Training method and system applied to a printed defect detection model

Abstract

The present application relates to the technical field of printing defect detection, and particularly relates to a training method and system applied to a printing defect detection model. The method comprises the following steps: collecting appearance data of a printing product and performing image optimization processing to construct a printing defect detection model; analyzing distortion abnormal states and detail gradient fuzzy conditions of a printing pattern through the optimized image data, evaluating a visual performance decreasing trend; determining an abnormal running condition of a printing plate, locating a printing plate defect, and evaluating a structure aging degree of the printing plate; predicting an error growth trend of the printing plate according to the aging degree, and combining the pattern visual performance decreasing to identify a printing abnormal defect; and optimizing and training the defect detection model by using the defect conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of printing defect detection, and in particular to a training method and system applied to a printing defect detection model. BACKGROUND

[0002] The printing quality detection method usually relies on manual visual inspection, which is difficult to comprehensively identify complex and diversified printing defects, especially in continuous detection of high-precision and complex patterns, there are obvious limitations. The defect types of printed matter include pattern blur, proportion distortion, detail loss, structure distortion, text and image stretching and compression, color block misplacement, etc. These defects affect the aesthetics and recognizability of the finished product, leading to subsequent problems such as loss of traceability information and failure of automatic system recognition. Computer vision and deep learning algorithms are gradually introduced to improve the detection efficiency and accuracy of printing defects, and a defect detection model suitable for different scenarios is constructed. The existing detection system lacks dynamic updating and self-optimization mechanism, which leads to the model training staying in the static stage and cannot adapt to the gradual evolution of factors such as surface wear and stress fatigue caused by long-term changes in printing conditions. However, the traditional printing defect detection has the problems of inaccurate evaluation of the decreasing trend of the visual performance of the printing pattern and inaccurate detection of the abnormal defect situation of the printing. SUMMARY

[0003] Therefore, it is necessary to provide a training method and system applied to a printing defect detection model to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a training method applied to a printing defect detection model comprises the following steps:

[0005] Step S1: obtaining printed product appearance data; collecting printed product images from the printed product appearance data to obtain printed product image data; performing image optimization processing on the printed product image data to obtain printed product image optimization data; constructing a printing defect detection model according to the printed product image optimization data to obtain the printing defect detection model;

[0006] Step S2: determining the abnormal state of the printing pattern distortion according to the printed product image optimization data; detecting the gradient blur condition of the printing pattern details according to the printed product image optimization data; evaluating the decreasing trend of the visual performance of the printing pattern based on the gradient blur condition of the printing pattern details and the abnormal state of the printing pattern distortion;

[0007] Step S3: determining the abnormal state of the printing plate operation based on the decreasing trend of the visual performance of the printing pattern; performing printing plate defect positioning processing using the abnormal state of the printing plate operation to obtain printing plate defect positioning data; evaluating the aging degree of the printing plate structure according to the printing plate defect positioning data;

[0008] Step S4: predicting a progressive error growth trend of the printing plate according to the aging degree of the printing plate structure; determining a printing abnormal defect condition according to the progressive error growth trend of the printing plate and the visual performance degradation trend of the printing pattern; and performing model training optimization on the printing defect detection model by using the printing abnormal defect condition, to obtain an optimized training model of the printing defect detection.

[0009] The present application can effectively improve the detection accuracy of printing defects by accurately collecting and optimizing the appearance data of the printed products. Through image optimization, image noise caused by external environment or equipment factors is reduced, ensuring the accuracy of the defect detection model. On this basis, the detection of printing pattern distortion abnormal state and detail gradient fuzzy condition provides a basis for evaluating the visual performance degradation of the printing pattern, and further reveals the potential problems and development trend of the printing quality. Through the detection of the printing plate running abnormal condition and the evaluation of the structure aging degree, potential faults of the printing plate can be found in time and defects can be accurately located to prevent the outflow of defective products. The analysis of the structure aging provides a prediction of the performance degradation of the printing plate during long-term use, which helps to develop maintenance and replacement plans. After further combining the prediction of the progressive error growth trend, accurate monitoring and early warning of the printing plate are realized, the defect detection model is accurately identified and optimized, and the stability and precision of the printing production line are improved. This system not only optimizes the existing detection process, but also greatly improves the detection efficiency and quality of printing defects, ensuring the consistency of product quality and maximizing production efficiency. Therefore, the present application optimizes the training of the traditional printing defect detection model, solves the problems of inaccurate evaluation of the visual performance degradation trend of the printing pattern and inaccurate detection of the printing abnormal defect condition in the traditional training of the printing defect detection model, and improves the accuracy of the evaluation of the visual performance degradation trend of the printing pattern and the accuracy of the detection of the printing abnormal defect condition.

