie group data points, and obtain the matrix transformation distance on the Lie group manifold as:

[0080] ΔE=exp(m h (V)) (6)

[0081] Step S44: Assign the new matrix to the original matrix, and move the initial position point C along the Mean-Shift vector to obtain the updated point on the Lie group manifold:

[0082] C′=C+Cexp(m h (V)) (7)

[0083] Step S45: Repeat steps S41, S42, S43, and S44 until ||m h (V)||≤ε, where ε is a set threshold, the loop ends, and the convergence value C is obtained. * In this invention, the threshold ε = 1.

[0084] Preferably, in step S4, obtaining the center point of each category of Lie algebra space point further includes:

[0085] If the convergence value C of the Lie algebraic space points of multiple busbar laser cascade image samples * If the points are the same or the distance is less than the set threshold ε, then the Lie algebra space points of these busbar laser overlay image samples are clustered into the same class;

[0086] By calculating the average value of the same type of Lie algebra space points, the Lie algebra center points of these busbar laser slab welding image samples are obtained, thus completing the covariance matrix clustering process on the Lie group manifold.

[0087] Step S5: Acquire the laser slab image of the busbar to be tested, and extract the covariance matrix of the laser slab image of the busbar to be tested to form a second Lie group manifold. Map the second Lie group manifold to the Lie algebra space to form a Lie algebra test point set, wherein the Lie algebra test point set includes the Lie algebra test points of all the laser slab images of the busbar to be tested.

[0088] It should be noted that the specific implementation of step S5 is completely consistent with the specific implementation of step S3, which extracts the covariance matrix of the image and maps it to the Lie algebra space. This will not be elaborated further here.

[0089] Step S6: Calculate the Euclidean distance between each Lie algebra test point in the Lie algebra test point set and the center point of each category of Lie algebra space point, and determine the category of the Lie algebra space point center point with the shortest Euclidean distance as the category of the laser overlay image of the busbar to be detected.

[0090] Preferably, step S6 further includes:

[0091]

[0092] Among them, i * This indicates the defect category of the laser overlay image of the busbar to be inspected. V represents the Lie algebra center of the i-th defect category. x Let c represent the Lie algebraic space point of the laser overlay image of the busbar to be inspected, c be the total number of defect categories, and arg be the logical NOT sign function.

[0093] Based on the above method, defect detection is performed on the laser overlay image of the busbar to be inspected. For example, the final output cluster category is insufficient tin defect.

[0094] Table 1 compares the accuracy of the proposed method with the Euclidean distance method. As can be seen from Table 1, the accuracy of the proposed method is significantly higher than that of the traditional detection method, demonstrating a very high accuracy.

[0095] Table 1 compares the accuracy of the method of this invention with that of the Euclidean distance method, showing a significant improvement of more than 3%.

[0096] Table 1

[0097] Defect Category Accuracy of Lie Group Manifold Method Euclidean distance method accuracy Shaoxi 98.61％ 95.55％ cold solder joint 97.46％ 94.44％ half-welding 93.92％ 89.89％ Missing weld 96.32％ 93.22％

[0098] Table 2 compares the computational efficiency of the proposed method with that of the Support Vector Machine (SVM) method. As can be seen from Table 2, the proposed method exhibits significantly higher computational efficiency.

[0099] Table 2 compares the computational efficiency of the method of the present invention with that of Support Vector Machine (SVM), showing a significant improvement of more than double.

[0100] Table 2

[0101] Sample set number Time taken by the Lie group manifold method Time taken by Euclidean distance method 300 2.0s 4.7s 400 2.7s 6.8s 500 3.9s 9.4s 600 5.7s 12.4s

[0102] Therefore, this invention features high accuracy and high efficiency.

