Packaging material printing quality detection method and system based on machine vision

Abstract

The invention relates to the technical field of image processing, and discloses a packaging material printing quality detection method and system based on machine vision. The method comprises the following steps: acquiring a multispectral image sequence and three-dimensional shape data of a moving packaging and printing material under different illumination, and constructing a dynamic three-dimensional physical attribute field; generating a virtual reference image and a dynamic reference image, constructing a multi-modal reference image, carrying out space-time registration on the multi-modal reference image and the dynamic three-dimensional physical attribute field, and calculating the difference between the multi-modal reference image and the dynamic three-dimensional physical attribute field in different dimensions to generate a multi-dimensional difference quality field; each dimension difference is enhanced through local statistics, and the comprehensive defect confidence coefficient is calculated based on the enhanced dimension difference; and extracting a defect region based on the comprehensive defect confidence, generating a defect evolution sequence and a defect track, analyzing defect track characteristics, constructing a correlation model in combination with process parameter time sequence data of the printing equipment, and positioning a defect reason. According to the invention, high-precision, multi-dimensional and self-adaptive printing defect detection and traceability can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a packaging material printing quality detection method and system based on machine vision. BACKGROUND

[0002] As an important carrier of commodity value and brand image, the printing quality of packaging materials directly affects the market competitiveness of products. In high-speed printing production lines, due to factors such as ink properties, material deformation, equipment wear and tear, and process parameter fluctuations, printed matter often has various defects such as color difference, ink spots, scratches, inaccurate overprinting, and texture loss. Traditional manual visual inspection methods are inefficient and cannot meet the strict requirements of modern industry for quality and production data traceability.

[0003] With the development of machine vision technology, image-based automated detection systems have gradually been applied in the field of printing quality detection. However, most existing technologies still remain in simple comparison and analysis of two-dimensional plane images, with many inherent limitations. First, two-dimensional images cannot effectively capture defects related to height and topography, such as slight concave-convex, blistering, indentation, or uneven ink layer thickness. Second, packaging materials come in a variety of types, including grating film, metal foil, and laser paper, which have complex optical properties and significant appearance differences under different lighting and viewing angles, resulting in poor stability and high false positive rate of two-dimensional image detection systems based on fixed lighting. Third, packaging materials on the production line inevitably undergo physical changes such as stretching and shaking, and static and rigid reference template images cannot effectively compensate for these changes, causing registration difficulties and easy misjudgment of normal deformation as defects.

[0004] In addition, existing detection systems mostly stop at defect discovery and positioning, lacking the ability to diagnose and predict defects. The appearance of defects is often a lagging reflection of abnormal equipment and process states, for example, a periodic scratch may be caused by damage to the cylinder, and ink color gradient may be related to a small fault in the ink key. Without deep correlation analysis of the detected defect sequence and real-time process parameters of the printing equipment, it will be difficult to quickly locate the root cause of the fault and miss the best opportunity for predictive maintenance. Therefore, the present application provides a packaging material printing quality detection method and system based on machine vision. SUMMARY

[0005] The present application provides a packaging material printing quality detection method and system based on machine vision, which is used to realize multi-dimensional accurate detection of packaging printing materials.

[0006] In a first aspect, the present application provides a packaging material printing quality detection method based on machine vision, which comprises: Step S1, a multispectral image sequence of the packaging printed material in movement under different illuminations is collected, three-dimensional topographic data of the packaging material surface is acquired in real time, a dynamic three-dimensional physical property field is constructed based on the multispectral image sequence and the three-dimensional topographic data; Step S2, a virtual reference image and a dynamic reference image are generated, a multi-modal reference image is constructed, the dynamic three-dimensional physical property field is spatiotemporally registered with the multi-modal reference image, a multi-dimensional differential quality field is generated by calculating the difference between the two in different dimensions; Step S3, for each single-dimensional difference in the multi-dimensional differential quality field, a local neighborhood is defined, the local statistical characteristics of the single-dimensional difference in each local neighborhood are calculated, and a linear superposition based on the local statistical characteristics is performed to obtain an enhanced dimensional difference, and a geometric mean of the enhanced dimensional difference is calculated to obtain a comprehensive defect confidence; Step S4, based on the comprehensive defect confidence, a defect region is extracted, a defect evolution sequence is generated, an evolution trajectory of the same defect is associated based on the defect evolution sequence, defect trajectory features are obtained by analyzing each evolution trajectory, a correlation model of the defect trajectory features and the process parameter time series data of the printing equipment is established, and the defect cause is obtained based on the correlation model.

