Printed matter visual detection method and device aiming at geometrical morphology deviation

Abstract

The invention discloses a printed matter visual detection method and device aiming at geometric morphology deviation. The method comprises the following steps: S10, carrying out offline modeling; s20, carrying out online detection; the step S10 comprises the following sub-steps: S11, collecting a good product sample; s12, geometric element identification and parameterization extraction are carried out; s13, element sampling representation is carried out; s14, carrying out first-stage statistical morphology modeling; s15, carrying out second-stage statistical morphology modeling; the S20 comprises the following steps: S21, acquiring and preprocessing a test image; s22, geometric elements are rapidly extracted and parameterized; s23, the morphology deviation is calculated; s24, performing defect identification and positioning; and S25, carrying out defect classification and visualization. According to the method, the morphological statistical characteristics of each geometric element under the normal production condition are learned based on multiple good product samples; using the learned statistical morphology model to intelligently compare the to-be-detected products, and accurately identifying geometric deviations exceeding an allowable fluctuation range; providing a defect diagnosis report detailed to specific elements and positions; and the high-speed real-time processing requirement of an industrial field is met while the high precision is ensured.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of printing detection, and particularly relates to a printing visual detection method and device for geometric shape deviation. BACKGROUND

[0002] In high-precision printing industries, such as circuit board printing, RFID antenna printing, precision packaging, and anti-counterfeiting label fields, the geometric shape precision of the printed pattern is one of the core indicators for measuring product quality. Small deviations in geometric parameters such as the concentricity of a circular ring, the uniformity of line width, and the accuracy of pattern angle can lead to product functional failure or a significant decline in appearance quality. Traditional machine vision detection methods face great challenges when dealing with such geometric shape defects.

[0003] Existing machine vision-based geometric detection schemes are mostly based on traditional image processing and rigid template comparison. The most similar implementation schemes mainly include methods based on edge extraction and geometric fitting, methods based on pixel-level difference, and methods based on global deformation field analysis.

[0004] The method based on edge extraction and geometric parameter measurement: This is the most widely used geometric detection scheme in the industry. Its technical route is as follows: First, pre-process the standard template image (such as Gaussian filter denoising) and edge enhancement (such as Canny operator), then obtain the contour pixel point set of the target pattern (such as circular ring, straight line) through edge tracking or contour extraction algorithm (such as Suzuki's topological structure analysis algorithm) . Next, use these point sets for geometric fitting, for example, fit a circle with the least squares method to get the center and radius of the standard circle; or fit a straight line to get the equation of the standard straight line . In online detection, perform the same edge extraction and fitting operations on the image to be detected to obtain the geometric parameters of the detected pattern, such as the center and radius of the detected circle. Finally, judge by calculating the absolute difference of the parameters, for example, the center offset , the radius change . If or If the preset threshold is exceeded, it is determined that the product is a defective product. The advantage of this method is that the principle is intuitive and effective for obvious overall deformation. However, it has the following disadvantages: First, its accuracy is heavily dependent on the accuracy of edge extraction. When there is uneven lighting, blurred edges or noise, the fitting result will fluctuate greatly, leading to unstable measurement. Second, it only focuses on a few global geometric parameters (such as the center of the circle and the radius), and cannot perceive subtle deformation in local areas of the pattern, such as a concave or convex (elliptical) section on the ring, but the overall center and average radius may not change much, resulting in missed detection. Finally, it is very tedious to manually set multiple geometric parameter thresholds, and there is a lack of consideration of the correlation between parameters.

