Printing defect detection method and system based on computer vision

Abstract

The invention relates to the technical field of image detection, and discloses a printing defect detection method and system based on computer vision. According to the method, a multi-mode imaging device collects a printing surface multi-channel fusion image, distortion is corrected through a dynamic segmentation module, a detection area mask is extracted, and a standard image is output. Inputting the standard image into a flaw detection model, extracting microscopic lattice points and macroscopic texture features, and matching with a template library to generate a probability distribution diagram; and performing multi-level judgment according to the image, quantitatively evaluating the flaws, and finally outputting flaw coordinates with space-time labels and type labels. The system comprises a multi-source acquisition module, a geometric correction module, a feature matching module, a grading decision-making module and a closed-loop control module, and real-time detection and feedback control of printing flaws are achieved. The method improves the detection precision and efficiency, is suitable for different types of printed matters, and has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of image detection, and particularly to a printing defect detection method and system based on computer vision. Background Art

[0002] In the printing industry, the quality of printed products is of crucial importance. However, traditional printing defect detection techniques often rely on manual visual inspection, which is not only time-consuming and laborious, but also easily affected by human factors, resulting in low detection accuracy and efficiency. With the continuous development of computer vision technology, more and more researchers have begun to attempt to apply it to printing defect detection. However, existing printing defect detection techniques based on computer vision still have some problems, such as insufficient detection accuracy and poor adaptability to complex printed products. Therefore, there is an urgent need for a more efficient and accurate printing defect detection method to meet the needs of the printing industry. Summary of the Invention

[0003] Aiming at the above-mentioned existing technical deficiencies, the purpose of the present invention is to provide a printing defect detection method and system based on computer vision, which solves the problems of low detection accuracy and poor adaptability to complex printed products in the prior art.

[0004] To solve the above technical problems, the present invention adopts the following technical solutions: In the first aspect, the present invention provides a printing defect detection method based on computer vision, and the method includes: Collecting a multi-channel fusion image of the printing surface containing ink physical characteristic parameters through a multi-modal imaging device, and the multi-modal imaging device uses a multi-spectral composite light source for sequential excitation; Inputting the multi-channel fusion image of the printing surface into a dynamic segmentation module, correcting mechanical transmission distortion based on preset dot parameters and overprint reference lines, extracting a detection area mask through morphological closing operation, and outputting a standard image after geometric alignment; Inputting the geometrically aligned standard image into a defect detection model, extracting microscopic dot features and macroscopic texture features, and generating a probability distribution map containing the position and confidence information of abnormal areas through cross-scale matching with a template library; Performing multi-level discrimination according to the probability distribution map, where the edge diffusion deviation amount is quantitatively evaluated at the dot level, and the topological structure is analyzed at the pattern level, and finally outputting the printing defect coordinates and type labels with spatio-temporal annotations; Mapping the printing defect coordinates to the printing unit phase coordinate system, triggering the sorting mechanism to perform phase synchronization rejection operation, and associating ink pressure adjustment parameters and temperature control parameters to generate a closed-loop feedback instruction.

[0005] Preferably, in a possible implementation manner of the first aspect, the multimodal imaging device adopts an LED light source group distributed in an annular array, including a multispectral composite light source in the visible light band, near-infrared band, and short-wave infrared band; The imaging unit consists of a high-resolution area array CCD camera and a multi-channel filter wheel, and constructs a multi-channel fusion image containing physical characteristic parameters such as ink thickness, curing degree, and adhesion by synchronously collecting reflectivity, fluorescence intensity, and thermal radiation response data.

[0006] Preferably, in a possible implementation manner of the first aspect, the dynamic segmentation module specifically includes: Generating a reference grid template based on preset dot parameters, where the dot parameters include screen angle, dot shape, and screen ruling; Detecting the actual position of the overprint reference line through the Hough transform, calculating the offset vector between its actual position and the theoretical position, and using an affine transformation matrix to correct the stretching and rotation distortions caused by mechanical transmission; Performing multi-scale morphological closing operations on the corrected image, selecting a circular structural element to perform dilation and erosion operations with radii of 3 pixels and 5 pixels in sequence, eliminating local noise and filling microscopic pores, and generating a detection area mask with a complete connected domain; Performing a bitwise operation alignment on the mask and the original image, and outputting a standard image with eliminated position deviation, whose pixel coordinate system establishes a sub-pixel level mapping relationship with the encoder signal of the printing unit.

