Universal defect point aggregation detection method and system

Abstract

The invention provides a universal defect point aggregation detection method and system, and relates to the field of intelligent manufacturing and artificial intelligence, and the method comprises the steps: carrying out the preprocessing of panel data; performing clustering processing on the preprocessed panel data by using a defect point clustering algorithm to obtain a plurality of defect point clusters; analyzing and judging each defect point cluster from two dimensions of a basic position and a basic shape so as to obtain category labels of the basic position and the basic shape of each defect point cluster; and converting and mapping the category label of the basic position and the basic shape of each defect point cluster to a self-defined defect category name so as to judge the final defect category name of each defect point cluster. According to the method, the defect point aggregation condition is analyzed from the two independent dimensions of the basic position and the basic shape, and the problems that in the prior art, industrial panel defect point aggregation detection cannot be intelligently achieved, and defect point aggregation detection lacks universality, generalization, robustness and flexibility are solved.

Description

Technical Field

[0001] The present invention relates to the fields of intelligent manufacturing and artificial intelligence, and more particularly, to a general method and system for detecting defect point aggregation. Background Art

[0002] Defect point aggregation on industrial panels refers to the fact that during the production and processing of industrial panels, defect points of uneven sizes will appear on the panel surface. These defect points usually have a certain regional tendency, that is, they concentrate and occur in a certain area. During the production and processing of industrial panels, if the aggregation of defect points on the industrial panel cannot be accurately observed, it is easy to cause missed inspections and increase the mass production risk of defective products. At the same time, the aggregation situation and aggregation type of these defect points are often closely related to the process. It is also of great significance for tracing and analyzing the root cause of defects and adjusting the process. Therefore, accurately detecting the aggregation area and aggregation type of defect points on industrial panels and realizing the aggregation analysis of defect point images (Map diagrams) is the core problem to be solved urgently in this field.

[0003] Currently, the analysis of defect point aggregation on industrial panels mainly includes two tasks: the detection and recognition of the defect point aggregation area and the determination of the defect point aggregation type.

[0004] Existing technical means mainly include defect clustering algorithms based on machine vision and traditional defect clustering algorithms. For defect clustering algorithms based on machine vision, first, a defect point image needs to be made, then the defect point area is identified from the defect point image, and the location of the defect point area is carried out. Finally, a classification model is trained based on the defect category name labels defined by the user to classify the aggregation situation of defect points. For traditional defect clustering algorithms, first, clustering is directly performed based on the coordinates of defect points to directly identify multiple aggregation clusters, and then the aggregation situation of defect points is further identified by means of constructing a classification model or constructing rules, etc.

[0005] There are still some problems with existing technical means, which are specifically manifested as follows:

[0006] 1. Lack of generalization, universality, and robustness: The distribution of defect points on industrial panels is very complex. Industrial panels have clustering clusters of different shapes, sizes, and surface densities. Existing technologies often detect specific defect aggregation categories, and the determination effect for other defect aggregation categories is not good, with weak generalization, universality, and robustness.

[0007] 2. Lack of flexibility: Existing technologies usually involve complex algorithms and data processing processes, lacking the ability to directly operate on data. In particular, it is impossible to directly monitor and intervene in the detection and determination processes during the detection process, with poor flexibility.

[0008] 3. Subjectivity is too strong and the detection and determination cost is too high: Some existing technologies rely on experienced determination personnel, lack objective, unified, and clear determination criteria, and the cost of determination results is too high to achieve automation. Summary of the Invention

[0009] The present invention provides a general method and system for detecting the clustering of defect points, which solves the problems that the existing technology cannot intelligently realize the detection of the clustering of defect points on industrial panels, and the detection of the clustering of defect points lacks generality, generalization, robustness, and flexibility.

[0010] In a first aspect, an embodiment of the present invention provides a general method for detecting the clustering of defect points, and the method includes the following processes:

[0011] Obtain panel data and preprocess the panel data;

[0012] Use a defect point clustering algorithm to cluster the preprocessed panel data to obtain several defect point clusters;

[0013] Analyze and judge each defect point cluster from two dimensions of the basic position and the basic shape to obtain the category labels of the basic position and the basic shape of each defect point cluster;

[0014] Convert and map the category labels of the basic position and the basic shape of each defect point cluster to a custom defect category name to determine the final defect category name of each defect point cluster.

[0015] In the above embodiment, the present invention maps the basic features (position and shape) to the defect types defined by the factory. This design can adapt to the naming habits and classification requirements of different factories, and can handle the situation of semantic overlap, with strong flexibility.

[0016] As some optional embodiments of the present application, the process of using a defect point clustering algorithm to cluster the preprocessed panel data to obtain several defect point clusters is as follows:

[0017] Perform neighborhood screening processing on the panel data based on a preset neighborhood radius and minimum neighborhood points to obtain several core objects;

[0018] Based on the principles of density reachability and density connection, recursively cluster and divide the panel data within the neighborhood radius of the core objects to obtain several defect point clusters.