[0010] The present application also provides a training system for a printing defect detection model, which is used to execute the training method for the printing defect detection model as described above, and the training system for the printing defect detection model comprises:

[0011] A model construction module is configured to acquire appearance data of printed products, perform image collection on the appearance data of the printed products to obtain image data of the printed products, perform image optimization processing on the image data of the printed products to obtain optimized image data of the printed products, and construct a printing defect detection model according to the optimized image data of the printed products, to obtain the printing defect detection model.

[0012] a visual performance degradation trend evaluation module configured to determine a printing pattern distortion abnormality state according to the printing finished product image optimization data, detect a printing pattern detail gradient blur condition according to the printing finished product image optimization data, and evaluate a printing pattern visual performance degradation trend based on the printing pattern detail gradient blur condition and the printing pattern distortion abnormality state;

[0013] a printing plate structure aging degree evaluation module configured to determine a printing plate running abnormality condition based on the printing pattern visual performance degradation trend, perform printing plate defect positioning processing using the printing plate running abnormality condition to obtain printing plate defect positioning data, and evaluate a printing plate structure aging degree according to the printing plate defect positioning data;

[0014] a model training optimization module configured to predict a printing plate progressive error growth trend according to the printing plate structure aging degree, determine a printing abnormality defect condition according to the printing plate progressive error growth trend and the printing pattern visual performance degradation trend, and perform model training optimization on a printing defect detection model using the printing abnormality defect condition to obtain a printing defect detection optimization training model.

[0015] The present application is applied to a training system of a printing defect detection model, which can realize any application of the training method of the printing defect detection model of the present application, is used as a medium for joint operation and signal transmission between various modules, and is used to complete the training method applied to the printing defect detection model. The modules in the system cooperate with each other, accurately collect and optimize printing finished product image data, combine visual performance degradation, printing plate aging evaluation and progressive error prediction, effectively improve the precision and efficiency of printing defect detection, optimize the defect detection model, and improve the printing quality control capability. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 It is a step flowchart of a training method applied to a printing defect detection model;

[0017] Figure 2 It is a step flowchart of a training method applied to a printing defect detection model; Figure 1 It is a detailed implementation step flowchart of step S3 in the method;

[0018] Figure 3 It is a step flowchart of a training method applied to a printing defect detection model; Figure 1 It is a detailed implementation step flowchart of step S4 in the method;

[0019] The implementation of the present application, functional features and advantages will be further described with reference to the accompanying drawings. DETAILED DESCRIPTION

[0020] The technical method of the present application will be described clearly and completely in combination with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0021] In addition, the drawings are only schematic illustrations of the present application, and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated description thereof will be omitted. Some block diagrams shown in the drawings are functional entities, which do not necessarily have to correspond to physically or logically independent entities. The functional entities can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0022] It should be understood that although the terms "first", "second" and the like can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the exemplary embodiments, a first element can be called a second element, and similarly a second element can be called a first element. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0023] To achieve the above-mentioned purpose, please refer to Figures 1 to 3 A training method applied to a printing defect detection model, comprising the following steps:

[0024] Step S1: obtaining printing product appearance data; collecting printing product images from the printing product appearance data to obtain printing product image data; performing image optimization processing on the printing product image data to obtain printing product image optimization data; constructing a printing defect detection model according to the printing product image optimization data to obtain the printing defect detection model;