[0005] Method based on image difference and morphological analysis: This method attempts to find morphological differences at the pixel level. After accurate registration of the template and the image to be detected, the difference image is calculated , and the binary difference region is obtained through threshold processing. Then, morphological analysis is performed on the difference region, for example, by analyzing the shape, aspect ratio, direction, etc. of the connected region, to infer whether there is geometric deformation. For example, a long difference region may indicate that the line is thick or thin. This method can capture local deformation to some extent. However, it is still a pixel-level comparison and requires high registration accuracy. Any small registration error will generate a large number of difference pixels in the edge region, forming false defects and seriously interfering with the judgment of real geometric deformation. More importantly, it cannot quantify and classify geometric deviations, for example, it cannot distinguish between a difference region caused by ring deformation or simply uneven ink color, and has weak diagnostic ability.

[0006] Method based on global deformation field (optical flow method / DIC): More advanced solutions draw on the ideas of Digital Image Correlation (DIC) or optical flow method, which describe the overall deformation by calculating a dense displacement vector field between the template image and the image to be detected. By analyzing the divergence, curl, etc. of this vector field, stretching, compression, shear, etc. deformation modes can be identified. This method can provide comprehensive deformation information and is theoretically very powerful. However, it has extremely high computational complexity, making it difficult to meet the real-time requirements of high-speed printing production lines. At the same time, the interpretation of the vector field requires deep professional knowledge, and the setting of algorithm parameters is also very complex, making it difficult to promote and apply in industrial sites. Moreover, it is also difficult to directly relate the macroscopic deformation field to specific, local geometric elements (such as the width of a certain line).

[0007] Through in-depth analysis of the above existing technical solutions, it can be concluded that they have several key common shortcomings in dealing with geometric and morphological defects: 1. Feature characterization capability is insufficient: existing methods are either too global (e.g. overall parameters in solution one), losing local detail information; or too low-level (e.g. pixel difference in solution two), lacking abstract understanding of geometric shapes. They cannot effectively decompose and describe the geometric elements of the pattern, resulting in insensitivity to local and subtle morphological changes.

[0008] 2. Weak anti-interference ability and poor stability: edge extraction-based methods are easily affected by image quality; pixel difference-based methods have zero tolerance for registration errors. These methods are prone to false positives in the face of normal fluctuations caused by factors such as lighting, vibration, and materials in production environments, resulting in poor stability.

[0009] 3. Threshold setting lacks statistical basis: The tolerance threshold of geometric parameters (such as the allowed center offset and line width variation) is usually set by experience, lacking data support. It cannot be adjusted adaptively according to the actual production process capability, resulting in false positives or missed detections.

[0010] 4. Defect diagnosis information is ambiguous: When an anomaly is detected, the system usually only gives a general conclusion that there is a geometric deviation, and cannot accurately inform which part of which circle has undergone what deformation (such as ovality deviation or local indentation) or which section of which line has uneven width, which is not conducive to rapid problem root analysis by production personnel.

[0011] 5. Contradiction between computational efficiency and real-time performance: High-precision deformation field analysis methods have heavy computational burden, making it difficult to meet the real-time detection requirements of high-speed production lines while ensuring accuracy. SUMMARY

[0012] To overcome the defects of the prior art, the present application provides a printed matter visual inspection method for geometric morphological deviation, comprising the following steps: S10, offline modeling; S20, online detection; S10 comprises the following steps: S11, collecting good product samples; S12, geometric element identification and parameterized extraction; S13, element sampling representation; S14, first-stage statistical topography modeling; S15, second-stage statistical topography modeling; S20 comprises the following steps: S21, acquisition and preprocessing of test images; S22, fast extraction and parameterization of geometric elements; S23, topography deviation calculation; S24, defect identification and positioning; S25, defect classification and visualization.

[0013] Preferably, the S12 geometric element identification and parameterization extraction specifically includes: S121, preprocessing and sub-pixel edge extraction: filtering and denoising each good image, and then using a sub-pixel edge detection algorithm to obtain edge point positions with higher accuracy than integral pixels, forming an edge point set; S122, element decomposition and fitting: according to a pre-set region of interest and a graphic template, segmenting the point set belonging to different geometric elements from the edge point set.

[0014] Preferably, the S13 includes equiangular sampling and equidistant radius sampling.