[0007] In a second aspect, the application provides a packaging material printing quality detection system based on machine vision, which comprises: An acquisition and processing module, which collects a multispectral image sequence of the packaging printed material in movement under different illuminations, acquires three-dimensional topographic data of the packaging material surface in real time, and constructs a dynamic three-dimensional physical property field based on the multispectral image sequence and the three-dimensional topographic data; A difference calculation module, which generates a virtual reference image and a dynamic reference image, constructs a multi-modal reference image, spatiotemporally registers the dynamic three-dimensional physical property field with the multi-modal reference image, and generates a multi-dimensional differential quality field by calculating the difference between the two in different dimensions; A defect detection module, which, for each single-dimensional difference in the multi-dimensional differential quality field, defines a local neighborhood, calculates the local statistical characteristics of the single-dimensional difference in each local neighborhood, and performs a linear superposition based on the local statistical characteristics to obtain an enhanced dimensional difference, and calculates a geometric mean of the enhanced dimensional difference to obtain a comprehensive defect confidence; A deep analysis module, which extracts a defect region based on the comprehensive defect confidence, generates a defect evolution sequence, associates an evolution trajectory of the same defect based on the defect evolution sequence, analyzes each evolution trajectory to obtain defect trajectory features, performs detailed analysis on the defect trajectory features, establishes a correlation model of the defect trajectory features and the process parameter time series data of the printing equipment, and obtains the defect cause based on the correlation model.

[0008] Compared with the prior art, the application has at least the following advantages: The technical solution provided in this application constructs a dynamic three-dimensional physical property field by combining multispectral images with three-dimensional topography data, enabling comprehensive perception of color, texture, shape, and optical properties, thereby improving the dimensionality and accuracy of defect detection. It also constructs a multimodal reference benchmark by combining virtual and dynamic reference images, effectively addressing material deformation, illumination changes, and equipment status fluctuations, thus reducing false alarm rates. Furthermore, it enhances the difference across dimensions through local statistical characteristics and physical causal logic, and calculates the comprehensive defect confidence level using geometric mean, improving the ability to identify real defects. Finally, it constructs defect trajectories based on defect evolution sequences and establishes a correlation model based on equipment process parameters, enabling defect type classification, development trend prediction, and precise fault source location. It also supports predictive maintenance and process optimization. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of one embodiment of the packaging material printing quality inspection method based on machine vision in this application. Figure 2 This is a schematic diagram of one embodiment of the machine vision-based packaging material printing quality inspection system in this application. Detailed Implementation

[0011] This application provides a machine vision-based method and system for inspecting the printing quality of packaging materials. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0012] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the machine vision-based packaging material printing quality inspection method in this application includes: Step S1: Collect multispectral image sequences of the moving packaging printing material under different lighting conditions, obtain three-dimensional morphological data of the packaging material surface in real time, and construct a dynamic three-dimensional physical property field based on the multispectral image sequence and three-dimensional morphological data.

[0013] Specifically, multi-view cameras are deployed on the packaging material printing production line to simultaneously acquire multispectral image sequences of moving packaging printing materials under different lighting conditions. Additionally, high-resolution laser displacement sensors are used to acquire real-time three-dimensional morphological data of the packaging material surface. This three-dimensional morphological data refers to a precise description of the microscopic three-dimensional geometry of the packaging material surface, including not only color and brightness information but also height information. A dynamic three-dimensional physical property field is constructed based on the multispectral image sequences and the three-dimensional morphological data. This field includes color, texture, morphology, and optical properties. Morphology refers to the three-dimensional geometric topology of the packaging material surface. Morphological data can detect height-related defects such as dents, indentations, bubbles, and foreign matter adhesion. It can also accurately measure the uniformity of printing thickness and identify local or global deformation caused by material stretching and mechanical pressure.