[0019] In the above embodiment, the present invention clusters defect points into several defect point clusters through a density-based clustering algorithm, which is convenient for later detection of the clustering of defect points.

[0020] In some alternative embodiments of the present application, the basic positions of the defect point clusters include the central position, the corner positions, the edge positions, and no specific position; among them, the corner positions include the upper left corner position, the lower left corner position, the upper right corner position, and the lower right corner position; the edge positions include the upper edge position, the lower edge position, the left edge position, and the right edge position.

[0021] In some alternative embodiments of the present application, the basic shapes of the defect point clusters include linear shapes, cluster shapes, and planar shapes; among them, the linear shapes include horizontal lines, vertical lines, and oblique lines.

[0022] In some alternative embodiments of the present application, the process of analyzing and judging each defect point cluster from two dimensions of the basic position and the basic shape to obtain the category labels of the basic position and the basic shape of each defect point cluster is as follows:

[0023] Use a rule-based determination method to classify the positions of each defect point cluster to obtain the position category names of each defect point cluster;

[0024] Use a geometric feature-based determination method to classify the shapes of each defect point cluster to obtain the shape category names of each defect point cluster.

[0025] In the above embodiments, the present invention abstracts and generalizes the judgment requirements of different factories and different users in industrial inspection, so as to uniformly process the judgment requirements of different factories and different users, effectively improving the generalization, universality, and robustness of the defect point aggregation detection.

[0026] In some alternative embodiments of the present application, the process of using a rule-based determination method to classify the positions of each defect point cluster to obtain the position category names of each defect point cluster is as follows:

[0027] Obtain the cluster center of each defect point cluster and calculate the Euclidean distance from the cluster center to the center of the panel;

[0028] If the Euclidean distance from the cluster center to the center of the panel is less than the center threshold, it is determined that the position category of the defect point cluster is the central position;

[0029] If the central position condition is not met, calculate the Euclidean distance from the cluster center to the four corners of the panel. If there is a case where the Euclidean distance from the cluster center to each corner of the panel is less than the corner threshold, it is determined that the position category of the defect point cluster is the corresponding corner position;

[0030] If the corner position condition is not met, calculate the Euclidean distance from the cluster center to the four sides of the panel. If there is a case where the Euclidean distance from the cluster center to the four sides of the panel is less than the corner threshold, it is determined that the position category of the defect point cluster is the corresponding edge position;

[0031] If none of the above conditions are met, it is determined that the position category of the defect point cluster is no specific position.

[0032] As some alternative embodiments of the present application, the process of classifying the shape of each defect point cluster by using a geometric feature-based determination method to obtain the shape category name of each defect point cluster is as follows:

[0033] Obtain the minimum bounding rectangle of each defect point cluster and calculate the ratio of the long side to the short side of the minimum bounding rectangle;

[0034] If the ratio of the long side to the short side of the minimum bounding rectangle is greater than the linear threshold, it is determined that the shape category of the defect point cluster is a linear shape;

[0035] If the linear shape condition is not met, calculate the area of the minimum bounding rectangle. If the area of the minimum bounding rectangle is greater than the area threshold, it is determined that the shape category of the defect point cluster is a planar shape; if the area of the minimum bounding rectangle is less than the area threshold, it is determined that the shape category of the defect point cluster is a cluster shape.

[0036] In a second aspect, the present invention provides a general defect point aggregation detection system, which includes:

[0037] A panel data acquisition unit, which is used to acquire panel data and preprocess the panel data;

[0038] A defect point clustering unit, which uses a defect point clustering algorithm to cluster the preprocessed panel data to obtain several defect point clusters;

[0039] A category label analysis unit, which analyzes and judges each defect point cluster from two dimensions of basic position and basic shape to obtain the category labels of the basic position and basic shape of each defect point cluster;

[0040] A category label mapping unit, which is used to convert and map the category labels of the basic position and basic shape of each defect point cluster to a custom defect category name to determine the final defect category name of each defect point cluster.

[0041] In a third aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the general defect point aggregation detection method.

[0042] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the general defect point aggregation detection method.

[0043] The beneficial effects of the present invention are as follows:

[0044] (1) The present invention decouples the analysis framework of two independent dimensions of the basic position and the basic shape. This decoupling design makes the analysis process clearer and is convenient for expansion. Users can independently adjust or optimize the determination method of any dimension without affecting the other dimension.

[0045] (2) The present invention abstracts and generalizes the judgment requirements of different factories and different users in industrial inspection, so as to uniformly process the judgment requirements of different factories and different users, effectively improving the generalization, universality and robustness of the defect point clustering detection.