[0025] In the embodiment of the present application, an industrial-grade linear array CCD high-speed camera and a backlight module are deployed to perform synchronous image acquisition at a fixed detection section of the printing product conveyor belt path. A PLC control module is used to synchronously control the image acquisition beat and the conveyor belt speed to ensure that each printing product is in a non-displacement and non-shaking state during image acquisition. The image acquisition format is set to 16-bit grayscale image and three-channel RGB image parallel acquisition, and the image resolution is set to 0.01 mm / pixel to ensure that the subsequent image features can be accurately separated. The acquired image data is transmitted to an edge computing node through a gigabit Ethernet, and the image optimization process relies on OpenCV and Halcon image processing libraries. A high-pass filter is used in the edge node to perform printing image sharpening processing, and a multi-scale Retinex algorithm is used to dynamically adjust the image brightness unevenness and color saturation imbalance. After image enhancement, Canny edge extraction and morphological opening operation algorithm are used to remove image noise, generate clear edge and color balanced "printing product image optimization data", and store it in.tiff lossless format. The printing defect detection model is constructed based on the geometric distribution characteristics of the alignment mark area, the printing boundary area and the pattern entity area contained in the image optimization data. An area positioning module for defect detection is constructed based on image geometry matching and contour recognition method. The module realizes the pre-screening logic of abnormal areas such as uneven color block area, pattern boundary shedding and local disconnection by setting specific threshold rules. The area features are used as input sources to construct a defect type judgment process and generate a "printing defect detection model". The model structure is established based on rule logic tree and does not involve any AI learning mechanism. The model output includes: abnormal area coordinates, abnormal type classification labels and boundary offset data, which are saved as structured data files for subsequent processing.

[0026] Step S2: determining the printing pattern distortion abnormality state according to the printing product image optimization data; detecting the printing pattern detail gradient blur condition according to the printing product image optimization data; and evaluating the printing pattern visual performance decreasing trend based on the printing pattern detail gradient blur condition and the printing pattern distortion abnormality state;

[0027] In the embodiment of the present application, the pixel arrangement difference between the main visual area and the edge reference area of the printed pattern is extracted from the "printed product image optimization data", and the spatial offset between the actual printed pattern and the design reference pattern is analyzed by the pattern template matching and image coordinate alignment method. In the specific implementation process, the template transformation function in the Halcon image matching tool is used to perform affine matching processing on the original template, and then combined with corner point recognition and translation vector calculation, the "printed pattern distortion abnormal state" data is generated. This data quantifies the area of the current pattern that has undergone geometric deformation, the offset direction, the offset amount (in mm), and the offset slope gradient (in degrees). Then, the detail gradient blur condition analysis is performed, and the image high-frequency gradient energy detection method based on Laplace transform is used to extract the texture clarity parameters of the key details of the pattern (such as frame curves, text strokes, and color superimposed areas). If the gradient energy value is lower than the set threshold (the threshold is determined to be below 14.2 units of intensity according to experimental statistics), it is determined that the "printed pattern detail gradient blur" occurs. The gradient intensity analysis of multiple detail areas in the image is performed respectively, and a blur condition identification map is generated, recording the coordinates, area, gradient distribution difference, and relative standard deviation of each blurred area. By combining the above two data, i.e., the printed pattern distortion abnormal state and the detail blur condition, the visual performance degradation trend of the printed pattern is evaluated according to the level division standard of visual performance degradation. The specific method includes constructing the abnormal area ratio index (the ratio of the defect area in the unit area), the detail loss rate index (the gradient attenuation ratio of the detail area), and the color gamut deviation index (the mean deviation of the color block area), to form a complete "printed pattern visual performance degradation trend" evaluation data set, which provides a basis for the next step of printed plate abnormal state judgment.