[0014] Step S2: Generate a virtual reference image and a dynamic reference image, construct a multimodal reference image, perform spatiotemporal registration of the dynamic three-dimensional physical property field with the multimodal reference image, and calculate the differences between the two in different dimensions to generate a multidimensional differential mass field.

[0015] Specifically, based on the printed digital artwork to be inspected and combined with the physical property data of the packaging material, a virtual reference image is generated under different lighting and deformation conditions using a fiber optic tracing algorithm. A generative adversarial network is then used to perform style transfer and deformation compensation on historical qualified product images, generating a dynamic reference image that matches the current production conditions. The virtual and dynamic reference images together constitute a multimodal reference image. Multidimensional difference data includes the differences between the dynamic three-dimensional physical property data and the multimodal reference image in four dimensions: color, texture, morphology, and the optical properties of the material.

[0016] Step S3: For each single-dimensional difference in the multidimensional difference mass field, define a local neighborhood, calculate the local statistical properties of the single-dimensional difference in each local neighborhood, and perform linear superposition based on the local statistical properties to obtain the enhanced dimensional difference. Calculate the geometric mean of the enhanced dimensional difference to obtain the comprehensive defect confidence. Specifically, to enhance the real defects, a fixed-size region, such as 9*9 pixels, is defined as the local neighborhood centered on each pixel in the image to be detected. The local statistical characteristics of each single-dimensional difference within each local neighborhood are calculated. The single-dimensional difference includes color, texture, shape, and optical properties. The local statistical characteristics include the mean and standard deviation. A linear superposition based on the local statistical characteristics is then performed to obtain the enhanced dimensional difference. The specific methods for enhancing each dimensional difference will be explained in detail later. Assuming that the enhanced dimensional differences are color dimensional difference C_e, texture dimensional difference T_e, shape dimensional difference S_e, and optical property dimensional difference O_e, the geometric mean of the enhanced dimensional differences is used to calculate the comprehensive defect confidence score, which is the average of the four enhanced dimensional differences. The formula for calculating the comprehensive defect confidence score by calculating the geometric mean of the enhanced dimensional differences is: M=(C_e+T_e+S_e+O_e) / 4.

[0017] Step S4: Extract the defect region based on the comprehensive defect confidence level, generate the defect evolution sequence, associate the evolution trajectory of the same defect based on the defect evolution sequence, analyze each evolution trajectory to obtain defect trajectory features, conduct a detailed analysis of the defect trajectory features, establish a correlation model between the defect trajectory features and the time series data of the printing equipment process parameters, obtain the defect cause based on the correlation model, and conduct a detailed analysis of the defect based on the defect trajectory features. Specifically, pixels with a comprehensive defect confidence score greater than a preset threshold are designated as defect areas. The comprehensive defect confidence scores of N consecutive packaging printing materials are obtained as a defect evolution sequence. The evolution trajectory of the same defect is correlated by calculating the overlap rate of defect areas in adjacent printing materials. Each defect trajectory is analyzed to obtain defect trajectory characteristics, including lifecycle, confidence score intensity change curve, and spatial movement path. A detailed analysis of these characteristics is conducted, with specific methods explained later. When a correlation is detected between the defect trajectory characteristics and the time-series data of a certain process parameter, the corresponding process parameter and its associated process configuration or process step are designated as the defect cause. This facilitates subsequent adjustments to the printing equipment's process parameters based on the defect cause and the corresponding process configuration.

[0018] In one specific embodiment, generating a virtual reference image includes the following steps: The process involves acquiring original digital design drafts and material physical property data, including parameters such as gloss, roughness, ink absorption, and thickness of the packaging material. Using ray tracing algorithms, the actual lighting conditions on the production line and the reflection, refraction, and scattering behavior of light on the material surface are simulated to generate multiple virtual reference images. These virtual reference images contain realistic highlights, shadows, and color gradients caused by minute material deformations.