[0103] As another embodiment of the present invention, a solar panel busbar laser overlay defect detection system is provided, wherein the solar panel busbar laser overlay defect detection system includes:

[0104] The acquisition module is used to acquire image samples of busbar laser overlay welding and perform image preprocessing;

[0105] The labeling module is used to label the preprocessed busbar laser lap welding image samples, select the busbar laser lap welding image samples with defects, and register the defect categories to obtain the training set;

[0106] The first constitutive module is used to extract the covariance matrix of each bus laser slab welding image sample in the training set to form a first Lie group manifold, and to map the first Lie group manifold to the Lie algebra space to form a Lie algebra training point set.

[0107] The clustering module is used to cluster the Lie algebra space points of all busbar laser slab welding image samples in the Lie algebra training point set using the Mean-Shift algorithm, and obtain the center point of each category of Lie algebra space points;

[0108] The second component module is used to acquire the laser slab image of the busbar to be tested, extract the covariance matrix of the laser slab image of the busbar to be tested to form a second Lie group manifold, and map the second Lie group manifold to the Lie algebra space to form a Lie algebra test point set, wherein the Lie algebra test point set includes the Lie algebra test points of all the laser slab images of the busbar to be tested.

[0109] The calculation module is used to calculate the Euclidean distance between each Lie algebra test point in the Lie algebra test point set and the center point of each category of Lie algebra space point, and to determine the category of the Lie algebra space point center point with the shortest Euclidean distance as the category of the laser overlay image of the busbar to be detected.

[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for detecting defects in laser overlay welding of busbars in solar panels, characterized in that, include: Step S1: Acquire busbar laser overlay image samples and perform image preprocessing; Step S2: Mark the preprocessed busbar laser lap welding image samples, select the busbar laser lap welding image samples with defects, and register the defect categories to obtain the training set; Step S3: Extract the covariance matrix of each bus laser slab welding image sample in the training set to form a first Lie group manifold, and map the first Lie group manifold to the Lie algebra space to form a Lie algebra training point set; Step S4: In the Lie algebra space, the Mean-Shift algorithm is used to cluster the Lie algebra space points of all busbar laser slab welding image samples in the Lie algebra training point set, and the center point of each category of Lie algebra space points is obtained. Step S5: Acquire the laser slab image of the busbar to be tested, and extract the covariance matrix of the laser slab image of the busbar to be tested to form a second Lie group manifold. Map the second Lie group manifold to the Lie algebra space to form a Lie algebra test point set, wherein the Lie algebra test point set includes the Lie algebra test points of all the laser slab images of the busbar to be tested. Step S6: Calculate the Euclidean distance between each Lie algebra test point in the Lie algebra test point set and the center point of each category of Lie algebra space point, and determine the category of the Lie algebra space point center point with the shortest Euclidean distance as the category of the laser overlay image of the busbar to be detected.

2. The method for detecting defects in laser overlay welding of busbars in solar panels according to claim 1, characterized in that, In step S3, extracting the covariance matrix of each bus laser shing image sample in the training set to construct a first Lie group manifold, and mapping the first Lie group manifold to the Lie algebra space to construct a Lie algebra training point set, further includes: Let I be a busbar laser shing image sample with resolution M×N. Its color features are represented by R(x,y), G(x,y), and B(x,y), where x = 0, 1, ..., N-1, y = 0, 1, ..., M-1. Converting from RGB space to YCbCr space yields the grayscale feature Gray = 0.229R + 0.587G + 0.114B. Its gradient feature is the first derivative of the image, Gray. x (x,y),Gray x (x,y), its second derivative is Its spatial information is the distance between the pixel and the center of the entire image region, that is... Therefore, the associated 6-dimensional feature vector of the image is: Image covariance features C 6×6 It is the statistical information of equation (1), that is: Where μ is the mean vector of the feature vector f(x,y); The covariance matrix is ​​a symmetric positive definite matrix that does not belong to Euclidean space but to a Lie group manifold M, with respect to any point C. i For any ∈ M, there exists a Lie algebra space. C is mapped through exponential mapping i Tangent vector Mapped to a point C on a Lie group manifold M i Starting from a geodesic line of equal length and in the same direction, that is: There also exists an inverse mapping, namely a logarithmic mapping: Among them, V i That is, point C of the Lie group. i The corresponding Lie algebra space point.