[0007] Preferably, in a possible implementation manner of the first aspect, the construction of the affine transformation matrix satisfies the following relationship:

[0008] where , are the image pixel coordinates, is the angle deviation amount of the overprint reference line, , are the translation compensation amounts; The mathematical expression of the morphological closing operation is:

[0009] where is the output image matrix after the morphological closing operation, is the input image matrix, is the structural element, represents the dilation operation, represents the erosion operation.

[0010] Preferably, in a possible implementation manner of the first aspect, the working process of the defect detection model includes: A dual-branch deep convolutional neural network is adopted. The first branch extracts microscopic dot morphology features through a 3×3 small kernel convolutional layer to capture edge serration, dot area ratio, and ink diffusion coefficient; The second branch extracts macroscopic texture features through a dilated convolutional layer to analyze color difference distribution, moiré fringes, and overprint misregistration; The multi-scale features are input into a cross-modal attention mechanism module for similarity matching with standard samples in the template library. The template library establishes a hierarchical index according to printing color plates and printing material types; The probability distribution map of the abnormal area is reconstructed through a deconvolution layer, and its confidence value is normalized to the interval by the Softmax function.

[0011] Preferably, in a possible implementation manner of the first aspect, the similarity matching adopts the following metric method: Microscopic feature similarity:

[0012] Macroscopic feature similarity:

[0013] where is the th microscopic feature value of the test sample, is the th standard microscopic feature value in the template library, is the microscopic feature dimension, is the macroscopic feature vector of the test sample, is the standard macroscopic feature vector in the template library, is the maximum Euclidean distance of all sample macroscopic feature vectors in the template library; The final probability value is determined by the weighted fusion formula where and are weight coefficients, , and when is greater than the preset probability threshold, it is determined as an abnormal area.

[0014] Preferably, in a possible implementation manner of the first aspect, the specific content of the multi-level discrimination includes: At the dot level, the contour tracking algorithm is executed on the abnormal area in the probability distribution map, the curvature extreme points are extracted along the dot edge, and the deviation amount between the actual contour and the theoretical Gaussian diffusion curve is calculated. When the deviation amount exceeds the deviation threshold, it is marked as an edge diffusion defect; At the pattern level, a topological skeleton graph of the abnormal area is constructed, and the node connectivity, the number of loops, and the subgraph isomorphism are analyzed through graph theory algorithms. When a key path break or an adjacency relationship mismatch is detected, it is marked as a structural defect; Generate spatio-temporal annotation data including the phase angle of the printing unit, defect size, and type code by integrating spatio-temporal information.

[0015] Preferably, in a possible implementation manner of the first aspect, the quantization formula for the edge diffusion deviation amount is:

[0016] where is the curvature function of the actual contour, is the curvature function of the ideal Gaussian diffusion curve, is the arc length of the th contour segment, is the total number of contour segments; The topological structure analysis adopts the adjacency matrix eigenvalue comparison method: The adjacency matrix eigenvalue of the standard pattern is , and the eigenvalue of the measured abnormal area is , then the topological similarity is calculated by the following formula:

[0017] where is the adjacency matrix eigenvalue of the th standard pattern, , is the eigenvalue of the th measured abnormal area, , is the number of adjacency matrix eigenvalues. When is less than the topological similarity threshold, it is determined as a structural defect.

[0018] Preferably, in a possible implementation manner of the first aspect, mapping the printing defect coordinates to the printing unit phase coordinate system includes: parsing the current printing unit phase angle through the encoder signal, and converting the defect coordinates into the rejection trigger moment in the phase coordinate system, where is the current phase angle of the printing unit, is the angular velocity of the cylinder, is the delay compensation amount.

[0019] In a second aspect, the present invention provides a printing defect detection system based on computer vision. The system includes: A multi-source acquisition module that acquires a multi-channel fusion image of the printing surface containing ink physical property parameters through a multi-modal imaging device. The multi-modal imaging device uses a multi-spectral composite light source for sequential excitation; A geometric correction module, which is used to input the multi-channel fusion image of the printing surface into the dynamic segmentation module, correct the mechanical transmission distortion based on the preset dot parameters and overprint reference lines, extract the detection area mask through morphological closing operation, and output the standard image after geometric alignment; A feature matching module, which is used to input the standard image after geometric alignment into the defect detection model, extract the microscopic dot features and macroscopic texture features, and generate a probability distribution map containing the position and confidence information of the abnormal area through cross-scale matching with the template library; A hierarchical decision-making module, which performs multi-level discrimination according to the probability distribution map. Among them, the edge diffusion deviation amount is quantitatively evaluated at the dot level, and the topological structure is analyzed at the pattern level. Finally, the printing defect coordinates and type labels with spatio-temporal annotations are output; A closed-loop control module, which is used to map the printing defect coordinates to the phase coordinate system of the printing unit, trigger the sorting mechanism to perform phase synchronization rejection operation, and generate a closed-loop feedback instruction by associating the ink pressure adjustment parameter and the temperature control parameter.

