An intelligent printing quality detection method based on image recognition

Abstract

The present application relates to the technical field of intelligent printing quality detection, in particular to an intelligent printing quality detection method based on image recognition; containing image preprocessing, physical model construction, deformation iteration calculation, energy decoupling analysis and closed-loop feedback control module; the method generates a feature point matrix and a to-be-measured feature field by acquiring standard and real-time image data; the core is to construct a virtual tensor spring network model, map image differences to an external force field, and use finite element iteration to solve deformation potential energy distribution; the Laplacian operator is used to perform energy decoupling, separate low-frequency isotropic stress and high-frequency topological potential singular points, and generate a residual heat map to locate defects and adjust transmission tension accordingly; the present application realizes the transformation from simple image comparison to physical potential evolution, breaks the mutual exclusion limit of deformation tolerance and detection sensitivity, and can effectively filter out nonlinear deformation interference in high-speed transmission of flexible materials.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent printing quality detection, in particular to an intelligent printing quality detection method based on image recognition. BACKGROUND

[0002] In the modern high-speed gravure and flexographic printing production environment, online quality detection of flexible substrates such as plastic film and paper is a core link to ensure production efficiency.

[0003] To realize automatic quality inspection, existing solutions generally use visual detection technology based on pixel gray scale contrast or traditional pattern recognition. Although such solutions have certain processing capabilities in low-speed scenarios, due to the non-linear physical deformation and random tension fluctuations of flexible materials during high-speed transmission, there is a significant displacement deviation between the to-be-detected image and the reference master version. Existing algorithms cannot effectively decouple normal physical stretching and real printing defects at the mathematical level, often leading to technical contradictions between high robustness and high sensitivity, that is, improving deformation tolerance will cause defect omission, and strengthening small feature capture will produce a large number of false positives. In addition, existing technologies focus on discrete image analysis and lack real-time feedback mechanisms for converting detection results into physical mechanism control instructions, making it difficult to dynamically correct the transmission state from the source to suppress deformation.

[0004] Therefore, how to realize intelligent tolerance of complex physical deformation and accurate positioning of topological defects, while improving the closed-loop cooperative control capability of the detection system and the physical transmission mechanism, has become a technical problem to be solved. SUMMARY

[0005] To solve the above technical problems, the present application provides an intelligent printing quality detection method based on image recognition. Specifically, the technical solution of the present application includes:

[0006] S1. Obtain standard reference image data and real-time image data to be detected, and pre-process the standard reference image data and the real-time image data to be detected to generate a standard feature point matrix and to-be-detected feature field data.

[0007] S2. Construct a virtual tensor spring network model based on the standard feature point matrix, define the virtual connection relationship and virtual spring coefficient between each quality node in the virtual tensor spring network model, and form an initial stable grid.

[0008] S3. Map the to-be-detected feature field data as an external force field to the virtual tensor spring network model, perform finite element iteration calculation, solve the deformation potential distribution generated by the virtual tensor spring network model to match the to-be-detected feature field data, and generate global potential field data.

[0009] S4. performing energy decoupling processing on the global potential field data, separating a low-frequency isotropic stress component and a high-frequency topological potential singularity component by using a Laplacian operator, and generating a potential residual heat map based on the high-frequency topological potential singularity component;

[0010] S5. locating a local anomaly coordinate according to the potential residual heat map and outputting a quality detection result, and generating a tension closed-loop feedback control instruction based on the low-frequency isotropic stress component to perform real-time parameter adjustment on a physical transmission mechanism.

[0011] Preferably, the S1 comprises the following steps:

[0012] S11. acquiring an original image signal of a continuous medium surface by a high-speed linear array acquisition device;

[0013] S12. performing hardware-level preprocessing on the original image signal by using a field programmable gate array to generate digitized standard reference image data and real-time image data to be detected;

[0014] S13. converting the standard reference image data into the standard feature point matrix and converting the real-time image data to be detected into the feature field data to be detected with force field attributes by a data conversion and encoding module running in a processing unit, the feature field data to be detected being used to represent a real-time physical deformation state.

[0015] Preferably, the S2 comprises the following steps:

[0016] S21. extracting key feature points in the standard feature point matrix as quality nodes to construct a grid topological structure;

[0017] S22. dynamically determining virtual spring coefficients between the quality nodes according to local texture complexity of the standard reference image data, and assigning a virtual spring coefficient with a larger value to a region with a texture complexity higher than a preset complexity threshold;

[0018] S23. establishing a virtual tensor spring network model configured to simulate an elastic deformation response of a continuous medium under the action of an external force, rather than directly calculating a pixel gray difference.