[0046] (3) The present invention can flexibly customize various defect types according to the business analysis requirements, and can also adjust the parameters of the model to monitor and adjust the model judgment process, effectively improving the flexibility of the defect point clustering detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 is the overall flowchart of the method described in the embodiments of the present invention;

[0049] Figure 2 is the comprehensive analysis framework diagram described in the embodiments of the present invention;

[0050] Figure 3 is the defect type mapping determination flowchart described in the embodiments of the present invention;

[0051] Figure 4 is the basic position determination flowchart described in the embodiments of the present invention;

[0052] Figure 5 is the basic shape determination flowchart described in the embodiments of the present invention;

[0053] Figure 6 is the schematic diagram of the panel data described in the embodiments of the present invention;

[0054] Figure 7 is the schematic diagram of the DBSCAN clustering result of the panel data described in the embodiments of the present invention;

[0055] Figure 8It is a schematic diagram of the basic position judgment result described in the embodiment of the present invention;

[0056] Figure 9 It is a schematic diagram of the basic shape judgment result described in the embodiment of the present invention. Detailed implementation manners

[0057] It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0058] To solve the problems that the prior art cannot intelligently implement the clustering detection of defect points on industrial panels, and the clustering detection of defect points lacks generality, generalization, robustness, and flexibility. The embodiments of the present invention provide a general method for clustering detection of defect points. Please refer to Figure 1 , Figure 1 For the overall flowchart of the method, the method process is as follows:

[0059] (1) Obtain panel data and preprocess the panel data.

[0060] In the embodiment of the present invention, the panel data includes but is not limited to: the coordinate information of the defect points on the panel surface and the size information of the panel (such as length, width, and center point coordinates, etc.).

[0061] Specifically, the process of preprocessing the panel data is as follows:

[0062] (1.1) Data deduplication. Since manual recording operations may cause some defect point coordinates to be recorded repeatedly, redundant points can be eliminated by calculating the distance between two defect points. Among them, the value of the distance detection can be set according to business requirements for different problems, and the embodiments of the present invention do not limit this. Preferably, in the embodiment of the present invention, the defect points with a distance between two points less than 1 μm are selected as duplicate points.

[0063] (1.2) Data removal. Since during the cutting process of the industrial panel, the separation positions of the upper, lower, left, and right have been determined, there are certain gaps between different small pieces (i.e., panels in the industrial panel). In this case, if there are defect points in the gap area of the small pieces, it will not affect the quality of the entire panel. Therefore, the relevant data of the defect points that have no impact on the panel quality can be removed.

[0064] (2) The comprehensive analysis framework uses a defect point clustering algorithm to cluster the preprocessed panel data to obtain several defect point clusters. Please refer to Figure 2 , Figure 2 For the comprehensive analysis framework diagram.

[0065] Specifically, the defect point clustering algorithm includes, but is not limited to, DBSCAN algorithm, K-means algorithm, OPTICS algorithm, etc.; preferably, the embodiments of the present invention select to use the DBSCAN algorithm for clustering processing.

[0066] In the embodiments of the present invention, the process of using the DBSCAN algorithm to cluster the preprocessed panel data is as follows:

[0067] (2.1) Perform neighborhood screening processing on the panel data based on a preset neighborhood radius and minimum neighborhood points to obtain a number of core objects.

[0068] (2.2) Based on the principles of density reachability and density connectivity, use a recursive method to cluster and divide the panel data within the neighborhood radius of the core objects to obtain a number of defect point clusters. At the same time, for those defect points that do not meet the core object conditions and cannot form a density reachable and density connected relationship with any cluster, they are marked as noise points.

[0069] (3) Analyze and judge each defect point cluster from two dimensions of basic position and basic shape to obtain the category labels of the basic position and basic shape of each defect point cluster.

[0070] Specifically, in the traditional defect point aggregation detection process, it is often necessary to develop a specific aggregation type recognition algorithm model according to the labeled data of specific defect types by users for the determination of specific aggregation types. However, different factories and different users often have different definitions of aggregation types, resulting in that when changing a set of defect type names, it is necessary to relabel the panel data and retrain the aggregation type recognition algorithm model, with a long model training cycle and high cost.

[0071] Therefore, the embodiments of the present invention analyze and judge each defect point cluster from two dimensions of basic position and basic shape, aiming to be able to flexibly respond to various defect recognition types defined by users, improve the generality and generalization of defect point aggregation recognition, and thus improve the efficiency and intelligent level of defect point aggregation analysis.

[0072] Specifically, the basic positions of the defect point clusters include, but are not limited to, the center position, corner positions, edge positions, and no specific position; among them, the corner positions include the upper left corner position, lower left corner position, upper right corner position, and lower right corner position; the edge positions include the upper edge position, lower edge position, left edge position, and right edge position.

[0073] Specifically, the basic shapes of the defect point clusters include linear shapes, cluster shapes (relatively concentrated in a certain position area of the panel), and planar shapes (relatively evenly and widely spread on the panel); among them, the linear shapes include horizontal lines, vertical lines, and diagonal lines.

[0074] In the embodiments of the present invention, the core of analyzing and judging each defect point cluster from two dimensions of the basic position and the basic shape lies in that regardless of what the defect aggregation name defined by the user is, first, the aggregation situation of the defect point cluster is analyzed and judged from two independent and general dimensions of the basic position and the basic shape; then it is correspondingly converted to the defect type name defined by the user. The specific categories included in the above basic position and basic shape are only the summary based on a large amount of practical experience in the embodiments of the present invention. In practice, the user can follow these two analysis dimensions and expand the categories of the basic position and the basic shape according to the actual business needs (especially for the category division of the basic shape, it can be further refined according to business needs). However, it should be noted that there should be no semantic overlap between different categories of the basic shape and the basic position, and the independence of the basic categories needs to be ensured.