[0019] Specifically, based on the original digital design manuscript of the printed matter, by introducing physical rendering models and materials science knowledge, a perfect image that should be presented under real physical conditions is simulated in a computer. The traditional method is to use the sample image obtained by taking pictures of real printed samples as a test benchmark for comparison. However, the printed samples themselves may have defects and cannot cover all production conditions. Therefore, this application generates an absolutely ideal and flawless theoretical benchmark through the above method. Moreover, since the appearance of grating materials, metallic luster films, holographic patterns, etc. varies greatly with viewing angle and lighting, the virtual reference image can accurately simulate these complex optical effects, so that the test benchmark and the image to be tested are in the same external environmental conditions, providing a rich comparison benchmark for the test system.

[0020] In one specific embodiment, generating a dynamic reference image includes the following steps: Images of similar products that have been deemed qualified are selected from historical inspection data. Image registration and fusion techniques are used to align and fuse the product images to generate template images. Then, a conditional generative adversarial network is used to generate dynamic reference images that match the current production status based on the template images and the current equipment status.

[0021] Specifically, the state of a production line is not static. Dynamic reference images can track these normal fluctuations, thereby avoiding misjudging normal, minor process changes as defects and reducing false alarm rates. Secondly, for flexible materials such as plastic films, deformation on the production line is inevitable. Dynamic reference images are obtained by learning the deformation patterns of historical qualified products, thus predicting and compensating for the expected deformation of the current product, making subsequent registration and differentiation more accurate. Finally, in some cases, perfect digital design drafts may not be available. In this case, dynamic reference images can be used as the main reference benchmark. Therefore, the above method is based on historical qualified product images accumulated during the production process, and uses production adversarial networks for learning and fusion to generate dynamic reference images that can adapt to minor changes in the production environment.

[0022] In one specific embodiment, generating a multidimensional difference mass field includes the following steps: The three-dimensional physical property field generated based on the multimodal reference image is used as the reference three-dimensional physical property field. Spatiotemporal registration is performed between the dynamic three-dimensional physical property field and the reference three-dimensional physical property field. After registration is completed, the differences in color dimension, texture dimension, shape dimension and optical property dimension are calculated. The calculated differences in the four dimensions are superimposed in the same three-dimensional spatial coordinate system to generate a four-dimensional difference mass field.

[0023] Specifically, for accurate comparison, the virtual reference image and dynamic reference image in the multimodal reference image are generated using the same physical rendering model to create a reference 3D physical property field with the same expression as the dynamic 3D physical property field. Rotation, translation, scaling, and even nonlinear deformation correction are performed based on topographic data and feature points. Feature points can be any easily detectable point, such as edge points and points with the highest height, ensuring that each point in the dynamic 3D physical property field to be detected is aligned with its corresponding point in the reference 3D physical property field in 3D space. After registration, the differences in four independent dimensions are calculated. Color dimension difference refers to calculating the difference in color values ​​between the field to be detected (dynamic 3D physical property field) and the reference field (reference 3D physical property field) for each pixel. For example, for each pixel, its three values ​​L1, L2, a1, a2, b1, and b2 in Lab mode are obtained for the field to be detected and the reference field. L1 and L2 represent brightness, and a1, a2, b1, and b2 represent color contrast dimensions. The color difference ΔC is calculated as sqrt((L1-L2)). 2 +(a1-a2) 2 +(b1-b2) 2 The calculation method for texture dimension difference can be based on gray-level co-occurrence matrix, which will not be explained in detail here. The calculation of morphology dimension difference is the difference between the field to be inspected and the reference field in three-dimensional geometry, especially the small local concavity and convexity changes. The specific calculation process will be explained in detail later. Optical characteristics are also compared, such as the difference in the bidirectional reflection distribution function value under different light source configurations, as the optical characteristic dimension difference. The difference in the ratio of specular reflection component to diffuse reflection component can also be simply compared as the optical characteristic dimension difference. Then, the difference fields of the above four dimensions are combined in the same three-dimensional spatial coordinate system to form a unified structured data entity. This data entity is the multidimensional difference mass field. In the data structure, each pixel corresponds to a four-dimensional vector.