[0020] The beneficial effects of the present invention are as follows: The multi-channel fusion image of the printing surface is collected by a multi-modal imaging device, and the dynamic segmentation module, defect detection model, hierarchical decision-making module, and closed-loop control module are used to realize the efficient and accurate detection of printing defects.

[0021] This method not only improves the detection accuracy and efficiency, but also has strong adaptability and robustness, and can be applied to different types of printed products and printing processes.

[0022] In addition, this method can also realize the real-time image detection and feedback control of printing defects, further strengthening the application value of image detection technology in the quality control and production optimization of printed products. Therefore, the present invention has broad application prospects and promotion value in the printing industry. Description of the Drawings

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0024] Figure 1 This is a flowchart of a printing defect detection method based on computer vision provided by this application.

[0025] Figure 2 This is a structural diagram of a printing defect detection system based on computer vision provided by this application.

[0026] Description of reference numerals: 1 - multi-source acquisition module, 2 - geometric correction module, 3 - feature matching module, 4 - hierarchical decision-making module, 5 - closed-loop control module. Detailed implementation manners

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] Embodiment 1: As Figure 1 shown, the present invention provides a printing defect detection method based on computer vision, including: Collecting a multi-channel fusion image of a printing surface containing ink physical characteristic parameters through a multi-modal imaging device, and the multi-modal imaging device uses a multi-spectral composite light source for sequential excitation.

[0029] In this embodiment, the multi-modal imaging device uses an LED light source group with a circular array distribution, and its spectral coverage range includes the visible light band (400 - 700 nm), the near-infrared band (700 - 1100 nm), and the short-wave infrared band (1100 - 2500 nm). The light source excitation modes of different bands are switched according to a preset time sequence through a time-sharing trigger control module. The light source group is distributed in a circular array to ensure that the irradiance deviation range of the illumination uniformity of each area on the printing surface reaches at the same time, the independent control of the light source intensity of each band is realized through pulse width modulation technology.

[0030] The imaging unit consists of a high-resolution area array CCD camera and a multi-channel filter wheel. The pixel size of the CCD camera is , and a 16-bit analog-to-digital converter is equipped to achieve high-dynamic-range imaging. The filter wheel is internally provided with eight groups of narrow-band filters, and the central wavelengths respectively correspond to the key spectral feature points for ink characteristic detection, including ink absorption peaks (such as 530 nm for magenta ink and 650 nm for cyan ink) and the fluorescence excitation wavelength of the printing material. During the image acquisition process, the imaging unit and the light source group are synchronously triggered through an encoder signal, and reflectivity data, fluorescence intensity data, and thermal radiation response data are sequentially captured with a time resolution of 5 ms. Among them, the reflectivity data is used to calculate the ink optical density, the fluorescence intensity data analyzes the ink curing degree through Stokes shift, and the thermal radiation response data inversely calculates the ink layer adhesion parameters based on the Stefan-Boltzmann law.

[0031] Through multi-source data fusion, map the three-channel physical characteristic parameters to a unified pixel coordinate system, and construct a multi-channel fusion image including ink thickness, curing degree, and adhesion. The fusion process adopts a weighted superposition method, where the ink thickness is calculated from the reflectance in the near-infrared band and the Mie scattering model, the curing degree is quantified by the fluorescence lifetime imaging technique to obtain the concentration of uncured monomers, and the adhesion is derived based on the empirical relationship between the thermal radiation relaxation time and the ink layer peeling strength.