[0075] It should be noted that the specific determination methods for the basic position and the basic shape of each defect point cluster include but are not limited to: ① rule-based determination method; ② statistic-based determination method; ③ supervised classification method based on machine learning (such as decision tree, random forest, support vector machine, neural network, etc.).

[0076] Generally speaking, among these classification methods, the method based on machine learning often depends on the user to manually label the categories of the defect point clusters in advance, and is generally applicable to the situation where the number of defect samples is large. The rule-based determination method often needs to combine the understanding of different categories to give relatively clear judgment rules. The statistic-based determination method often requires the user to provide a clear judgment threshold. In practice, the user can flexibly select a suitable category determination method for the defect point cluster in combination with the actual data volume, data characteristics, and business characteristics, etc.

[0077] In the embodiments of the present invention, the process of analyzing and judging each defect point cluster from two dimensions of the basic position and the basic shape is as follows:

[0078] (3.1) Use the rule-based determination method to classify the position of each defect point cluster to obtain the position category name of each defect point cluster.

[0079] (3.2) Use the geometric feature-based determination method to classify the shape of each defect point cluster to obtain the shape category name of each defect point cluster.

[0080] Please refer to Figure 4 , Figure 4 which is the flowchart of the basic position determination; specifically, the process of using the rule-based determination method to classify the position of each defect point cluster is as follows:

[0081] (3.11) Obtain the cluster center of each defect point cluster and calculate the Euclidean distance from the cluster center to the center of the panel.

[0082] (3.12) If the Euclidean distance from the cluster center to the center of the panel is less than the center threshold, then determine that the position category of the defect point cluster is the center position.

[0083] (3.13) If the center position condition is not satisfied, then calculate the Euclidean distances from the cluster center to the four corners of the panel. If there is a case where the Euclidean distance from the cluster center to each corner of the panel is less than the corner threshold, then determine that the position category of the defect point cluster is the corresponding corner position.

[0084] (3.14) If the corner position condition is not satisfied, then calculate the Euclidean distances from the cluster center to the four sides of the panel. If there is a case where the Euclidean distance from the cluster center to the four sides of the panel is less than the corner threshold, then determine that the position category of the defect point cluster is the corresponding edge position.

[0085] (3.15) If none of the above conditions are satisfied, then determine that the position category of the defect point cluster is no specific position.

[0086] Please refer to Figure 5 , Figure 5 for the basic shape determination flowchart; specifically, the process of classifying the shape of each defect point cluster using a determination method based on geometric features is as follows:

[0087] (3.21) Obtain the minimum bounding rectangle of each defect point cluster and calculate the ratio of the long side to the short side of the minimum bounding rectangle.

[0088] (3.22) If the ratio of the long side to the short side of the minimum bounding rectangle is greater than the linear threshold, then determine that the shape category of the defect point cluster is a linear shape.

[0089] (3.23) If the linear shape condition is not satisfied, then calculate the area of the minimum bounding rectangle. If the area of the minimum bounding rectangle is greater than the area threshold, then determine that the shape category of the defect point cluster is a planar shape; if the area of the minimum bounding rectangle is less than the area threshold, then determine that the shape category of the defect point cluster is a cluster shape.

[0090] (4) Map the category labels of the basic position and basic shape of each defect point cluster to custom defect category names to determine the final defect category name of each defect point cluster. Please refer to Figure 3 , Figure 3 for the defect type mapping determination flowchart.

[0091] It should be noted specifically that:

[0092] ①Multiple aggregation types may exist simultaneously on a panel. For example, a panel may simultaneously have the characteristics of linear aggregation and central aggregation. In this case, the embodiments of the present invention will give multiple determination results, that is, multiple category labels.

[0093] ②The embodiments of the present invention provide great flexibility. Users can freely define the names of defect point aggregation types according to business requirements, and semantic overlaps between different types are allowed. This design is based on the consideration of actual factory analysis requirements, because in actual applications, according to the user's focus, there may indeed be semantic overlaps between different defect category names. The embodiments of the present invention can handle this situation well and do not limit users. For example, users may define linear aggregation and central aggregation simultaneously. Obviously, there is a semantic overlap between these two names (because linear aggregation may also appear in the middle). Then, when linear aggregation appears in the central area of a panel, the determination result of the aggregation type of the panel will include both the type name labels of linear aggregation and central aggregation. The embodiments of the present invention can handle this situation well.

[0094] Specifically, users can define the mapping rules between the categories of basic positions and basic shapes and the user-defined defect category names according to their own business characteristics and business understandings. The embodiments of the present invention do not make restrictions. At the same time, users can continue to flexibly adjust this mapping correspondence in the later stage; the mapping rules can also be dynamically updated and managed through various automated methods such as Web interfaces and configuration files.