[0015] Preferably, the S14 first stage statistical topography modeling specifically includes, for each sampling point parameter of each geometric element, calculating its mean value in samples and standard deviation , establishing an allowable fluctuation range μ ± kσ, as the global sensitivity coefficient.

[0016] Preferably, the S15 second stage statistical topography modeling specifically includes calculating a feature interval and establishing a statistical morphological model.

[0017] Preferably, in the S22, sub-pixel edge extraction, geometric element fitting and sampling are performed on the to-be-inspected image to obtain a radius vector of the to-be-inspected ring and a width vector of the to-be-inspected line.

[0018] Preferably, in the S23, the parameter vector of the to-be-inspected element is compared with the mean value vector of the statistical topography model point by point, and a standardized residual vector is calculated.

[0019] Preferably, in the S24, a standardized residual vector threshold is set, for each geometric element, if there are continuous several sampling points whose standardized residual vector is greater than the standardized residual vector threshold, it is determined that there is a deformation defect in this local area of the geometric element; at the same time, the global parameters are calculated and compared with the tolerance learned based on historical data as an auxiliary judgment.

[0020] Preferably, in the S25, the defect classification includes ring deformation, uneven line thickness and angle offset.

[0021] To achieve the above objectives, the present invention also provides a visual inspection device for printed materials with geometric shape deviations. This device implements the above-mentioned inspection method and includes an area scan camera, a telecentric lens, a parallel backlight or coaxial light source, an image acquisition card, an industrial computer, a controller, and a motor. The telecentric lens is arranged in front of the area scan camera, and a parallel backlight or coaxial light source is arranged to the side below the telecentric lens. The area scan camera is sequentially connected to the image acquisition card, the industrial computer, the controller, and the motor to perform image acquisition, digital image processing, output control, and execution control on the object under test.

[0022] The present invention has at least the following beneficial effects: 1. Extremely high sensitivity to the detection of local geometric defects: Through “sampling point characterization” and “local Z value judgment”, this invention can capture the morphological deviations of small areas on geometric elements, such as local depressions of rings and local swelling of lines, which are easily missed by fitting methods that rely on global parameters.

[0023] 2. The test results are stable and reliable with strong anti-interference capabilities: This is due to the use of statistical learning to obtain the fluctuation range. Standardized residuals were used as the judgment criterion. The system can adaptively tolerate normal fluctuations in production, only triggering an alarm when "abnormal" deviations occur. Subpixel edge extraction technology also ensures the accuracy of input data, improving stability from the source.

[0024] 3. Specific and actionable diagnostic information: This invention not only provides a "non-conforming" conclusion, but also clearly identifies "which part" and "which type" of geometric problem occurred. This provides invaluable, actionable information for debugging and maintenance personnel on the production line, significantly reducing troubleshooting time.

[0025] 4. Scientific and adaptive threshold setting: detection threshold Sensitivity coefficient during modeling In essence, uniformity is a statistical confidence level. This gives threshold setting a clear statistical meaning, allowing it to be scientifically set based on product quality requirements (such as PPM levels) rather than through blind trial and error.

[0026] 5. Balancing accuracy and efficiency: Although detailed element decomposition is performed, the entire algorithm avoids computationally intensive methods such as optical flow / DIC. During the online detection phase, the core operations are vector subtraction and comparison, resulting in high computational efficiency and meeting the real-time requirements of high-speed production lines. Attached Figure Description

[0027] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 The flowchart of the printing product visual inspection method for geometric shape deviation of an embodiment of the present application; Figure 2 The structural diagram of the printing product visual inspection device for geometric shape deviation of an embodiment of the present application; Figure 3 The defect result visualization diagram of the printing product visual inspection device for geometric shape deviation of an embodiment of the present application. DETAILED DESCRIPTION

[0028] The preferred embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0029] Referring to Figure 1 , the printing product visual inspection method for geometric shape deviation of the present application comprises the following steps: S10, offline modeling; comprising: S11, collecting good product samples; the system collects confirmed good product images . .