[0024] In one specific embodiment, calculating the topographic dimension difference specifically includes the following steps: On the registered dynamic 3D physical property field and the reference 3D physical property field, a local window centered on each pixel is defined, and for each local window, the height distribution statistics within the local window are calculated. For each pixel, the absolute difference in statistics between the 3D physical property field and the reference 3D physical property field is calculated. The absolute difference is then merged into a single shape difference value through a fusion function. The combination of shape difference values ​​corresponding to all pixels is used as the corresponding shape dimension difference.

[0025] Specifically, to calculate the topographic dimension difference, a local window centered on each pixel is defined on the registered test field and reference field, such as 7*7 or 15*15 pixels. For each local window, statistics are calculated, including height mean, height standard deviation, and height deviation. The height mean reflects the overall height level of the local window, the height standard deviation reflects the roughness or fluctuation of the local window surface, and the height deviation reflects the symmetry of the height distribution. For each pixel, the absolute difference between the test field and the reference field in the above three statistics is calculated. The three absolute differences are combined into a single topographic difference value through a fusion function. The fusion function can be a weighted summation function, with different weights assigned to the three statistics. The specific weights can be adjusted based on the sensitivity of different defects. For example, for detecting flatness defects, a higher weight can be assigned to the height standard deviation, and for detecting concavity and convexity defects, a higher weight can be assigned to the height mean. The combination of the topographic difference values ​​corresponding to all pixels is used as the topographic dimension difference.

[0026] In one specific embodiment, the enhanced dimensionality difference is obtained by performing linear superposition based on local statistical properties, specifically including the following steps: Directional excitation is established based on predefined physical causal logic, texture dimension difference is enhanced based on color dimension difference, shape dimension difference is enhanced based on texture dimension difference, and optical property dimension difference is enhanced based on shape dimension difference.

[0027] Specifically, directional excitation is based on strict, predefined physical causal logic. Since ink is the common carrier for forming color and texture, the absence of ink dots or the ink color being too light will cause the corresponding area to become lighter in color. At the same time, the printing dot texture in the corresponding area will also become blurred, incomplete, or even disappear. Therefore, texture difference is enhanced based on color dimension difference. For printed materials, the absence or occurrence of certain textures may be related to changes in the thickness of the ink layer. For example, ink accumulation and foreign matter adhesion not only destroy the normal texture but may also greatly lead to thickening of the ink layer in the area or the formation of physical protrusions. Therefore, morphology dimension difference is enhanced based on texture dimension difference, since the surface geometry directly determines the optical reflection behavior. For example, scratches, protrusions, or dents can change the normal direction of a surface, and may also significantly alter the position, intensity, and diffuse reflection distribution of specular reflection spots. Therefore, definite morphological changes are a strong basis for predicting changes in optical properties. Thus, based on morphological dimensional differences, optical property dimensional differences can be enhanced. The unified enhancement formula is: Target Enhanced Dimensional Difference = Original Enhanced Dimensional Difference + Gain Coefficient * (Mean Local Difference of Source Enhancement Difference * (1 - Standard Deviation of Target Enhanced Dimensional Difference)). Taking color dimensional difference enhancement of texture dimensional difference as an example, the target enhanced dimensional difference is the texture dimensional difference, and the source enhancement difference is the color dimensional difference. The enhancement of texture dimensional difference is based on color dimensional difference. The enhancement formula is: T_e = T + k * (μ_C * (1 - σ_T)), where T_e represents the enhanced texture dimension difference, T represents the original texture dimension difference, k is the gain coefficient, a pre-set constant greater than 0 and less than 1, for example, set to 0.3, used to control the overall intensity of the excitation and prevent over-enhancement, μ_C represents the average value of the color dimension difference in the local neighborhood, and σ_T represents the standard deviation of the texture dimension difference in the local neighborhood. Using μ_C as the excitation signal, when the overall color of a region becomes darker or lighter, it is assumed that the texture of the center and surrounding areas of the corresponding region may also change due to the absence or increase of ink. 1-σ_T) serves as an adaptive adjustment factor, acting as an intelligent switch to prevent false alarms. For example, in smooth regions, the texture itself is simple, so σ_T is small, and therefore 1-σ_T≈1. If a color change is detected at this time, the system will strongly assume that the texture should also be synchronously abnormal, thus significantly increasing the texture dimension difference. In complex texture regions, the background is complex and variable, and σ_T itself is large, so 1-σ_T≈0, and the excitation is greatly suppressed. This is because, in areas with very complex textures, a slight ink loss may cause texture changes that are easily drowned out by background noise. If it is forcibly enhanced at this time, it will misjudge a large number of normal complex textures as defects, leading to a surge in the false alarm rate.