[0095] (5) Continuously optimize the accuracy of model determination, and deploy the model to the defect detection system of the factory to achieve the aggregation detection of the panel.

[0096] Specifically, compare the determination results given by the model with the aggregation type labels marked by users, calculate the accuracy rate of the determination results, and adjust the model parameters according to the accuracy rate results to continuously optimize the model results. That is, when the determination accuracy rate reaches the expected requirements, the model can be officially deployed to the defect detection system of the factory and adjusted in a timely manner according to user feedback, so as to achieve stable and reliable detection of the aggregation defects of the panel.

[0097] To more clearly show the purpose, technical solution and advantages of the present invention, the following takes the aggregation detection of defect points in the production and manufacturing process of glass panels as an example, and elaborates on the present invention in detail in combination with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0098] There are two prominent features of the defect points on the glass panel: ① Uneven distribution. The defect points on the glass panel may be unevenly distributed in different areas of the panel. The shape of the aggregation area may also be irregular, without an obvious geometric pattern. ② Spatial correlation. The distribution of defect points on the panel may show a tendency of aggregation or clustering. This spatial correlation is often related to the manufacturing process of the panel, material properties, or other factors. Therefore, the aggregation type of defect points is an important reference basis for assisting in analyzing the root causes of defects and improving the process during panel processing.

[0099] To address the limitations of traditional clustering analysis, the present invention proposes a method for detecting the clustering of defect points based on two independent dimensions of basic position and basic shape. First, perform clustering judgment on the defect points; then analyze the position and shape characteristics of each defect point cluster, and map the analysis results to the defect types defined by the factory, so as to quickly and accurately obtain the judgment result of the clustering situation of panel defect points.

[0100] Specifically, the detailed steps of the general method for detecting the clustering of defect points are as follows:

[0101] S1: Obtain the panel data and preprocess the panel data. Please refer to Figure 6 , Figure 6 As shown in the schematic diagram of the panel data, first calculate the distance between two defect points based on the COOR_X_GLS and COOR_Y_GLS fields as the coordinate points X and Y. If the distance is less than 1um, remove the defect point with the larger index among the two defect points; then remove the coordinates of the defect points that have no effect on the panel in the data according to the encoding rule of the last four characters of the PNL_ID field, which is the ID of each small piece. Specifically, according to the encoding business rules of the panel industry, when the last four characters of the PNL_ID field are all X, it is considered that the defect point is at the cutting gap, so remove the defect point from the panel data.

[0102] S2: Use the DBSCAN algorithm to cluster the preprocessed panel data to obtain several defect point clusters. Among them, the neighborhood radius of the DBSCAN algorithm is set to 5 mm, and the minimum number of neighborhood points is set to 10. For example, after a certain panel is processed by the DBSCAN algorithm, the defect points on the panel surface are clustered into 3 defect point clusters. Please refer to Figure 7 , Figure 7 As shown in the schematic diagram of the DBSCAN clustering result of the panel data of a certain panel.

[0103] S3: Analyze and judge each defect point cluster from two dimensions of basic position and basic shape to obtain two category labels of the basic position and basic shape of each defect point cluster. Please refer to Table 1, where Table 1 is the definition of basic position and basic shape.

[0104]

[0105] Table 1 shows the definitions of the basic positions and basic shapes.

[0106] It should be noted that the categories included in the above basic positions and basic shapes are only a summary based on a large amount of practical experience in the embodiments of the present invention. In practice, users can follow these two analysis dimensions and expand the categories of basic positions and basic shapes according to actual business needs.

[0107] Specifically, the embodiments of the present invention do not limit the specific position judgment and shape judgment algorithms. Users can flexibly select appropriate judgment methods according to actual business needs, including but not limited to: rule-based judgment methods, statistics-based judgment methods, machine learning-based classification methods (such as decision trees, random forests, support vector machines, neural networks, etc.).

[0108] The embodiments of the present invention select a rule-based judgment method based on geometric features, mainly considering the following: ① Reducing data dependence: Geometric feature rules do not require a large amount of labeled training data, only a small number of samples for verification; ② Strong parameter interpretability: The parameters of geometric feature rules have clear physical meanings, which are convenient for engineers to understand and adjust; ③ Flexible and adjustable: Each threshold parameter can be flexibly adjusted according to actual needs without retraining the model; ④ Small computational overhead: The calculation is relatively simple, the computational resource requirements are low, and it is suitable for real-time operation; ⑤ Good maintainability: The rules are clear and the logic is clear, which is convenient for later maintenance and upgrade.

[0109] Specifically, the determination process of the basic position is as follows:

[0110] S311: Obtain the cluster center coordinates (x c , y c ) of each defect point cluster, and calculate the Euclidean distance from the cluster center to the center of the panel.

[0111] Among them, (x i , y i ) is the coordinate of the i-th defect point, and n is the number of points in the defect point cluster.

[0112] S312: If the Euclidean distance from the cluster center to the center of the panel is less than the center threshold, it is determined that the position category of the defect point cluster is the center position.