[0030] S12, geometric element identification and parameterized extraction; pre-processing and sub-pixel edge extraction: filter and denoise each good product image, and then use a sub-pixel edge detection algorithm (such as a method based on Zernike moments) to obtain edge point positions with higher accuracy than integral pixels. Let the extracted edge point set be .

[0031] Element decomposition and fitting: according to a pre-set ROI (region of interest) and a graph template (such as “circle”, “straight line”, “arc”), the point set belonging to different geometric elements is segmented from the edge point set . For example, for a circular ring, the RANSAC (random sample consensus) algorithm or the least square method is used to fit the center and radius of each circular ring. For a straight line segment, the end points or equation parameters are obtained by fitting. The present application not only extracts global parameters, but more importantly, characterizes the geometric elements by sampling points.

[0032] For a circular ring, a point is sampled at a fixed angular interval (for example, 5°) along the circumference, and the radial distance of each sampling point relative to the center is recorded ( ). In this way, a circular ring is no longer described by the center and the radius, but by a radius vector , which contains the local shape information of the circular ring.

[0033] For a line, sample along the line centerline direction with fixed length interval points. At each sampling point, draw the normal of the line, and calculate the distance from the two edges of the line to the centerline (i.e. half width), to get the upper edge distance vector and the lower edge distance vector . The width vector of the line is . This contains the width information of each part of the line.

[0034] S13, element sampling representation, including equiangular sampling and equidistant radius sampling.

[0035] S14, first stage statistical topography modeling; assuming that all good product images have completed element-based registration (for example, taking the center of the first ring as the coordinate origin), align the parameters of each geometric element in samples.

[0036] For each sampling point parameter of each geometric element, calculate its mean value and standard deviation in samples.

[0037] For a ring, the average radius of the th sampling point is , and the standard deviation is . The statistical model of the entire ring is represented as , where , .

[0038] For a line, the average width of the th sampling point is , and the standard deviation is . The statistical model of the line width is represented as .

[0039] Establish the allowable fluctuation range: similar to the previous draft, we set an allowable fluctuation boundary for each sampling point parameter based on the standard deviation: . Where is the global sensitivity coefficient. This boundary defines the "normal topography" range of each point on the geometric element.

[0040] S15, second stage statistical topography modeling; including calculating feature intervals and establishing statistical morphology models.

[0041] S20, online detection; including the following steps: S21, acquisition and preprocessing of test images; for each product image to be detected , perform the following steps: S22, geometric element fast extraction and parameterization; same procedure as offline stage, sub-pixel edge extraction and geometric element fitting and sampling, get radius vector of the circle and width vector of the line .

[0042] S23, topography deviation calculation; compare the parameter vector of the inspected element with the mean vector of the statistical topography model, calculate the normalized residual vector

[0043] For circle, ; For line, ; This value quantifies the degree of deviation of the inspected element from the "normal topography" at each sampling point, is the multiple of standard deviation.

[0044] S24, defect identification and localization; set a Z value threshold (e.g. consistent with modeling). For each geometric element, traverse all its sampling points.

[0045] If there are continuous multiple sampling points (e.g., more than 3) whose value is greater than , it is determined that there is a deformation defect in this local area of the geometric element. For example, for a circle, the of continuous multiple sampling points indicates that there is a local concave or convex in this area. For a line, the of continuous multiple sampling points indicates that there is a width unevenness in this section.

[0046] At the same time, calculate the global parameters (such as overall circle center offset, average radius change, line overall angle offset) and compare with the tolerance learned based on historical data as auxiliary judgment.

[0047] S25, defect classification and visualization. The system will preliminarily classify the geometric defects according to the distribution and morphology of the out-of-tolerance points.

[0048] Circle deformation: can be subdivided into poor ellipticity (systematic deviation of radius in two opposite directions), local out-of-roundness (continuous out-of-tolerance points in a specific area).