[0028] In one specific embodiment, associating the evolutionary trajectories of the same defect based on the defect evolution sequence specifically includes the following steps: Based on the comprehensive defect confidence score of each defect in the defect evolution sequence, the corresponding defect region is obtained. An overlap rate threshold is defined, and the overlap rate of the defect regions of two adjacent printing materials is calculated. If the overlap rate is greater than the overlap rate threshold, the two adjacent printing materials are considered to belong to the same defect. The defect regions of multiple consecutive printing materials belonging to the same defect are combined to form a complete defect evolution trajectory.

[0029] Specifically, based on the comprehensive defect confidence score of each product in the defect evolution sequence, the defect area of ​​the corresponding printing material is obtained. An overlap rate threshold is defined, for example, 0.6. The overlap rate of the defect areas of two adjacent printing materials is calculated. Assuming that the defect area of ​​the previous printing material is A and the defect area of ​​the next product is B, the formula for calculating the overlap rate is overlap rate = OA / min(Area(A), Area(B)), where OA represents the overlap area of ​​defect area A and defect area B, and min(Area(A), Area(B)) represents the minimum area of ​​defect area A and defect area B. If the overlap rate is greater than the overlap rate threshold, the two adjacent printing materials are considered to belong to the same defect. The defect areas of multiple consecutive printing materials belonging to the same defect constitute a complete defect evolution trajectory.

[0030] In one specific embodiment, a detailed analysis of the defect trajectory characteristics is performed, specifically including the following steps: The defect trajectory characteristics include life cycle, confidence intensity change curve, and spatial movement path. The defect type is distinguished based on the length of the life cycle, the severity of the problem is assessed based on the defect type, the development stage of the defect is determined based on the confidence intensity change curve and a predictive maintenance basis is provided, and the fault source is accurately located based on the spatial movement path.

[0031] Specifically, the lifecycle refers to the number of products or the duration of time that the same defect spans from its first appearance to its final disappearance. This involves collecting time-series data on the process parameters of the printing equipment, including the ink key opening of each color group, pressure roller pressure, drying temperature, and printing speed. The lifecycle can be used to differentiate defect types. For example, defects with a lifecycle less than or equal to a preset threshold are considered short-lifecycle defects, typically considered random or transient issues, such as a single ink droplet. These types of problems usually do not require adjustments to the printing equipment parameters. Defects with a lifecycle greater than the preset threshold are considered long-lifecycle defects, considered persistent issues, such as worn doctor blades or damaged rollers. These situations require timely intervention from the printing equipment. Long-lifecycle defects are classified as severe defects and require timely intervention.

[0032] The confidence intensity change curve refers to the curve showing how the confidence level of a defect changes over time throughout its entire lifecycle. Based on this curve, the development stage of a defect can be determined. For example, a continuously rising curve indicates that the defect is worsening, such as increased wear on the scraper or expanding damage to the roller. A continuously falling curve suggests that the surface defect is weakening, possibly due to a minor disturbance or effective intervention by staff. A periodic oscillating curve indicates that the defect may be strongly correlated with a rotating component, whose damage point periodically contacts and leaves the substrate. Furthermore, by analyzing the slope of the confidence intensity change curve, it's possible to predict when the defect will exceed an unacceptable threshold and make adjustments before then, achieving predictive maintenance.