[0028] Step S3: determining the printed plate running abnormal condition based on the visual performance degradation trend of the printed pattern; performing printed plate defect positioning processing using the printed plate running abnormal condition to obtain printed plate defect positioning data; and evaluating the printed plate structure aging degree according to the printed plate defect positioning data;

[0029] In the embodiment of the present application, according to the "print pattern visual performance degradation trend" evaluation data generated in the previous step, the inference determination operation of the printing plate running abnormal state is performed. This operation is based on the comparison table of pattern attenuation characteristic parameters corresponding to different plate wear degrees in historical printing batches, and a parameter lookup table is deployed in the edge node for dynamic matching. By setting a threshold interval, the current pattern visual attenuation trend data is mapped to the plate running state level, such as "mild abnormality", "moderate wear", "serious deviation", etc., to generate "printing plate running abnormal condition" label data. According to the running abnormal state information, combined with the printing pattern area coordinates and pattern projection error recorded in the printing finished product image optimization data, the printing plate defect positioning processing is performed. Using the mapping algorithm based on geometric backstepping, the pattern deformation area is projected into the printing cylinder plate coordinate system in reverse, and the potential damage source position is located through contour mapping recognition. The defect area center coordinates, coverage range, error gradient direction, etc. form "printing plate defect positioning data" and are marked on the plate development diagram to assist subsequent maintenance. Based on the distribution density of each defect position in the defect positioning data and the historical cumulative area ratio, the material micro-loss trend, the pressure-resistant coating peeling trend and the plate deformation direction on the plate surface are analyzed, and the "printing plate structure aging degree" is calculated. The degree is expressed by an aging index, which is calculated by weighting the cumulative damage area, the number of damage times and the damage direction consistency coefficient, and the value range is 0-1. The higher the value, the more serious the aging. The result will be combined with the subsequent error trend evaluation for processing.

[0030] Step S4: predicting the progressive error growth trend of the printing plate according to the printing plate structure aging degree; determining the printing abnormal defect condition according to the printing plate progressive error growth trend and the print pattern visual performance degradation trend; training and optimizing the printing defect detection model using the printing abnormal defect condition to obtain an optimized printing defect detection model.

[0031] In the embodiment of the present application, the "printing plate structure aging degree" parameter is read, and according to the value range, it is determined which stage the current printing plate is in in the gradual error growth. The front feed differential incremental analysis method is used to calculate the incremental change trend of the pattern position error, pattern repeatability error and pattern stretching deformation error in the continuous printing process. The error growth trend curve is drawn with the time axis as the horizontal axis and the error amount as the vertical axis, the trend slope and acceleration are extracted, and the "printing plate gradual error growth trend" index data is obtained. The trend data is cross-compared with the previously obtained "decreasing trend of printing pattern visual performance", and an abnormal defect comprehensive mapping table containing time series and spatial superposition is constructed. Combined with the error source position and the visual abnormality degree, the "printing abnormal defect condition" is accurately determined. Such defect conditions include: periodic pattern stretching, intensified transverse boundary blur, double pattern ghosting, and continuously increasing boundary missing lines. The above printing abnormal defect condition data is fed back to the initially constructed "printing defect detection model" to realize the training and optimization processing of the model. On the basis of not relying on deep learning, three mechanisms of defect frequency weight adjustment, threshold dynamic self-adaptive adjustment and region discrimination logic enhancement are used to reconstruct the region recognition module and defect classification module of the original model. The newly added recognition features, abnormal range, structure aging law and error trend data are written into the judgment rule table as conditional logic to realize the generation of the "printing defect detection optimization training model". The optimization model is output in JSON format and loaded into the system running environment as the update configuration of the printing defect real-time monitoring system.

[0032] Preferably, step S1 comprises the following steps:

[0033] Step S11: set the resolution of the visible light camera to 1920x1080 pixels, the minimum light sensitivity to 0.01 lux, and the frame rate to 30 FPS;

[0034] In the embodiment of the present application, in the printing finished product image acquisition link, in order to ensure the complete capture of image detail information, an industrial visible light camera with a model of SONY IMX327 sensor is selected, the image acquisition resolution thereof is set to 1920x1080 pixels, so as to ensure that the image details have a definition of at least 300 pixels per square millimeter. In order to improve the imaging capability in low light environment, the minimum light sensitivity of the camera is set to 0.01 lux, so as to ensure that high-quality image acquisition can still be completed under the condition of insufficient workshop illumination (such as less than 30 lux). At the same time, the frame rate is set to 30 frames / second (30FPS), so as to avoid image tearing or acquisition delay phenomenon while maintaining the continuity and dynamic response performance of the image. The camera is connected with the main control processing system through a USB 3.0 industrial camera interface, data stream acquisition is carried out by using DirectShow protocol, and the acquisition stability is ensured. This configuration aims to provide hardware support for subsequent image data acquisition and image processing link, and ensure that the data quality meets the standardization processing requirements.