[0032] Input the multi-channel fusion image of the printing surface into the dynamic segmentation module, correct the mechanical transmission distortion based on the preset dot parameters and overprint reference line, extract the detection area mask through morphological closing operation, and output the standard image after geometric alignment. [[ID=X]]

[0033] In this embodiment, the dynamic segmentation module establishes a geometric correction benchmark through preset printing process parameters. The dot parameters include screen angle, dot shape (circular / square / rhombus), and screen ruling (such as 175 LPI). Based on these parameters, a reference grid template is generated. The template is composed of periodically arranged dot units, and the size of each unit is calculated from the screen ruling, and the grid spacing The system detects the actual spatial position of the overprint reference line through the Hough transform algorithm. The overprint reference line is composed of the cross marks preset by the printing unit, and an improved Radon transform method is used to enhance the detection robustness. When the deviation between the detected actual reference line and the theoretical position is detected, calculate its angular deviation α and translation compensation amounts Δx and Δy, and construct an affine transformation matrix:

[0034] where 、 are the image pixel coordinates, is the angular deviation amount of the overprint reference line, 、 are the translation compensation amounts. This matrix performs geometric correction on the input image through the bilinear interpolation algorithm to eliminate the stretching distortion and rotational deviation caused by the mechanical transmission system. The corrected image enters the morphological processing stage, and multi-scale closing operation is used to eliminate local noise and fill microscopic pores. The mathematical expression of the morphological closing operation is:

[0035] where is the output image matrix after the morphological closing operation, is the input image matrix, is the structuring element, represents the dilation operation, represents the erosion operation. In this embodiment, a circular structuring element with a radius of 3 pixels is first selected to perform the dilation operation , filling microscopic pores smaller than 6 pixels; then performing an erosion operation with the same structural element , while restoring the main contours and eliminating isolated noise points. To further enhance the integrity of the connected domain, a secondary closing operation is performed using a 5-pixel radius structural element, and finally a detection area mask is generated. This mask is precisely aligned with the original image through bit operations, so that the pixel coordinate system of the output standard image is mapped to the encoder signal of the printing unit. The mapping accuracy is controlled by a phase locking algorithm. Phase angle range.

[0036] During the specific implementation process, the system obtains the position information of the printed material by real-time acquisition of encoder pulse signals, and dynamically calibrates the origin of the image coordinate system with the phase origin of the printing unit. In the calibrated coordinate system, each pixel corresponds to a specific mechanical phase angle. (where n is the encoder count value and N is the total number of pulses corresponding to the drum circumference), achieving sub-pixel synchronization between the image space and the mechanical space. The geometric distortion residual error of the standard image after dynamic segmentation processing is controlled within The standard image data output by this module is stored in 16-bit grayscale format and comes with metadata including dot parameters, correction matrix coefficients and timestamp, providing a complete geometric feature benchmark for subsequent processing.

[0037] The geometrically aligned standard image is input into the defect detection model to extract microscopic dot features and macroscopic texture features. By performing cross-scale matching with the template library, a probability distribution map containing the location and confidence information of the abnormal area is generated.

[0038] In this embodiment, the defect detection model adopts a dual-branch deep convolutional neural network architecture to achieve parallel extraction and cross-scale fusion of micro features and macro textures. The first branch is dedicated to capturing the micro morphological features of printed dots. Its input layer receives a standard image after geometric correction, which is normalized and input into a feature extraction network consisting of five cascaded 3×3 small-kernel convolution layers. Each convolution layer uses a ReLU activation function and superimposes a batch normalization operation. The step size of the first layer is set to 1 to retain detail information, and the subsequent layers gradually increase the receptive field. Through this branch, three key features can be extracted: edge jaggedness, dot area ratio, and ink diffusion coefficient: edge jaggedness is quantified by the peak value of the contour gradient histogram, reflecting the smoothness of the dot edge; the dot area ratio is calculated by the proportion of the effective printing area after pixel binarization; the ink diffusion coefficient is obtained based on the optical density attenuation curve fitting from the dot boundary to the center, and its mathematical expression is ,in is the radial light intensity distribution, is the diffusion attenuation factor.

[0039] The second branch is designed as an atrous convolutional network for parsing the macroscopic texture features of printed patterns. This branch constructs a multi-scale feature pyramid using three groups of atrous convolutional layers with atrous rates of 2, 4, and 6 respectively, and fuses semantic information at different levels through skip connections. The macroscopic feature extraction focuses on three dimensions: color difference distribution, moiré fringes, and overprint misregistration amount. The color difference distribution is characterized by the standard deviation and skewness values of the a and b channels in the CIE-Lab color space. Moiré fringe detection uses two-dimensional Fourier transform to extract abnormal peak values in the frequency domain, and its intensity is quantified by the energy ratio of the fundamental frequency component ; the overprint misregistration amount is calculated by the cross-correlation matching of the four-color plate registration marks, and the offset vector corresponding to the maximum cross-correlation value is the misregistration amount.