[0019] Preferably, the S3 comprises the following steps:

[0020] S31. regarding the feature field data to be detected as a continuous external force field applied to the initial stable grid;

[0021] S32. Running finite element analysis algorithm in the graphics processor core, performing multiple fast iterations on the virtual tensor spring network model, calculating the total potential energy required for the initial steady-state grid to reach a matching state under the action of the external force field;

[0022] S33. Recording the stress distribution state of each node inside the grid, generating the global potential energy field data containing global potential energy distribution information, which reflects the physical energy required to stretch the standard reference image data to match the real-time image data to be detected.

[0023] Preferably, the S4 comprises the following steps:

[0024] S41. Applying Laplacian operator to the global potential energy field data for spatial filtering processing;

[0025] S42. Extracting the low-frequency energy part in the filtering result as the low-frequency isotropic stress component, which represents the overall physical deformation of the continuous medium;

[0026] S43. Extracting the high-frequency residual part in the filtering result as the high-frequency topological potential singular point component, which represents the topological tearing or local potential energy mutation that cannot be explained by continuous physical stretching;

[0027] S44. Mapping the high-frequency topological potential singular point component to the visual potential residual heat map, where the heat value represents the severity of local defects.

[0028] Preferably, the S5 comprises the following steps:

[0029] S51. Setting an abnormality determination threshold, identifying the area in the potential residual heat map where the heat value is greater than the abnormality determination threshold;

[0030] S52. Reverse mapping the identified area to calculate its position in the original image coordinate system, marking it as a defect coordinate, and outputting a rejection signal to the controller;

[0031] S53. Analyzing the distribution characteristics of the low-frequency isotropic stress component, calculating the tension deviation value in the current physical transmission process;

[0032] S54. Converting the tension deviation value into a servo motor adjustment signal as the tension closed-loop feedback control instruction, dynamically adjusting the roller speed or pressure on both sides of the physical transmission mechanism to eliminate nonlinear deformation.

[0033] Preferably, the abnormality determination threshold in S51 is dynamically adjusted according to the following rules:

[0034] If the average of the low-frequency isotropic stress component is greater than a preset stress reference value, it indicates that the overall deformation is large, and the abnormality determination threshold is automatically increased to reduce the false positive rate.

[0035] If the average of the low-frequency isotropic stress component is less than or equal to the stress reference value, it indicates that the overall operation is stable, and the preset reference abnormality determination threshold is maintained.

[0036] Preferably, the method is applied to a flexible substrate detection scene with a continuous physical surface, and is configured to perform the following determination logic:

[0037] Deformation tolerance determination: for the difference identified as the low-frequency isotropic stress component, even if it causes pixel-level mismatch, it is determined as normal physical deformation, and no defect alarm is triggered;

[0038] Topology tear determination: for the difference identified as the high-frequency topological potential singularity component, no matter how large the pixel gray scale change amplitude is, as long as the grid topological continuity is destroyed, it is determined as a printing quality defect;

[0039] Real-time control: the processing and feedback control of a single frame image are limited to be completed within a preset time window to match the operation cycle of the physical transmission mechanism.

[0040] Compared with the prior art, the present application has the following beneficial effects:

[0041] 1. The method creatively converts image differences into physical potential fields by constructing a virtual tensor spring network model, and realizes the orthogonal separation of normal physical deformation and real printing defects at the mathematical level by using mathematical operators; this method breaks the limitation that deformation tolerance and detection sensitivity are mutually restricted in traditional detection schemes, and can effectively filter out nonlinear physical deformation interference caused by high-speed transmission of flexible materials, while ensuring high tolerance for material stretching, it can accurately capture small topological defects.

[0042] 2. The method not only identifies defects, but also uses the decoupled low-frequency isotropic stress component to analyze the tension deviation in the physical transmission process in real time, and generates closed-loop feedback control instructions to dynamically adjust the transmission mechanism; this mechanism realizes the deep cooperation of the detection system and the production mechanism, can dynamically correct the transmission state and suppress deformation from the source, thereby changing from simply rejecting waste to reducing waste production, and significantly improving the stability of the whole production process.