[0113] S313: If the center position condition is not satisfied, calculate the Euclidean distance from the cluster center to the four corners of the panel. If there is a case where the Euclidean distance from the cluster center to each corner of the panel is less than the corner threshold, it is determined that the position category of the defect point cluster is the corresponding (i.e., upper left, lower left, upper right, lower right) corner position.

[0114] S314: If the corner position condition is not satisfied, calculate the Euclidean distances from the cluster center to the four sides of the panel. If there is a case where the Euclidean distance from the cluster center to any of the four sides of the panel is less than the corner threshold, then determine that the position category of the defect point cluster is the corresponding (i.e., upper, lower, left, or right) edge position.

[0115] S315: If none of the above conditions are satisfied, then determine that the position category of the defect point cluster is no specific position.

[0116] It should be noted that the three key threshold parameters (center distance threshold, corner distance threshold, and edge distance threshold) in this embodiment can be adjusted according to actual business requirements. In actual applications, appropriate thresholds can be determined by analyzing historical data or expert experience.

[0117] Based on the above basic position determination process, the basic positions of the 3 defect point clusters in step S2 can be judged. For the results, please refer to Figure 8 , Figure 8 , which is the schematic diagram of the basic position judgment result.

[0118] Specifically, the basic shape determination process is as follows:

[0119] S321: Obtain the minimum bounding rectangle of each defect point cluster, and calculate the ratio R of the long side to the short side of the minimum bounding rectangle.

[0120]

[0121] Among them, L max and L min are the lengths of the long side and the short side of the minimum bounding rectangle respectively. In actual implementation, the minimum bounding rectangle can be directly calculated using a computer vision library.

[0122] S322: If the ratio of the long side to the short side of the minimum bounding rectangle is greater than the linear threshold, then determine that the shape category of the defect point cluster is a linear shape. When the ratio R is greater than the threshold (recommended value: 3.0), it is determined to be linear.

[0123] Specifically, for the defect point cluster determined to be linear, first calculate the absolute value of the main direction slope through principal component analysis (PCA), and then judge the linear direction according to the threshold. That is, first centralize the point set and calculate the covariance matrix; then calculate the eigenvectors of the covariance matrix, and the eigenvector corresponding to the largest eigenvalue is the main direction. If this eigenvector is set as (v x , v y ), then the absolute value of the slope |k| is:

[0124]

[0125] S323: If the linear shape condition is not satisfied, calculate the area of the minimum bounding rectangle. If the area of the minimum bounding rectangle is greater than the area threshold, determine that the shape category of the defect point cluster is planar; if the area of the minimum bounding rectangle is less than the area threshold, determine that the shape category of the defect point cluster is globular.

[0126] Specifically, use the convex hull algorithm to calculate the area of the minimum bounding rectangle. Assume that the convex hull vertices are in counterclockwise order as (x 1 , y 1 ), (x 2 , y 2 ),...,(x n , y n ), then the area A is:

[0127]

[0128] where n is the number of convex hull vertices, (x i , y i ) is the coordinate of the i-th vertex, and |·| represents taking the absolute value.

[0129] Specifically, to eliminate the influence of the panel size on the area calculation, it is recommended to normalize the area:

[0130]

[0131] where W is the panel width, H is the panel height, A normalized is the normalized area, and the value range is [0, 1]. In this way, the area threshold A threshold should also be a normalized value (recommended value: 0.25, that is, 25% of the panel area).

[0132] It should be noted that the threshold parameters in this embodiment can be adjusted according to actual business requirements. The above determination method based on geometric features is only one of many feasible solutions. In practical applications, users can choose a deep learning model to directly learn features from the original coordinate data; or use traditional machine learning methods (such as SVM, random forest, etc.); or design other rule-based determination methods; or combine multiple methods. The choice of determination method depends on factors such as data characteristics, computing resources, and accuracy requirements.

[0133] Specifically, based on the above basic shape determination process, the basic shapes of the 3 defect point clusters in step S2 can be judged. The results can be referred to Figure 9 , Figure 9 which is the schematic diagram of the basic shape judgment result.

[0134] S4: Map and transform the class labels of the basic positions and basic shapes of each cluster of defect points into custom defect class names to determine the final defect class names for each cluster of defect points.

[0135] The embodiments of the present invention provide a flexible mapping mechanism that can transform and map the combination of the basic positions and basic shapes of each cluster of defect points obtained in step S3 to the names of various defect types customized by the factory. For example, a certain factory has customized the following defect class names, and their corresponding relationships are shown in Table 2.

[0136] Defect aggregation type name Basic position requirement Basic shape requirement Cluster quantity requirement Linear aggregation Any position Linear ≥1 Corner aggregation Corner position Mass-like ≥1 Central aggregation Center position Any shape ≥1 Scattered aggregation Any position Planar ≥2

[0137] Table 2 is the defect aggregation types customized by the factory

[0138] The detailed explanations of the above table are as follows:

[0139] ① When there is a linear cluster at any position, it is determined as linear aggregation;

[0140] ② When there is a cluster in the shape of a mass at the corner position, it is determined as corner mass aggregation;

[0141] ③ When there is a cluster at the central position, it is determined as central aggregation, regardless of the shape of the cluster;

[0142] ④ When there are two or more planar clusters at any position, it is determined as scattered aggregation.