[0040] The multi-scale features extracted by the two branches are fused and enhanced through a cross-modal attention mechanism. This module first performs channel-level concatenation on the microscopic feature map and the macroscopic feature map , and after dimensionality reduction through 1×1 convolution, it is input into the multi-head self-attention layer. The attention weight calculation uses the scaled dot-product formula , where the query matrix is generated by projecting the microscopic features, and the key matrix and the value matrix are derived from the macroscopic features. After parallel processing by eight groups of attention heads, the output enhanced feature vector retains both microscopic details and macroscopic context information.

[0041] In the feature matching stage, the system measures the similarity between the fused multi-modal features and the standard samples in the template library. The template library builds a three-dimensional hierarchical index structure according to the printing color plate type, printing substrate (copperplate paper / graphic art paper / special paper), and printing process parameters, supporting fast retrieval based on hash coding. For the th microscopic feature value of the test sample, its similarity calculation with the th standard sample in the template library adopts a composite measurement strategy: The microscopic feature similarity is calculated based on cosine similarity:

[0042] where is the dimension of the microscopic features.

[0043] The macroscopic feature similarity is measured using the normalized Euclidean distance:

[0044] where is the macroscopic feature vector of the test sample, is the standard macroscopic feature vector in the template library, and is the maximum Euclidean distance of all sample macroscopic feature vectors in the template library.

[0045] The final matching probability value is determined by the weighted fusion formula where the weight coefficients 、 are determined by grid search optimization and satisfy the constraint conditions. When the fusion probability value exceeds the preset probability threshold of 0.85, it is determined that there is an abnormality in the area. The probability distribution map reconstruction process is implemented using a deconvolution network. The network consists of three levels of transposed convolutional layers, each equipped with a LeakyReLU activation function and Dropout regularization. Finally, the confidence is normalized to interval through the Softmax function to generate a probability heat map with the same spatial resolution as the input image.

[0046] The model training adopts a two-stage optimization strategy: first, pre-train on the ImageNet dataset to obtain the basic feature extraction ability; then, fine-tune on the self-built printed defect dataset, which contains 35,000 labeled images of 12 common defects. The loss function is designed as a weighted combination of Focal Loss and Dice loss:

[0047] where 、 . The Adam optimizer is used during the training process, and the initial learning rate is set to , which decays by 0.5 times every 10 epochs and converges after training for 150 epochs.

[0048] Multi-level discrimination is performed according to the probability distribution map. Among them, the edge diffusion deviation amount is quantitatively evaluated at the dot level, and the topological structure is analyzed at the pattern level. Finally, the coordinates and type labels of the printed defects with spatio-temporal annotations are output.

[0049] In this embodiment, the multi-level discrimination process realizes accurate defect classification through the collaborative analysis of spatial features and topological relationships. For the abnormal regions in the probability distribution map with confidence exceeding the threshold, the system adopts a hierarchical progressive detection strategy: focusing on the morphological distortion of a single dot at the micro scale and analyzing the topological integrity of the pattern structure at the macro scale. For the analysis at the dot level, the system first executes a contour tracking algorithm to extract the closed contour of the abnormal region, uses the Douglas-Peucker algorithm to discretize the contour into a piecewise linear approximation curve, and samples 1000 points at equal intervals along the curve arc length parameter s. The curvature value of each sampling point is calculated by the three-point method, that is, using the reciprocal of the local circular arc radius fitted by the coordinates of three adjacent points as the curvature estimation value, and the calculation formula is where 、 are the coordinate differences of adjacent points. The actual curvature distribution curve of the contour is compared point by point with the ideal curvature curve generated by the theoretical Gaussian diffusion model , where is the center position of the dot, which is determined by the printing process parameters. The edge diffusion deviation amount quantifies the morphological difference between the actual and ideal profiles through an integral form:

[0050] where is the number of divided contour segments, is the arc length of the th contour segment. The system presets a dynamic deviation threshold ( is the theoretical radius of the dot), and when , it is determined as an edge diffusion defect and an ink viscosity adjustment instruction is triggered.