[0043] 3. By introducing dynamic spring coefficients based on texture complexity and adaptive decision threshold logic, the method can simulate the cognitive characteristics of human observers, performing strict monitoring on sensitive areas with rich details, and showing intelligent tolerance on flat areas; this adaptive mechanism enables the system to adjust sensitivity in real time according to line speed fluctuations or material batch changes, maintaining extremely low false positive rates and extremely high detection rates under complex and variable production conditions.

[0044] 4. The method uses hardware-level preprocessing and high-performance computing architecture, through the parallel cooperation of field programmable gate arrays and graphics processors, to ensure that massive image data completes the whole process of acquisition, physical simulation and feedback control within a time window of milliseconds; this efficient computing chain enables complex finite element analysis to perfectly match the operating rhythm of modern high-speed printing machines, solving the problem of real-time high-precision algorithms in large-scale industrial production. BRIEF DESCRIPTION OF DRAWINGS

[0045] The present application will be further explained in conjunction with the accompanying drawings and examples:

[0046] Figure 1 is a flow chart of the method of the present application. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in conjunction with specific examples.

[0048] Example 1:

[0049] Please refer to Figure 1 An intelligent printing quality detection method based on image recognition, comprising the following steps:

[0050] S1. Obtain standard reference image data and real-time image data to be detected, and pre-process the standard reference image data and real-time image data to be detected to generate a standard feature point matrix and a feature field data to be detected;

[0051] S2. Construct a virtual tensor spring network model based on the standard feature point matrix, define the virtual connection relationship and virtual spring coefficient between each quality node in the virtual tensor spring network model, and form an initial stable grid;

[0052] S3. Map the feature field data to be detected as an external force field to the virtual tensor spring network model, perform finite element iterative calculation, solve the deformation potential distribution generated by the virtual tensor spring network model to match the feature field data to be detected, and generate global potential field data;

[0053] S4. Energy decoupling processing is performed on the global potential energy field data, a Laplace operator is used to separate a low-frequency isotropic stress component and a high-frequency topological potential energy singularity component, and a potential energy residual heat map is generated based on the high-frequency topological potential energy singularity component;

[0054] S5. Local abnormal coordinates are located according to the potential energy residual heat map, and a quality detection result is output, and a tension closed-loop feedback control instruction is generated based on the low-frequency isotropic stress component, and real-time parameter adjustment is performed on the physical transmission mechanism.

[0055] The embodiment discloses a general flow of an intelligent printing quality detection method, which is applied to an online detection system of a high-speed gravure printing machine or a flexographic printing machine;

[0056] S1, data acquisition and preprocessing, the system acquires standard reference image data and real-time image data to be detected on the current production line in parallel through an industrial camera; the standard reference image data refers to a digital master plate pre-stored and confirmed by manual confirmation to be defect-free, and the standard reference image data is used as a static reference for physical simulation; the real-time image data to be detected refers to flow data collected in real time during the production process and containing potential deformation and defects; the preprocessing module aligns and converts the two into a data format suitable for physical calculation to generate a standard feature point matrix and feature field data to be detected;

[0057] S2, a virtual physical model is constructed, a virtual tensor spring network model is constructed by the system based on the standard feature point matrix; the virtual tensor spring network model refers to a calculation architecture that does not directly compare pixel gray values, but simulates material physical characteristics; in the model, the feature points of the image are defined as quality nodes, and the nodes are connected by virtual springs; the key of this step is to define the virtual connection relationship between the quality nodes and the virtual spring coefficients, so as to form an initial stable grid in the lowest energy state;

[0058] S3, finite element potential energy calculation, the system regards the feature field data to be detected as an external force field and forcibly maps it to the initial stable grid; the energy accumulated in the entire network when the grid is deformed to match the data to be detected is calculated by using a finite element analysis algorithm, and global potential energy field data is generated; the global potential energy field data refers to a matrix, and the value of each element represents the physical deformation potential energy generated at the position to match the real-time image;

[0059] S4, energy decoupling and feature separation, which is the core step of the application; the system performs energy decoupling processing on the global potential energy field data; using Laplace operator as a spatial filter, the total energy field is separated into two orthogonal components: low-frequency isotropic stress component representing large-scale, smooth stretching or twisting; high-frequency topological potential singularity component representing local, abrupt energy peaks; based on the latter, the system generates a potential energy residual heat map, which directly reflects the distribution of non-physical differences, i.e. defects;

[0060] S5, result output and closed-loop control, the system locates the local abnormal coordinates according to the potential energy residual heat map, and outputs an alarm or rejection signal if the heat value exceeds the standard; at the same time, the system creatively uses the separated low-frequency isotropic stress component to generate tension closed-loop feedback control instructions, directly adjusting the physical transmission mechanism such as roller, to suppress the deformation from the source;

[0061] By converting image differences into potential energy fields, the application successfully performs orthogonal separation of normal deformation as low-frequency stress and printing defects as high-frequency potential singularities in mathematics; this allows the system to tolerate soft material nonlinear deformation while still accurately capturing subtle defects, solving the technical contradiction in the prior art that high robustness inevitably leads to low sensitivity.