[0143] ⑤ When none of the above conditions are met, but there is at least one cluster, it is determined as other.

[0144] Based on the above rules, the three clusters in step S2 can be judged, and the results are shown in Table 3:

[0145]

[0146] Table 3 is the judgment result of the defect aggregation type

[0147] Specifically, the final determination result of the defect aggregation type given by the panel is: linear aggregation and corner aggregation.

[0148] It should be emphasized that Table 2 given here is only an example. Users can also set other mapping rules according to their own business class labels. For example, define a bull's-eye aggregation as 1 cluster in the shape of a mass, appearing at the central position, and at the same time, there is at least 1 cluster in the shape of a mass at each of the 4 corners (upper left, upper right, lower left, lower right). Regardless of how the aggregation type names customized by the user change, the mapping rules can be determined based on the two dimensions of the basic shape and basic position proposed by the present invention, so as to achieve the generality and flexibility in the mapping of the aggregation type.

[0149] S5: Continuously optimize the accuracy of model determination, and deploy the model to the defect detection system in the factory to achieve the clustering detection of panels.

[0150] After completing the above steps, it is necessary to evaluate the accuracy of the determination results. For example: The user collects the defect data of 1000 glass panels from the production line and analyzes them using the method of the embodiment of the present invention to obtain the determination results of the defect clustering types of 1000 panels. Then, the user compares these results with the actual defect situations and finds that:

[0151] ① Some clusters are far from the center (such as more than 300 mm away from the center), but are determined to be centrally clustered. Then, the center distance threshold for position determination needs to be adjusted from the original 350 mm to 250 mm to improve the accuracy of position determination.

[0152] ② It is found that some horizontal linear clusters (the absolute value of the slope is less than 0.1) are misjudged as diagonal clustering. Then, the slope threshold in shape determination needs to be adjusted, and the determination criterion for horizontal linearity is adjusted from |k| < 0.2 to |k| < 0.1 to improve the accuracy of shape determination.

[0153] Through continuous optimization and adjustment, when the determination accuracy rate is increased from the initial 85% to over 95%, the model can be officially deployed to the defect detection system in the factory.

[0154] Moreover, after the system deployment is completed, a normalized monitoring mechanism needs to be established to regularly evaluate the system operation situation. For example:

[0155] ① Randomly select 100 panel samples from the production line every week, and have experts make manual determinations and compare them with the system determination results to calculate the determination accuracy rate. If the accuracy rate in a certain week is lower than 90%, it is necessary to immediately investigate the reasons and make optimization adjustments;

[0156] ② On a certain day, the factory replaces the equipment or adjusts the process parameters (such as the cutting accuracy is improved from ±0.1 mm to ±0.05 mm), which may cause changes in the distribution characteristics of defects. Then, the determination rules need to be adjusted accordingly (such as adjusting the ∈ parameter of DBSCAN from 5 mm to 4 mm) to ensure the practicality of the system.

[0157] Through the above measures, stable and reliable detection of the clustering defects of panels can be achieved. At the same time, the determination results of the system can also provide an important reference basis for the factory to improve the process and increase the yield.

[0158] In summary, the embodiment of the present invention solves the problems that the prior art cannot intelligently achieve the clustering detection of industrial panel defect points, and the clustering detection of defect points lacks generality, generalization, robustness, and flexibility.

[0159] In addition, in one embodiment, based on the same inventive concept as the foregoing embodiment, an embodiment of the present invention provides a general defect point clustering detection system, which corresponds one-to-one with the method of the first embodiment. The system includes:

[0160] A panel data acquisition unit, which is used to acquire panel data and preprocess the panel data;

[0161] A defect point clustering unit, which uses a defect point clustering algorithm to cluster the preprocessed panel data to obtain several defect point clusters;

[0162] A category label analysis unit, which analyzes and judges each defect point cluster from two dimensions of basic position and basic shape to obtain the category labels of the basic position and basic shape of each defect point cluster;

[0163] A category label mapping unit, which is used to convert and map the category labels of the basic position and basic shape of each defect point cluster to a custom defect category name to determine the final defect category name of each defect point cluster.

[0164] It should be noted that each unit in the defect point clustering detection system in this embodiment corresponds one-to-one with each step in the defect point clustering detection method in the foregoing embodiment. Therefore, the specific implementation manner and the achieved technical effects of this embodiment can refer to the implementation manner of the foregoing defect point clustering detection method, which will not be elaborated here.

[0165] In addition, in one embodiment, the present application further provides a computer device, which includes a processor, a memory, and a computer program stored in the memory. When the computer program is run by the processor, the method in the foregoing embodiment is implemented.