[0051] In the topological structure analysis at the pattern level, the system first performs skeletonization on the abnormal area, and uses a thinning algorithm to iteratively remove boundary pixels until a topological skeleton with a single-pixel width is generated. An adjacency matrix is constructed based on the skeleton graph, where the number of nodes is determined by the total number of skeleton branch points and endpoints, and the matrix element indicates that there is a direct connection between nodes i and j, otherwise it is 0. By solving the matrix eigenvalue equation , the eigenvalue set is obtained, and this eigen-spectrum has a unique representativeness for the connectivity, number of loops, and sub-graph distribution of the topological structure. The eigenvalue set of the measured abnormal area is matched with the of the standard pattern in the template library to calculate its topological similarity:

[0052] where is the adjacency matrix eigenvalue of the th standard pattern, , is the eigenvalue of the th measured abnormal area, , is the number of eigenvalues of the adjacency matrix. When Stopo is less than the topological similarity threshold of 0.82, it is determined that there is a structural defect. The system automatically retrieves defect cases with similar topological features in the historical data and generates optimization suggestions by combining the ink key opening and cylinder pressure parameters of the current printing unit. For the detected critical path break (such as the existence of isolated nodes in the skeleton diagram) or adjacency relationship mismatch (eigenvalue deviation exceeding 3 standard deviations), the system classifies it as a register misalignment or mechanical damage defect and triggers the forced rejection instruction of the sorting mechanism.

[0053] The spatio-temporal annotation data includes the phase angle, the size of the defect envelope rectangle, and the type code, integrating the phase synchronization information of the printing unit and the geometric features of the defect, and finally outputting the coordinates and type labels of the printing defects with spatio-temporal annotation.

[0054] Map the coordinates of the printing defect to the phase coordinate system of the printing unit, trigger the sorting mechanism to perform the phase synchronization rejection operation, and generate a closed-loop feedback instruction by associating the ink pressure adjustment parameter and the temperature control parameter.

[0055] In this embodiment, by constructing an accurate spatio-temporal correlation model between the spatial position of the printing defect and the mechanical movement of the printing unit, the phase synchronization control of the defect positioning and rejection actions is realized. The system collects the angular displacement signal of the main drive shaft of the printing unit in real time through a high-speed encoder and analyzes the current cylinder phase angle . For the detected defect coordinates , the system maps them to the phase coordinate system of the printing unit according to the pre-calibrated conversion relationship between the image coordinate system and the mechanical coordinate system. Specifically, through the sub-pixel mapping model established by the geometric correction module, the mechanical phase angle corresponding to the defect coordinates is calculated by the following formula:

[0056] where is the circumference of the printing cylinder, is the effective detection range in the printing width direction, is the phase angle increment between adjacent detection rows in the cylinder axial direction. This model is established by combining offline calibration and online calibration. During the calibration process, the reference marking points on the standard calibration plate are used for two-way coordinate registration, and finally the phase angle mapping accuracy of is achieved.

[0057] In the phase synchronization rejection control, the system dynamically calculates the trigger moment t of the rejection action according to the current cylinder angular velocity and the mechanical delay characteristics of the sorting mechanism. The mathematical model of the rejection trigger moment is:

[0058] where is the current phase angle of the printing unit, is the angular velocity of the cylinder, is the delay compensation amount, , is the inherent mechanical delay of the sorting mechanism (including solenoid valve response and cylinder action time), which is measured by a step response experiment; is the system processing delay (including image processing and communication transmission time), which is obtained by real-time performance test statistics.

[0059] The generation of the closed-loop feedback command is based on the multi-dimensional analysis of the defect type and spatial distribution. By statistically analyzing the occurrence frequency and distribution pattern of various defects in the current production batch, and combining the real-time data of the ink zone pressure sensor and temperature sensor, an adaptive adjustment model of ink pressure-temperature control parameters is constructed. Specifically, for edge diffusion type defects, the system calculates the deviation between the ink diffusion coefficient D of the dot layer and the theoretical threshold, and updates the ink fountain roller pressure setting value according to the proportional-integral control algorithm :

[0060] where , are the PID control parameters, which are optimized and determined by the orthogonal experiment method. The adjustment of the temperature control parameters is based on the temperature difference between the ink layer curing temperature T collected by the thermal imager and the target value . A fuzzy control strategy is adopted to generate the heater power adjustment command, and the temperature control accuracy is .

[0061] The system sends the adjustment command to the printing press PLC control system through the OPC UA protocol, and at the same time records the historical adjustment data to optimize the control model. Every time 1000 meters of printed sheets are produced, the system automatically performs self-check and calibration of the control parameters: by injecting a preset test pattern, evaluating indicators such as ink layer uniformity and overprint accuracy, and dynamically correcting the phase compensation amount and mechanical delay in the mapping model to ensure that the closed-loop control system maintains stable detection accuracy and feedback response speed throughout its life cycle.