[0062] Embodiment 2:

[0063] S1 includes the following steps:

[0064] S11. Obtain the original image signal of the continuous medium surface by the high-speed linear array acquisition device;

[0065] S12. Use the field programmable gate array to perform hardware-level preprocessing on the original image signal to generate digitized standard reference image data and real-time image data to be detected;

[0066] S13. Convert the standard reference image data into a standard feature point matrix by running a data conversion and encoding module in the processing unit, and convert the real-time image data to be detected into a feature field data to be detected with force field attributes, which is used to represent the real-time physical deformation state.

[0067] This embodiment is a specific implementation of data acquisition and preprocessing, focusing on hardware acceleration and data flow construction;

[0068] S11, image acquisition, this embodiment uses a high-speed linear array acquisition device to obtain the original image signal of the continuous medium surface; the high-speed linear array acquisition device specifically selects a linear array camera with a line frequency of The linear array camera transmits data through or interface to adapt to a high-speed production line of per minute;​

[0069] S12. FPGA hardware preprocessing: To reduce latency, this embodiment utilizes a field-programmable gate array (FPGA). Hardware-level preprocessing of the raw image signal; Internally configured with a parallel pipeline, it performs illumination correction, noise reduction, and distortion correction, directly generating digitized standard reference image data and real-time image data to be detected at the bitstream level, without requiring a host computer. resource;

[0070] S13. Force field attribute conversion: The image data is converted into a physical model input through a preset feature extraction and encoding software algorithm module. Specifically, the algorithm module converts the texture gradient of the standard image into a standard feature point matrix as grid node coordinates, and converts the real-time image data to be detected into the feature field data to be measured. In this embodiment, the feature field data to be measured is defined as a vector field, the value of which represents the optical flow vector or gray-level gradient vector of the corresponding pixel in the real-time image, used to characterize the real-time physical deformation state. Based on this, the vector field is converted into a physical model input through a preset conversion operator. Mapped as external force The mapping logic follows the formula ,in, These are force field mapping coefficients used to convert pixel displacement intensity into mechanical amplitudes in the physical simulation domain, within this virtual physical model. The unit is defined as a virtual energy unit;

[0071] pass Hardware anchoring with a high-speed camera ensures that the throughput of data acquisition and preprocessing meets the above industrial requirements, providing a low-latency real-time data stream for subsequent physics field calculations.

[0072] Example 3:

[0073] S2 includes the following steps:

[0074] S21. Extract key feature points from the standard feature point matrix and define them as quality nodes to construct a gridded topology;

[0075] S22. Based on the local texture complexity of the standard reference image data, dynamically determine the virtual spring coefficient between quality nodes, and assign a larger virtual spring coefficient to regions with texture complexity higher than the preset complexity threshold;

[0076] S23. Establish a virtual tensor spring network model. The virtual tensor spring network model is configured to simulate the elastic deformation response of a continuous medium under external force, rather than directly calculating pixel grayscale differences.

[0077] This embodiment is a concretization of the construction of a virtual physical model, and details the virtual tensor spring network. The construction logic is key to achieving intelligent deformation tolerance;

[0078] S21. Mesh Topology Construction; The system extracts key feature points from the standard feature point matrix, such as edge corners and texture centers, and defines them as quality nodes, based on... Triangulation algorithm constructs a meshed topology;

[0079] S22. Dynamic definition of virtual spring coefficient; To accurately simulate the physical properties of printed materials, this embodiment introduces a texture-adaptive stiffness model; For any connected edge in the mesh, its virtual spring coefficient... The specific calculation formula is as follows:

[0080]

[0081] in, For texture weighting factors, It is the basic stiffness constant;

[0082] Among them, nodes Local texture complexity at the location The calculation method is as follows:

[0083]

[0084] in, For pixels grayscale value, This represents the average gray level within that neighborhood.