3. The method for detecting defects in laser overlay welding of busbars in solar panels according to claim 2, characterized in that, In step S4, the clustering of Lie algebra space points of all busbar laser shing image samples in the Lie algebra training point set using the Mean-Shift algorithm further includes: Suppose we have a set of Lie group data points C i , where C i ∈M, i=1,...,n, and the initial position of the iteration is C; Step S41: First, store all Lie group data points C on the Lie group manifold M. i Multiplying C by its inverse matrix on the left transforms the initial position point C into the identity element E. Taking the logarithmic mapping then yields the Lie algebraic space point V. i The Lie algebra space point V corresponding to the identity element E; Step S42: The Mean-Shift vector form of the Lie algebra space point V corresponding to the identity element E is: Where g(V) is the negative derivative of the kernel function K(V), h is the bandwidth, and the Mean-Shift vector always points in the direction of the maximum increase in probability density; Step S43: Use the exponential mapping to map the Lie algebra space points back to the Lie group data points, and obtain the matrix transformation distance on the Lie group manifold as: ΔE=exp(m h (V)) (6) Step S44: Assign the new matrix to the original matrix, and move the initial position point C along the Mean-Shift vector to obtain the updated point on the Lie group manifold: C′=C+Cexp(m h (V)) (7) Step S45: Repeat steps S41, S42, S43, and S44 until ||m h (V)||≤ε, where ε is a set threshold, the loop ends, and the convergence value C is obtained. * .

4. The method for detecting defects in laser overlay welding of busbars in solar panels according to claim 3, characterized in that, In step S4, obtaining the center point of each category of Lie algebra space point further includes: If the convergence value C of the Lie algebraic space points of multiple busbar laser cascade image samples * If the points are the same or the distance is less than the set threshold ε, then the Lie algebra space points of these busbar laser overlay image samples are clustered into the same class; By calculating the average value of the same type of Lie algebra space points, the Lie algebra center points of these busbar laser slab welding image samples are obtained, thus completing the covariance matrix clustering process on the Lie group manifold.

5. The method for detecting defects in laser overlay welding of busbars in solar panels according to claim 1, characterized in that, Step S6 further includes: Among them, i * This indicates the defect category of the laser overlay image of the busbar to be inspected. V represents the Lie algebra center of the i-th defect category. x Let c represent the Lie algebraic space point of the laser overlay image of the busbar to be inspected, c be the total number of defect categories, and arg be the logical NOT sign function.

6. A solar panel busbar laser overlay defect detection system, used to implement the solar panel busbar laser overlay defect detection method according to any one of claims 1 to 5, characterized in that, The solar panel busbar laser overlay defect detection system includes: The acquisition module is used to acquire image samples of busbar laser overlay welding and perform image preprocessing; The labeling module is used to label the preprocessed busbar laser lap welding image samples, select the busbar laser lap welding image samples with defects, and register the defect categories to obtain the training set; The first constitutive module is used to extract the covariance matrix of each bus laser slab welding image sample in the training set to form a first Lie group manifold, and to map the first Lie group manifold to the Lie algebra space to form a Lie algebra training point set. The clustering module is used to cluster the Lie algebra space points of all busbar laser slab welding image samples in the Lie algebra training point set using the Mean-Shift algorithm, and obtain the center point of each category of Lie algebra space points; The second component module is used to acquire the laser slab image of the busbar to be tested, extract the covariance matrix of the laser slab image of the busbar to be tested to form a second Lie group manifold, and map the second Lie group manifold to the Lie algebra space to form a Lie algebra test point set, wherein the Lie algebra test point set includes the Lie algebra test points of all the laser slab images of the busbar to be tested. The calculation module is used to calculate the Euclidean distance between each Lie algebra test point in the Lie algebra test point set and the center point of each category of Lie algebra space point, and to determine the category of the Lie algebra space point center point with the shortest Euclidean distance as the category of the laser overlay image of the busbar to be detected.

Images