[0166] In addition, in one embodiment, the present application further provides a computer storage medium, on which a computer program is stored. When the computer program is run by the processor, the method in the foregoing embodiment is implemented.

[0167] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disc, or CD-ROM; or it may be various devices including one or any combination of the above memories. The computer may be various computing devices including intelligent terminals and servers.

[0168] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0169] As an example, executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0170] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices located at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0171] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0172] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0173] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a multimedia terminal device (which can be a mobile phone, a computer, a television receiver, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0174] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A general defect point aggregation detection method, characterized in that: The method comprises the following steps: Obtain panel data and preprocess the panel data; The defect point clustering algorithm is used to cluster the pre-processed panel data to obtain several defect point clusters; Analyze and judge each defect point cluster from the two dimensions of basic position and basic shape to obtain the category labels of the basic position and basic shape of each defect point cluster; The category labels of the basic position and basic shape of each defect point cluster are converted and mapped to a custom defect category name to determine the final defect category name of each defect point cluster.

2. A general defect point aggregation detection method according to claim 1, characterized in that: The process of using the defect point clustering algorithm to cluster the preprocessed panel data to obtain several defect point clusters is as follows: Based on the preset neighborhood radius and minimum number of neighborhood points, the panel data is processed by neighborhood screening to obtain several core objects; Based on the principles of density reachability and density connectivity, a recursive method is used to cluster the panel data within the neighborhood radius of the core object to obtain several defect point clusters.

3. A universal defect point aggregation detection method according to claim 1, characterized in that: The basic positions of the defect point cluster include a center position, a corner position, an edge position and an unspecified position; wherein the corner positions include an upper left corner position, a lower left corner position, an upper right corner position and a lower right corner position; and the edge positions include an upper edge position, a lower edge position, a left edge position and a right edge position.

4. A universal defect point aggregation detection method according to claim 1, characterized in that: The basic shapes of the defect point cluster include linear shapes, cluster shapes and surface shapes; wherein the linear shapes include horizontal lines, vertical lines and oblique lines.

5. A universal defect point aggregation detection method according to claim 1, characterized in that: The process of analyzing and judging each defect point cluster from the two dimensions of basic position and basic shape to obtain the category labels of the basic position and basic shape of each defect point cluster is as follows: A rule-based determination method is used to classify the location of each defect point cluster to obtain the location category name of each defect point cluster; The shape of each defect point cluster is classified by using a judgment method based on geometric features to obtain the shape category name of each defect point cluster.

6. A universal defect point aggregation detection method according to claim 5, characterized in that: The process of classifying the location of each defect point cluster by using a rule-based determination method to obtain the location category name of each defect point cluster is as follows: Get the cluster center of each defect point cluster and calculate the Euclidean distance from the cluster center to the panel center; If the Euclidean distance from the cluster center to the panel center is less than the center threshold, the location category of the defect point cluster is determined to be the center location; If the center position condition is not met, the Euclidean distance from the cluster center to the four corners of the panel is calculated. If the Euclidean distance from the cluster center to each corner of the panel is less than the corner threshold, the position category of the defect point cluster is determined to be the corresponding corner position; If the corner position condition is not met, the Euclidean distance from the cluster center to the four edges of the panel is calculated. If the Euclidean distance from the cluster center to the four edges of the panel is less than the corner threshold, the position category of the defect point cluster is determined to be the corresponding edge position; If the above conditions are not met, the location category of the defect point cluster is determined to be no specific location.

7. A universal defect point aggregation detection method according to claim 5, characterized in that: The process of classifying the shape of each defect point cluster by using a judgment method based on geometric features to obtain the shape category name of each defect point cluster is as follows: Obtain the minimum bounding rectangle of each defect point cluster and calculate the ratio of the long side to the short side of the minimum bounding rectangle; If the ratio of the long side to the short side of the minimum circumscribed rectangle is greater than the linear threshold, the shape category of the defect point cluster is determined to be a linear shape; If the linear shape condition is not met, the area of ​​the minimum circumscribed rectangle is calculated. If the area of ​​the minimum circumscribed rectangle is greater than the area threshold, the shape category of the defect point cluster is determined to be planar; if the area of ​​the minimum circumscribed rectangle is less than the area threshold, the shape category of the defect point cluster is determined to be clumpy.

8. A universal defect point aggregation detection system, characterized in that: The system comprises: A panel data acquisition unit, wherein the panel data acquisition unit is used to acquire the panel data and pre-process the panel data; A defect point clustering unit, wherein the defect point clustering unit performs clustering processing on the pre-processed panel data using a defect point clustering algorithm to obtain a number of defect point clusters; A category label analysis unit, wherein the category label analysis unit analyzes and judges each defect point cluster from two dimensions of basic position and basic shape to obtain a category label of the basic position and basic shape of each defect point cluster; The category label mapping unit is used to convert and map the category labels of the basic position and basic shape of each defect point cluster into a custom defect category name to determine the final defect category name of each defect point cluster.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the universal defect point aggregation detection method described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the universal defect point aggregation detection method described in any one of claims 1 to 7 is implemented.

Images