[0085] Areas with complex textures, such as text and intricate patterns, are extremely sensitive to deformation; even the slightest misalignment should be considered a defect, and therefore require greater precision. The value represents high stiffness; while flat textured areas, such as large color blocks, are not sensitive to slight stretching and are assigned a smaller value. The value represents low stiffness;

[0086] S23. Physical simulation configuration; Establish a virtual tensor spring network model; This model is configured to simulate the elastic deformation response of a continuous medium under external force; Unlike the traditional algorithm that directly subtracts pixel values, this model calculates the internal forces required to maintain the integrity of the topology.

[0087] The spring coefficient is dynamically determined by the texture complexity. The model can understand images like a human observer: it is strict (high stiffness) in areas with rich detail and lenient (low stiffness) in areas with background. This biomimetic design significantly reduces the false alarm rate caused by the natural stretching of materials.

[0088] Example 4:

[0089] S3 comprises the following steps:

[0090] S31. The to-be-tested feature field data is regarded as a sustained external force field applied to the initial steady-state grid;

[0091] S32. A finite element analysis algorithm is run in a graphics processor core to perform multiple rapid iterations on the virtual tensor spring network model to calculate the total potential energy required for the initial steady-state grid to reach a matching state under the action of the external force field;

[0092] S33. The stress distribution state of each node in the grid is recorded to generate global potential energy field data containing global potential energy distribution information, which reflects the physical energy required to stretch the standard reference image data to match the to-be-tested real-time image data.

[0093] This embodiment is a specific implementation of finite element potential energy calculation, which describes how to solve the potential energy field in the by finite element analysis;

[0094] S31, external force field application, the system regards the to-be-tested feature field data as a sustained external force field applied to the initial steady-state grid; that is, it is assumed that the to-be-tested image is the result of the standard image after being subjected to certain mechanical action, and this step aims to reverse the size of such force;

[0095] S32, GPU accelerated iteration; a simplified finite element analysis algorithm is run in the core of a graphics processor; the system performs rapid iterations on the virtual tensor spring network model, aiming to solve the total potential energy minimization problem; if the preset energy convergence threshold is not reached within the preset number of iterations, the state with the lowest total potential energy in the current iteration period is taken as the approximate matching solution to prioritize the real-time control requirements of the high-speed production line;

[0096] In each iteration process, the update of the coordinate position of the node follows the force balance equation:

[0097]

[0098] wherein, is a preset virtual node mass, is a virtual damping coefficient, the values of both are pre-set based on the unit consistency of the virtual physical domain to adjust the smoothness of the simulation evolution; and represent the first-order velocity vector and the second-order acceleration vector of the node in the virtual physical simulation evolution process, respectively; is external force; the calculation of the global potential energy field data is based on the following generalized formula of Hooke's law:

[0099]

[0100] wherein, is the cumulative potential energy at node , in virtual energy units, obtained by iterative calculation; is the set of neighborhood nodes of node ; is the spatial coordinate position of node and after iteration; is the initial Euclidean distance of node and in the standard reference image;

[0101] By parallel computing the above formula in , the system can complete the evolution simulation of the physical field within ; the global potential energy field data quantitatively reflects the physical energy required to stretch the standard image into the real-time image, providing a unique physical benchmark for subsequent differentiation between normal stretching and abnormal defects.

[0102] Embodiment 5:

[0103] S4 comprises the following steps:

[0104] S41. Apply Laplace operator to the global potential energy field data for spatial filtering processing;

[0105] S42. Extract the low-frequency energy part in the filtering result as the low-frequency isotropic stress component, which represents the overall physical deformation of the continuous medium;

[0106] S43. Extract the high-frequency residual part in the filtering result as the high-frequency topological potential singular point component, which represents the topological tearing or local potential energy mutation that cannot be explained by continuous physical stretching;

[0107] S44. Map the high-frequency topological potential singular point component to a visual potential residual heat map, where the heat value represents the severity of the local defect.