[0035] Step S12: acquiring printing finished product appearance data;

[0036] In the embodiment of the present application, the acquisition of printing finished product appearance data is carried out by the visible light camera acquisition system configured above. The acquisition window is set in the production line, and the image acquisition station is set at the first inspection position after printing. Each time the finished product passes through this position, the system triggers the shooting instruction, and records the front image. The acquisition area is set to A4 paper (210mmx297mm), and image calibration is carried out through an optical calibration device to eliminate the edge deformation error caused by lens distortion. In the acquisition process, an angle adjusting platform is additionally set, so that the sample image acquisition angle is controlled within vertical ±2°, so as to avoid the influence of angle deviation on the subsequent image processing accuracy. The acquired printing finished product appearance data is stored in the original RGB format, the pixel array size corresponding to each image is 1920x1080, the color bit depth is 8bit, and the frame buffer mechanism is used to upload the image to the image processing server.

[0037] Step S13: acquiring printing finished product image data by the visible light camera according to the printing finished product appearance data;

[0038] In the embodiment of the present application, after obtaining the appearance image basis data, the image acquisition processing flow is entered. During the execution process of the acquisition link, the image acquisition control unit is used to schedule signals for the camera, and the PLC (Programmable Logic Controller) is used to synchronously control the production rhythm, so as to ensure that the image acquisition is accurately matched with the time of passing through the detection point of the finished product. The acquired image is subjected to preliminary mean filtering processing to eliminate pulse noise, and then the ROI (Region of Interest) region extraction operation is performed on the image, according to the set image coordinate parameters, the non-printing area such as the frame, the binding hole and other interference information is stripped from the image, and only the core printing pattern area is reserved. The extracted image is stored in the lossless PNG format through the image compression function of the OpenCV image processing library, and is attached with metadata such as a timestamp, a product batch number and a printing task number, and is constructed as a printing finished product image data set, which is used as an input source for subsequent image optimization processing.

[0039] Step S14: performing image optimization processing on the printing finished product image data to obtain printing finished product image optimization data;

[0040] In the embodiment of the present application, during the image optimization processing, the brightness normalization processing is performed on the acquired printing finished product image data, so as to eliminate the pixel gray offset caused by the change of illumination. The histogram equalization algorithm (Histogram Equalization) is used to adjust the overall brightness of the image, so as to improve the contrast between the pattern boundary and the layers. Then, the Sobel edge enhancement processing method is used to sharpen the image contour, so as to retain the key printing pattern feature edge and remove the background noise interference. For the artifact noise and the gray fuzzy area existing in the image, the bilateral filter algorithm (Bilateral Filter) is used for edge-preserving filter processing, so as to avoid that the detail structure is blurred and removed. After the above operations are completed, the standard image template is compared through the structural similarity detection algorithm (SSIM), so as to check whether the clarity of the optimized image reaches the set threshold (for example, SSIM>0.85), the printing finished product image optimization data is output, and the unified image size and image bit depth format are standardized for storage, so as to be used for defect detection tasks.

[0041] Step S15: constructing a printing defect detection model according to the printing finished product image optimization data to obtain the printing defect detection model.

[0042] In the embodiment of the present application, based on the printing finished product image optimization data generated in step S14, the image feature set required for constructing the printing defect detection model is extracted by the gray level gradient distribution analysis method, the edge distribution graph in each image is extracted, and the image texture feature matrix is constructed by combining the gray level distribution mean, standard deviation and other statistical features. Subsequently, the image block sliding window analysis technology is used to scan the entire image area in a 10x10 pixel sliding window manner, the color offset degree in the local block, the boundary fracture feature and the local texture discontinuity are counted, and the corresponding defect label coordinate set is formed. The above feature data and image coordinates are labeled one by one, and are organized into a structured training sample set in the form of a matrix. A feature rule set-based discrimination method is used for model construction, wherein the standard defect types defined by the printing industry (such as “broken line”, “raw edge”, “overprint offset”, “ghosting”, “color spot” and the like) are used as output labels, and the feature abnormality degree of the corresponding image area is used as an input judgment factor to construct a defect detection model in the form of a decision tree. After the model is constructed, the model structure and parameter rule set are saved in JSON format, which serves as the basis for subsequent training optimization and abnormality detection processes.