[0049] Line unevenness: according to whether the out-of-tolerance points are distributed on the upper edge, lower edge or both sides, it is judged whether the line is overall widened / narrowed or single-sided burr.

[0050] Angle offset: judge by comparing the difference between the overall direction angle of the line and the mean value of the model.

[0051] ​​As shown in the accompanying drawings Figure 3 , the detection result interface will highlight the elements with geometric defects on the original drawing, and indicate the defect type and specific location with different colors or marks (for example, use red line segments to mark the deformed area on the ring, and use blue shadows to mark the section with too wide lines), and generate a detection report containing specific quantitative data (such as maximum deformation, position coordinates).

[0052] Referring to Figure 2 , the present application also provides a printed matter visual detection device for geometric shape deviation, which implements the above-mentioned detection method, and comprises a face array camera 11, a telecentric lens 12, a parallel backlight or coaxial light source 13, an image acquisition card 20, an industrial computer 30, a controller 40 and a motor 51, wherein the telecentric lens 12 is arranged in front of the face array camera 11, the parallel backlight or coaxial light source 13 is arranged on the lower side of the telecentric lens 12, and the face array camera 11 is sequentially connected with the image acquisition card 20, the industrial computer 30, the controller 40 and the motor 51 to perform image acquisition, digital image processing, output control and execution control (the motor 51 removes the measured object 60 that does not meet the requirements from the conveying path through the screening rod 52, thereby realizing automatic sorting).

[0053] The protection scope of the present application is determined by the appended claims, and the core innovation point and key protection point are as follows: 1. Geometric element parameterization representation method based on sampling points: the technical scheme of densely sampling geometric elements such as rings and lines, and describing the detailed topography with vectors (such as radius vector, width vector).

[0054] 2. Construction method of geometric element statistical topography model: the specific process of forming model is realized by using multiple good images to calculate the mean value and standard deviation of each sampling point parameter of the geometric element.

[0055] 3. Geometric defect identification algorithm based on standardized residual vector : the specific judgment logic and mathematical formula for positioning and identifying local geometric deformation are realized by calculating the point-by-point deviation of the to-be-detected element from the model mean value, and comparing it with the model standard deviation.

[0056] 4. Local defect judgment rule for continuous over-standard points: the decision mechanism for confirming local geometric defects is realized by judging whether there are continuous multiple points whose Z values exceed the threshold in the sampling point vector.

[0057] 5. Geometric feature-based defect classification and visualization system: The system integrates the functions of automatic classification of distribution patterns according to Z-value outliers, circular ring deformation, uneven line thickness, etc., and highlights the defects on the image and generates a detailed diagnostic report.

[0058] 6. Complete geometric detection system architecture integrating element decomposition, statistical learning, and model comparison: From hardware selection (such as telecentric lens) to offline modeling, to online detection and visualization, the system provides a complete solution.

[0059] Without departing from the core concept of the present application, the following alternatives can be implemented: Alternative sampling method: In addition to equi-angle / equi-distance sampling, adaptive sampling can be performed according to the curvature variation of geometric elements, increasing the sampling density in areas with high curvature (such as corners) and reducing the sampling density in flat areas to optimize calculation and representation efficiency.

[0060] Alternative topographic representation parameters: For circular rings, in addition to radial distance, tangential angle variation can be used for representation; for complex patterns, Fourier descriptors or Zernike moments can be extracted as shape features, and statistical models can be built based on these features.

[0061] Alternative deviation measurement method: In addition to using Z-value based on standard deviation, Mahalanobis distance can be used to measure the overall deviation between the topographic vector of the element to be detected and the distribution of good sample, which can consider the correlation between parameters.

[0062] Alternative model updating mechanism: Online learning mechanism can be introduced, under strict control conditions (such as multiple detections of good products and fluctuations within controllable range), the geometric parameters of the current product are used to incrementally update the statistical topographic model and , so that the model has the ability to slowly adapt to the drift of the production line.