[0033] The spatial movement path refers to the trajectory of the center point of the defect on the packaging material over time. Accurately locating the fault source based on the spatial movement path means, for example, if the defect always appears in the same position on the material, it indicates that there is a problem with the static equipment component corresponding to that position, such as a fixed ink key or nozzle. If the path moves periodically in the lateral direction, it indicates that the fault source is a scanning printhead. Based on the above analysis, the corresponding analysis results can be obtained, and a detailed inspection report can be generated in conjunction with the cause of the defect to help relevant personnel locate and repair problems related to the printing equipment.

[0034] The above describes the machine vision-based packaging material printing quality inspection method in the embodiments of this application. The following describes the machine vision-based packaging material printing quality inspection system in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the machine vision-based packaging material printing quality inspection system in this application includes: The acquisition and processing module acquires multispectral image sequences of moving packaging and printing materials under different lighting conditions, obtains three-dimensional morphological data of the packaging material surface in real time, and constructs a dynamic three-dimensional physical property field based on the multispectral image sequences and three-dimensional morphological data. The differential calculation module generates virtual and dynamic reference images, constructs a multimodal reference image, performs spatiotemporal registration of the dynamic three-dimensional physical property field with the multimodal reference image, and calculates the differences between the two in different dimensions to generate a multidimensional differential mass field. The defect detection module defines a local neighborhood for each single-dimensional difference in the multidimensional differential mass field, calculates the local statistical characteristics of the single-dimensional difference in each local neighborhood, and performs linear superposition based on the local statistical characteristics to obtain the enhanced dimensional difference. The geometric mean of the enhanced dimensional difference is then calculated to obtain the comprehensive defect confidence. The deep analysis module extracts defect regions based on comprehensive defect confidence, generates defect evolution sequences, associates the evolution trajectories of the same defect based on the defect evolution sequences, analyzes each evolution trajectory to obtain defect trajectory features, conducts detailed analysis of defect trajectory features, establishes a correlation model between defect trajectory features and time-series data of printing equipment process parameters, and obtains the cause of defects based on the correlation model.

[0035] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0036] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0037] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A machine vision-based method for inspecting the printing quality of packaging materials, characterized in that, The method includes: Step S1: Collect multispectral image sequences of the moving packaging printing material under different illuminations, obtain three-dimensional morphological data of the packaging material surface in real time, and construct a dynamic three-dimensional physical property field based on the multispectral image sequence and three-dimensional morphological data. Step S2: Generate virtual reference image and dynamic reference image, construct multimodal reference image, perform spatiotemporal registration of dynamic three-dimensional physical property field with multimodal reference image, calculate the difference between the two in different dimensions to generate multidimensional differential mass field; Step S3: For each single-dimensional difference in the multidimensional difference mass field, define a local neighborhood, calculate the local statistical properties of the single-dimensional difference in each local neighborhood, and perform linear superposition based on the local statistical properties to obtain the enhanced dimensional difference. Calculate the geometric mean of the enhanced dimensional difference to obtain the comprehensive defect confidence. Step S4: Extract the defect region based on the comprehensive defect confidence level, generate the defect evolution sequence, associate the evolution trajectory of the same defect based on the defect evolution sequence, analyze each evolution trajectory to obtain defect trajectory features, conduct detailed analysis of the defect trajectory features, establish a correlation model between the defect trajectory features and the time series data of the printing equipment's process parameters, and obtain the cause of the defect based on the correlation model.

2. The method according to claim 1, characterized in that, Generate a virtual reference image, including: The process involves acquiring original digital design drafts and material physical property data, including parameters such as gloss, roughness, ink absorption, and thickness of the packaging material. Using ray tracing algorithms, the actual lighting conditions on the production line and the reflection, refraction, and scattering behavior of light on the material surface are simulated to generate multiple virtual reference images. These virtual reference images contain realistic highlights, shadows, and color gradients caused by minute material deformations.