[0108] This embodiment is a specific implementation of energy decoupling and feature separation, which describes how to use mathematical operators to automatically separate defects and deformations;

[0109] S41, Laplace spatial filtering, applying discrete Laplace operator to the global potential energy field data for spatial filtering; ;

[0110] Since the embodiment adopts the non-regular grid topology based on Delaunay triangulation, the discrete Laplace operator is defined in the form of graph Laplace; for the potential value at node , the output of the spatial filtering thereof is calculated as follows:

[0111]

[0112] wherein, is the degree of node , i.e. the number of neighbor nodes; is the neighbor node set of node ; the operation realizes the orthogonal separation of high and low frequency energy components by extracting the potential difference of each node and its neighborhood;

[0113] S42, extract the low-frequency isotropic stress, extract the low-frequency component in the filtering result; since the physical stretching of paper or film is usually globally smooth, it is represented as a low-frequency signal in the potential field; this part is extracted as a low-frequency isotropic stress component, representing the overall physical deformation of the continuous medium, such as the overall lengthening ;

[0114] S43, extract the high-frequency topological potential singularity, extract the high-frequency residual part in the filtering result; when there is a printing defect such as a dot or a missing print, the standard image cannot be changed into the to-be-tested image through continuous physical stretching, which will cause the grid to tear or greatly distort at the defect, thereby forming a singularity of energy mutation in the potential field ; this part is extracted as a high-frequency topological potential singularity component;

[0115] S44, heat map generation, map the high-frequency topological potential singularity component to a visual potential residual heat map; the deeper the color in the heat map, the more serious the topological tearing at that place, i.e. the greater the possibility of defects;

[0116] By utilizing the natural separation ability of the Laplace operator for high and low frequency signals, the present application realizes the automatic decoupling of deformation and defects; without complex deep learning training, only the mathematical and physical characteristics can accurately identify those image differences that cannot be explained by physical stretching, i.e. real defects.

[0117] Embodiment 6:

[0118] S5 includes the following steps:

[0119] S51. Set an abnormality judgment threshold, identify the area in the potential residual heat map where the heat value is greater than the abnormality judgment threshold;

[0120] ​S52. The identified area is inversely mapped to determine its position in the original image coordinate system, marked as defect coordinates, and an elimination signal is output to the controller;

[0121] S53. The distribution characteristics of the low-frequency isotropic stress component are analyzed, and the tension deviation value in the current physical transmission process is calculated;

[0122] S54. The tension deviation value is converted into a servo motor adjustment signal as a tension closed-loop feedback control instruction to dynamically adjust the roller speed or pressure on both sides of the physical transmission mechanism to eliminate nonlinear deformation.

[0123] This embodiment is a specific implementation of result output and closed-loop control, which details the application of detection results and the tension feedback control mechanism;

[0124] S61-S62, defect determination and elimination, set abnormality determination threshold; in the potential energy residual thermodynamic map, the area with a thermodynamic value is locked; the system inversely maps these areas to calculate their pixel positions in the original image coordinate system, marks them as defect coordinates, and outputs signals to drive the controller to perform waste elimination at the end of the production line;

[0125] S63, tension deviation calculation, using the separated low-frequency isotropic stress component, the system calculates its mean value in the horizontal and vertical direction to obtain the tension deviation value in the current physical transmission process; for example, if the low-frequency stress of the left area is generally higher than that of the right area, it indicates that the left roller tension is too large;

[0126] S64, closed-loop feedback control, convert the tension deviation value into a servo motor adjustment signal as a tension closed-loop feedback control instruction; the instruction is sent to the servo driver of the printing machine in real time to dynamically fine-tune the speed difference or pressure of the rollers on both sides of the physical transmission mechanism to eliminate nonlinear deformation;

[0127] The output speed compensation value of the tension closed-loop feedback control instruction The calculation formula uses a discrete PID algorithm:

[0128]

[0129] where, represents the current sampling time, the tension deviation value calculated at the historical sampling time ; and , respectively, are the proportional, integral, and derivative coefficients.

[0130] The embodiment is not only a quality inspection system, but also a process optimization system; it uses the by-product, i.e. low-frequency stress data, generated in the detection process to reversely optimize the production process, and realizes a technical leap from passive rejection of waste products to active reduction of waste product generation.

[0131] Embodiment 7:

[0132] The abnormality determination threshold in S51 is dynamically adjusted according to the following rules:

[0133] If the mean value of the low-frequency isotropic stress component is greater than the preset stress reference value, it indicates that the overall deformation is large, and the abnormality determination threshold is automatically increased to reduce the false positive rate;

[0134] If the mean value of the low-frequency isotropic stress component is less than or equal to the stress reference value, it indicates that the overall operation is stable, and the preset reference abnormality determination threshold is maintained.