[0063] Alternative solution combined with deep learning: The method can be used as a core detector with strong logic, and the structured geometric parameters and preliminary judgment output by the method can be combined with the analysis results of a lightweight neural network on the original image block to make a final decision, which can handle extremely complex or rare geometric defect patterns.

[0064] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present application.

Claims

1. A visual inspection method for printed materials to address geometric shape deviations, characterized in that, Includes the following steps: S10, offline modeling; S20, online testing; S10 includes the following steps: S11, Collect samples of good products; S12, Geometric feature identification and parametric extraction; S13, element sampling representation; S14, First stage statistical morphological modeling; S15, Second stage statistical morphological modeling; S20 includes the following steps: S21, Acquisition and preprocessing of test images; S22, Rapid extraction and parameterization of geometric features; S23, Calculation of morphological deviation; S24, Defect identification and location; S25, Defect Classification and Visualization.

2. The visual inspection method for printed materials targeting geometric shape deviations according to claim 1, characterized in that, The S12 geometric feature identification and parameter extraction specifically includes: S121, Preprocessing and subpixel edge extraction: Each good product image is filtered and denoised, and then a subpixel edge detection algorithm is used to obtain the position of edge points with a precision higher than that of whole pixels, forming an edge point set; S122, Element decomposition and fitting: Based on the preset region of interest and graphic template, segment the point set belonging to different geometric elements from the edge point set.

3. The visual inspection method for printed materials targeting geometric shape deviations according to claim 2, characterized in that, S13 includes isoangular sampling and equidistant radius sampling.

4. The visual inspection method for printed materials targeting geometric shape deviations according to claim 3, characterized in that, The first stage of statistical topography modeling in S14 specifically includes calculating the parameters of each sampling point for each geometric feature. The mean of the samples and standard deviation Establish the allowable fluctuation range μ ± kσ. This represents the global sensitivity coefficient.

5. The visual inspection method for printed matter targeting geometric shape deviations according to claim 4, characterized in that, The second stage of statistical morphology modeling in S15 specifically includes calculating feature intervals and establishing a statistical morphology model.

6. The visual inspection method for printed matter targeting geometric shape deviations according to claim 5, characterized in that, In step S22, subpixel edge extraction, geometric element fitting, and sampling are performed on the image to be inspected to obtain the radius vector of the ring to be inspected and the width vector of the line to be inspected.

7. The visual inspection method for printed matter targeting geometric shape deviations according to claim 6, characterized in that, In step S23, the parameter vector of the element to be inspected is compared point by point with the mean vector of the statistical morphology model to calculate the standardized residual vector.

8. The visual inspection method for printed matter with geometric shape deviation according to claim 7, characterized in that, In step S24, a standardized residual vector threshold is set. For each geometric element, all its sampling points are traversed. If there are several consecutive sampling points whose standardized residual vectors are greater than the standardized residual vector threshold, it is determined that there is a deformation defect in this local area of ​​the geometric element. At the same time, global parameters are calculated and compared with the tolerance learned based on historical data as an auxiliary judgment.

9. The visual inspection method for printed matter targeting geometric shape deviations according to claim 8, characterized in that, The defect classification in S25 includes ring deformation, uneven line thickness, and angular deviation.

10. A visual inspection device for printed materials to address geometric shape deviations, characterized in that, The device implements the detection method described in any one of claims 1-9, comprising an area scan camera, a telecentric lens, a parallel backlight or coaxial light source, an image acquisition card, an industrial computer, a controller, and a motor. The telecentric lens is positioned in front of the area scan camera, and a parallel backlight or coaxial light source is positioned to the side of the lower part of the telecentric lens. The area scan camera is sequentially connected to the image acquisition card, the industrial computer, the controller, and the motor to perform image acquisition, digital image processing, output control, and execution control on the object under test.

Images