3. The method according to claim 1, characterized in that, Generate a dynamic reference image, including: Images of similar products that have been deemed qualified are selected from historical inspection data. Image registration and fusion techniques are used to align and fuse the product images to generate template images. Then, a conditional generative adversarial network is used to generate dynamic reference images that match the current production status based on the template images and the current equipment status.

4. The method according to claim 1, characterized in that, Generating a multidimensional difference mass field includes: The three-dimensional physical property field generated based on the multimodal reference image is used as the reference three-dimensional physical property field. Spatiotemporal registration is performed between the dynamic three-dimensional physical property field and the reference three-dimensional physical property field. After registration is completed, the differences in color dimension, texture dimension, shape dimension and optical property dimension are calculated. The calculated differences in the four dimensions are superimposed in the same three-dimensional spatial coordinate system to generate a four-dimensional difference mass field.

5. The method according to claim 4, characterized in that, Calculating the difference in topographic dimensions includes: On the registered dynamic 3D physical property field and the reference 3D physical property field, a local window centered on each pixel is defined, and for each local window, the height distribution statistics within the local window are calculated. For each pixel, the absolute difference in statistics between the 3D physical property field and the reference 3D physical property field is calculated. The absolute difference is then merged into a single shape difference value through a fusion function. The combination of shape difference values ​​corresponding to all pixels is used as the corresponding shape dimension difference.

6. The method according to claim 1, characterized in that, The enhanced dimensionality difference is obtained by performing a linear superposition based on statistical properties, including: Directional excitation is established based on predefined physical causal logic, texture dimension difference is enhanced based on color dimension difference, shape dimension difference is enhanced based on texture dimension difference, and optical property dimension difference is enhanced based on shape dimension difference.

7. The method according to claim 1, characterized in that, The evolutionary trajectories of the same defect are associated based on the defect evolution sequence, including: Based on the comprehensive defect confidence score of each defect in the defect evolution sequence, the corresponding defect region is obtained. An overlap rate threshold is defined, and the overlap rate of the defect regions of two adjacent printing materials is calculated. If the overlap rate is greater than the overlap rate threshold, the two adjacent printing materials are considered to belong to the same defect. The defect regions of multiple consecutive printing materials belonging to the same defect are combined to form a complete defect evolution trajectory.

8. The method according to claim 1, characterized in that, A detailed analysis of the defect trajectory characteristics was conducted, including: The defect trajectory characteristics include life cycle, confidence intensity change curve, and spatial movement path. The defect type is distinguished based on the length of the life cycle, the severity of the problem is assessed based on the defect type, the development stage of the defect is determined based on the confidence intensity change curve and a predictive maintenance basis is provided, and the fault source is accurately located based on the spatial movement path.

9. A machine vision-based packaging material printing quality inspection system, used to implement the machine vision-based packaging material printing quality inspection method as described in any one of claims 1-8, characterized in that, The system includes: The acquisition and processing module acquires multispectral image sequences of moving packaging and printing materials under different lighting conditions, obtains three-dimensional morphological data of the packaging material surface in real time, and constructs a dynamic three-dimensional physical property field based on the multispectral image sequences and three-dimensional morphological data. The differential calculation module generates virtual and dynamic reference images, constructs a multimodal reference image, performs spatiotemporal registration of the dynamic three-dimensional physical property field with the multimodal reference image, and calculates the differences between the two in different dimensions to generate a multidimensional differential mass field. The defect detection module defines a local neighborhood for each single-dimensional difference in the multidimensional differential mass field, calculates the local statistical characteristics of the single-dimensional difference in each local neighborhood, and performs linear superposition based on the local statistical characteristics to obtain the enhanced dimensional difference. The geometric mean of the enhanced dimensional difference is then calculated to obtain the comprehensive defect confidence. The deep analysis module extracts defect regions based on comprehensive defect confidence, generates defect evolution sequences, associates the evolution trajectories of the same defect based on the defect evolution sequences, analyzes each evolution trajectory to obtain defect trajectory features, conducts detailed analysis of defect trajectory features, establishes a correlation model between defect trajectory features and time-series data of printing equipment process parameters, and obtains the cause of defects based on the correlation model.

Images