[0062] Embodiment 2: As shown in Figure 2 , the present invention provides a printing defect detection system based on computer vision, including: A multi-source acquisition module 1, which acquires multi-channel fusion images of the printing surface containing ink physical characteristic parameters through a multi-modal imaging device. The multi-modal imaging device uses a multi-spectral composite light source for sequential excitation; The geometric correction module 2 is used to input the multi-channel fusion image of the printing surface into the dynamic segmentation module, correct the mechanical transmission distortion based on the preset dot parameters and overprint reference lines, extract the detection area mask through morphological closing operation, and output the standard image after geometric alignment; The feature matching module 3 is used to input the geometrically aligned standard image into the defect detection model, extract the microscopic dot features and macroscopic texture features, and generate a probability distribution map containing the position and confidence information of the abnormal area through cross-scale matching with the template library; The hierarchical decision-making module 4 performs multi-level discrimination according to the probability distribution map. Among them, the edge diffusion deviation amount is quantitatively evaluated at the dot level, and the topological structure is analyzed at the pattern level. Finally, the printing defect coordinates and type labels with spatio-temporal annotations are output; The closed-loop control module 5 is used to map the printing defect coordinates to the phase coordinate system of the printing unit, trigger the sorting mechanism to perform phase synchronization rejection operation, and generate a closed-loop feedback instruction by associating the ink pressure adjustment parameter and the temperature control parameter.

[0063] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A printing defect detection method based on computer vision, characterized in that, The method includes: Collecting a multi-channel fusion image of the printed surface containing ink physical property parameters through a multi-modal imaging device, and the multi-modal imaging device uses a multi-spectral composite light source for sequential excitation; Inputting the multi-channel fusion image of the printed surface into a dynamic segmentation module, correcting mechanical transmission distortion based on preset dot parameters and overprint reference lines, extracting a detection area mask through morphological closing operation, and outputting a standard image after geometric alignment; Inputting the geometrically aligned standard image into a defect detection model, extracting microscopic dot features and macroscopic texture features, and generating a probability distribution map containing the position and confidence information of abnormal areas through cross-scale matching with a template library; Performing multi-level discrimination according to the probability distribution map, where the edge diffusion deviation amount is quantitatively evaluated at the dot level, and the topological structure is analyzed at the pattern level, and finally outputting the coordinates and type labels of the printed defects with spatio-temporal annotations; Mapping the printed defect coordinates to the phase coordinate system of the printing unit, triggering the sorting mechanism to perform phase-synchronized rejection operations, and associating ink pressure adjustment parameters with temperature control parameters to generate a closed-loop feedback command.

2. The method for detecting printing defects based on computer vision according to claim 1, wherein The multi-modal imaging device uses an LED light source group with a circular array distribution, including a multi-spectral composite light source in the visible light band, near-infrared band, and short-wave infrared band; The imaging unit consists of a high-resolution area array CCD camera and a multi-channel filter wheel, and constructs a multi-channel fusion image containing ink thickness, curing degree, and adhesion physical property parameters by synchronously collecting reflectivity, fluorescence intensity, and thermal radiation response data.

3. The method for detecting printing defects based on computer vision according to claim 1, characterized in that, The dynamic segmentation module specifically includes: Generating a reference grid template based on preset dot parameters, and the dot parameters include screen angle, dot shape, and screen ruling; Detecting the actual position of the overprint reference line through the Hough transform, calculating the offset vector between it and the theoretical position, and using an affine transformation matrix to correct the stretching and rotation distortions caused by mechanical transmission; Performing multi-scale morphological closing operations on the corrected image, selecting a circular structuring element to perform dilation and erosion operations with radii of 3 pixels and 5 pixels in sequence, eliminating local noise and filling microscopic pores, and generating a detection area mask with complete connected regions; Performing bitwise operation alignment on the mask and the original image, and outputting a standard image with eliminated position deviation, and establishing a sub-pixel level mapping relationship between its pixel coordinate system and the encoder signal of the printing unit.

4. The method for detecting printing defects based on computer vision according to claim 3, characterized in that, The construction of the affine transformation matrix satisfies the following relationship: Among them , are the image pixel coordinates, is the angle deviation amount of the overprint reference line, , are the translation compensation amounts; The mathematical expression of the morphological closing operation is: Among them is the output image matrix after morphological closing operation, is the input image matrix, is the structuring element, represents the dilation operation, represents the erosion operation.