[0135] The embodiment further limits the dynamic adjustment logic of the abnormality determination threshold to adapt to different working conditions;

[0136] In the embodiment, the abnormality determination threshold is not a fixed value, but is dynamically adjusted according to the following adaptive threshold formula:

[0137]

[0138] wherein, is the dynamic abnormality determination threshold used for the current frame; is the preset reference threshold; is the mean value of the low-frequency isotropic stress component of the current frame; is the preset stress reference value; is the adjustment coefficient;

[0139] wherein, the stress reference value is obtained as follows: in the no-load idle state of the production line, 100 standard images are continuously collected and the mean value of the global potential field is calculated, and the maximum value is taken as 1.2 times the safety reference; the adjustment coefficient is an empirical constant obtained by least squares fitting of the response curve of the known defect sample under different tensile stresses, and in the embodiment, the value range is ;

[0140] If , i.e. the mean value of the low-frequency stress is greater than the reference value, it indicates that the material is currently in a large deformation state such as an acceleration stage or a material batch change, at which time the system automatically increases the determination threshold; the purpose is to prevent edge false positives caused by global severe stretching and reduce the false positive rate; if , it indicates that the operation is stable, the preset reference threshold is maintained, and the high sensitivity is maintained.

[0141] The dynamic threshold mechanism gives the system environment adaptive ability, ensuring that the system will not produce a large number of false positives when the line speed fluctuates or the material is shaken violently, and can maintain a high micro-defect detection rate when running smoothly.

[0142] Embodiment 8:

[0143] The method is applied to the detection scene of flexible substrates with continuous physical surfaces, and is configured to perform the following decision logic:

[0144] Deformation tolerance decision: for the difference identified as a low-frequency isotropic stress component, even if it causes pixel-level mismatch, it is determined as normal physical deformation, and no defect alarm is triggered;

[0145] Topological tear decision: for the difference identified as a high-frequency topological potential singularity component, no matter how large the pixel gray scale change amplitude is, as long as it breaks the grid topological continuity, it is determined as a printing quality defect;

[0146] Real-time control: the processing and feedback control of a single frame image are limited to be completed within a preset time window to match the running tempo of the physical transmission mechanism.

[0147] This embodiment defines the applicable scene and core decision logic of the method;

[0148] The method is specially configured to be applied to the detection scene of flexible substrates with continuous physical surfaces such as plastic film, paper, and aluminum foil;

[0149] Deformation tolerance decision: for the difference separated as a low-frequency isotropic stress component, even if it causes a pixel displacement visible to the naked eye such as a pattern shift of pixels, the system determines it as normal physical deformation and does not trigger a defect alarm; Topological tear decision: for the difference separated as a high-frequency topological potential singularity component, no matter how large the pixel gray scale change amplitude is, as long as it breaks the grid topological continuity, i.e. the energy gradient is extremely large, the system determines it as a printing quality defect;

[0150] The system forces the processing of a single frame image, i.e. and the feedback control to be completed within a preset time window, for example , to strictly match the running tempo of the physical transmission mechanism of more than meters per minute;

[0151] Through clear logical definition, this embodiment ensures the usability of the algorithm in an industrial site, avoids misjudging harmless physical stretching as a quality problem, and at the same time guarantees the real-time response capability of a high-speed production line.