5. The method for detecting printing defects based on computer vision according to claim 1, wherein The working process of the defect detection model includes: Adopting a dual-branch deep convolutional neural network, the first branch extracts microscopic dot morphological features through a 3×3 small kernel convolutional layer, capturing edge serration, dot area ratio, and ink diffusion coefficient; The second branch extracts macroscopic texture features through a dilated convolutional layer, analyzing color difference distribution, moiré fringes, and overprint misregistration amount; Inputting the multi-scale features into a cross-modal attention mechanism module, and performing similarity matching with standard samples in the template library, and the template library establishes a hierarchical index according to printing color plates and substrate types; Reconstruct the probability distribution map of the abnormal area through the deconvolution layer, and its confidence value is normalized to the interval.

6. The method for detecting printing defects based on computer vision according to claim 5, wherein, The similarity matching adopts the following measurement method: Microscopic feature similarity: Macroscopic feature similarity: Among them is the th microscopic eigenvalue of the test sample, is the th standard microscopic eigenvalue in the template library, is the microscopic feature dimension, is the macroscopic feature vector of the test sample, is the standard macroscopic feature vector in the template library, is the maximum Euclidean distance of all sample macroscopic feature vectors in the template library; The final probability value is determined by the weighted fusion formula and and are weight coefficients, . When is greater than the preset probability threshold, it is determined as an abnormal area.

7. The method for detecting printing defects based on computer vision according to claim 1, wherein The specific content of the multi-level discrimination includes: At the dot level, perform a contour tracking algorithm on the abnormal area in the probability distribution map, extract the curvature extreme points along the dot edge, calculate the deviation between the actual contour and the theoretical Gaussian diffusion curve, and mark it as an edge diffusion defect when the deviation exceeds the deviation threshold; At the pattern level, construct a topological skeleton map of the abnormal area, analyze the node connectivity, the number of loops, and the subgraph isomorphism through graph theory algorithms, and mark it as a structural defect when a critical path break or adjacency relationship mismatch is detected; Generate spatio-temporal annotation data including the phase angle of the printing unit, the defect size, and the type code by integrating spatio-temporal information.

8. The method for detecting printing defects based on computer vision according to claim 7, wherein The quantization formula for the edge diffusion deviation is: wherein is the curvature function of the actual contour, is the curvature function of the ideal Gaussian diffusion curve, is the arc length of the th contour segment, and is the total number of contour segments; The topological structure analysis adopts the adjacency matrix eigenvalue comparison method: The eigenvalues of the adjacency matrix of the standard pattern are , and the eigenvalues of the measured abnormal area are . Then the topological similarity is calculated by the following formula: Among them is the eigenvalue of the adjacency matrix of the th standard pattern, , is the eigenvalue of the feature of the th measured abnormal area, , is the number of eigenvalues of the adjacency matrix. When is less than the topological similarity threshold, it is determined as a structural defect.

9. The method for detecting printing defects based on computer vision according to claim 1, characterized in that, Mapping the printing defect coordinates to the phase coordinate system of the printing unit includes: parsing the current phase angle of the printing unit through the encoder signal , converting the defect coordinates into the rejection trigger moment in the phase coordinate system , where is the current phase angle of the printing unit, is the angular velocity of the cylinder, is the delay compensation amount.

10. A printing defect detection system based on computer vision, characterized in that, The system includes: A multi-source acquisition module that acquires multi-channel fusion images of the printing surface containing ink physical property parameters through a multi-modal imaging device. The multi-modal imaging device uses a multi-spectral composite light source for sequential excitation; A geometric correction module that inputs the multi-channel fusion images of the printing surface into the dynamic segmentation module, corrects the mechanical transmission distortion based on preset dot parameters and the overprint reference line, extracts the mask of the detection area through morphological closing operation, and outputs the standard image after geometric alignment; A feature matching module that inputs the geometrically aligned standard image into the defect detection model, extracts microscopic dot features and macroscopic texture features, and generates a probability distribution map containing the position and confidence information of the abnormal area through cross-scale matching with the template library; A hierarchical decision-making module that performs multi-level discrimination based on the probability distribution map. Among them, the edge diffusion deviation is quantitatively evaluated at the dot level, and the topological structure is analyzed at the pattern level. Finally, it outputs the printing defect coordinates and type labels with spatio-temporal annotation; A closed-loop control module that maps the printing defect coordinates to the phase coordinate system of the printing unit, triggers the sorting mechanism to perform phase synchronization rejection operations, and generates a closed-loop feedback instruction by associating the ink pressure adjustment parameter and the temperature control parameter.

Images