[0152] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for intelligent printing quality inspection based on image recognition, characterized in that, Includes the following steps: S1. Acquire standard reference image data and real-time image data to be detected, and preprocess the standard reference image data and the real-time image data to be detected to generate a standard feature point matrix and feature field data to be tested; S2. Construct a virtual tensor spring network model based on the standard feature point matrix, define the virtual connection relationship and virtual spring coefficient between each mass node in the virtual tensor spring network model, and form an initial steady-state grid; S3. Map the data of the feature field to be measured as an external force field to the virtual tensor spring network model, perform finite element iterative calculation, solve the deformation potential energy distribution generated by the virtual tensor spring network model to match the data of the feature field to be measured, and generate global potential energy field data. S4. Perform energy decoupling processing on the global potential energy field data, use the Laplace operator to separate the low-frequency isotropic stress component and the high-frequency topological potential energy singularity component, and generate a potential energy residual thermogram based on the high-frequency topological potential energy singularity component. S5. Locate the coordinates of local anomalies based on the potential energy residual thermogram and output the quality detection results. At the same time, generate tension closed-loop feedback control commands based on the low-frequency isotropic stress components to adjust the parameters of the physical transmission mechanism in real time. S2 includes the following steps: S21. Extract the key feature points from the standard feature point matrix and define them as quality nodes to construct a gridded topology; S22. The virtual spring coefficient between the quality nodes is dynamically determined based on the local texture complexity of the standard reference image data, and a larger virtual spring coefficient is assigned to regions with texture complexity higher than a preset complexity threshold. S23. Establish the virtual tensor spring network model, which is configured to simulate the elastic deformation response of a continuous medium under external force, rather than directly calculating pixel grayscale differences. S4 includes the following steps: S41. Spatial filtering is performed on the global potential energy field data using the Laplace operator; S42. Extract the low-frequency energy component from the filtering result as the low-frequency isotropic stress component, wherein the low-frequency isotropic stress component characterizes the overall physical deformation of the continuous medium. S43. Extract the high-frequency residual part from the filtering result as the high-frequency topological potential singularity component, the high-frequency topological potential singularity component characterizing topological tearing or local potential abrupt change that cannot be explained by continuous physical stretching. S44. Map the high-frequency topological potential singularity component to a visualized potential energy residual thermogram, where the thermogram values ​​represent the severity of local defects.

2. The intelligent printing quality inspection method based on image recognition according to claim 1, characterized in that, S1 includes the following steps: S11. Acquire the original image signal of the continuous medium surface through a high-speed linear array acquisition device; S12. The original image signal is preprocessed at the hardware level using a field-programmable gate array to generate digitized standard reference image data and real-time image data to be detected; S13. The standard reference image data is converted into the standard feature point matrix by the data conversion and encoding module running in the processing unit, and the real-time image data to be detected is converted into the test feature field data with force field attributes. The test feature field data is used to characterize the real-time physical deformation state.

3. The intelligent printing quality inspection method based on image recognition according to claim 2, characterized in that, S3 includes the following steps: S31. The measured feature field data is regarded as a continuous external force field applied to the initial steady-state grid; S32. Run the finite element analysis algorithm in the graphics processor core, perform multiple fast iterations on the virtual tensor spring network model, and calculate the total potential energy required for the initial steady-state mesh to reach the matching state under the action of the external force field; S33. Record the stress distribution state of each node inside the mesh to generate global potential energy field data containing global potential energy distribution information. The global potential energy field data reflects the physical energy required to stretch the standard reference image data to match the real-time image data to be detected.

4. The intelligent printing quality inspection method based on image recognition according to claim 3, characterized in that, S5 includes the following steps: S51. Set an anomaly detection threshold and identify regions in the potential energy residual thermal map where the thermal value is greater than the anomaly detection threshold. S52. Perform inverse mapping on the identified region, calculate its position in the original image coordinate system, mark it as defect coordinates, and output the rejection signal to the controller; S53. Analyze the distribution characteristics of the low-frequency isotropic stress components and calculate the tension deviation value in the current physical transmission process; S54. The tension deviation value is converted into a servo motor adjustment signal, which serves as the tension closed-loop feedback control command to dynamically adjust the rotational speed or pressure of the rollers on both sides of the physical transmission mechanism to eliminate nonlinear deformation.

5. The intelligent printing quality inspection method based on image recognition according to claim 4, characterized in that, The anomaly detection threshold in S51 is dynamically adjusted according to the following rules: If the mean value of the low-frequency isotropic stress component is greater than the preset stress reference value, it indicates that the overall deformation is large, and the anomaly judgment threshold is automatically increased to reduce the false alarm rate. If the mean value of the low-frequency isotropic stress component is less than or equal to the stress reference value, it indicates that the overall operation is stable and the preset reference anomaly judgment threshold is maintained.

6. The intelligent printing quality inspection method based on image recognition according to claim 1, characterized in that, The method is applied to the inspection of flexible substrates with continuous physical surfaces and is configured to execute the following decision logic: Deformation tolerance determination: For the difference identified as the low-frequency isotropic stress component, even if it causes pixel-level mismatch, it is determined as normal physical deformation and no defect alarm is triggered; Topological tear determination: For differences identified as high-frequency topological potential energy singularity components, regardless of the magnitude of pixel grayscale change, as long as the mesh topological continuity is disrupted, it is determined to be a printing quality defect. Real-time control: The processing and feedback control of a single frame image are limited to a preset time window to match the operating rhythm of the physical transmission mechanism.

Images