[0043] Preferably, step S14 comprises the following steps:

[0044] Step S141: performing printing finished product image gradient amplitude distribution analysis on the printing finished product image data to obtain the printing finished product image gradient amplitude distribution;

[0045] In the embodiment of the present application, the gradient amplitude analysis is performed on the printing finished product image data, and the Sobel operator in the image processing software package OpenCV is used for calculation. The Sobel operator calculates the gradient amplitude of each pixel point by filtering the image in the horizontal and vertical directions. The gradient amplitude reflects the intensity of the edges and textures in the image, and can effectively identify the image details and contours. In this process, the original image is subjected to grayscale processing to remove color information and retain only brightness information, so as to reduce the calculation complexity. Then, the Sobel operator is used to perform convolution operation on the image to calculate the gradient amplitude of each pixel point in the image, and the gradient information of each pixel is obtained. The calculation result is stored in the form of a two-dimensional matrix, wherein the value of the matrix represents the gradient amplitude of each pixel in the image. The gradient amplitude data obtained in this analysis will serve as the basis for subsequent image sharpness evaluation.

[0046] Step S142: evaluating the printing finished product image sharpness based on the printing finished product image gradient amplitude distribution being greater than 60;

[0047] In the embodiment of the present application, according to the gradient amplitude distribution of the printed product image obtained in step S141, further image sharpness evaluation is performed, and the average value and standard deviation of the gradient amplitude of all image pixels are calculated. For the gradient amplitude of all pixels, the distribution is counted to obtain the overall "sharpness" or "clarity" of the image. When the average gradient amplitude of the image is greater than 60, it is considered that the clarity of the image is at a higher level. In the specific implementation process, Python script is used for automatic calculation, and the gradient amplitude of all pixels is summed and the standard deviation is calculated through the Numpy library. If the result meets the condition (i.e. the gradient amplitude is greater than 60), the image is evaluated as clear; if not, it is evaluated as not clear. The clarity evaluation value is one of the indicators for subsequent image quality detection.

[0048] Step S143: Local texture intensity identification of the printed product image is performed according to the printed product image data to obtain image local texture intensity data;

[0049] In the embodiment of the present application, the identification of the image local texture intensity is performed by using image blocking and local feature extraction technology, and the image is divided into multiple small blocks, and local analysis is usually performed by using a block size of 8x8 pixels. For each image block, a local binary pattern (LBP) algorithm is used for feature extraction, and the LBP algorithm generates a binary pattern to describe the texture features of the region by calculating the relationship between the pixel values of the surrounding neighborhood of each pixel point. The same operation is performed on all image blocks to obtain the texture intensity data of each image block. Then, through statistical analysis (such as mean, standard deviation) of these local texture intensity data, the local texture intensity features of the image are further obtained. These data will help to determine whether there is texture damage or loss in the local area of the image, and further provide a basis for image quality evaluation.

[0050] Step S144: The initial quality of the printed product image is evaluated based on the image local texture intensity data and the clarity of the printed product image.

[0051] In the embodiment of the present application, based on the results of steps S142 and S143, the initial quality of the printed product image is further evaluated. In the specific implementation process, by combining the analysis of the image sharpness (the result of step S142) and the local texture intensity (the result of step S143), the comprehensive quality score of the image is calculated. The calculation of the quality score adopts the weighted average method, in which the sharpness accounts for 70%, and the texture intensity accounts for 30%. For example, for an image with a sharpness score greater than 60, if its local texture intensity is within the specified range, the image quality is considered good; if the local texture intensity is too low, there is a printing defect or quality problem, and the image needs to be further optimized. The specific scoring process is realized by writing a Python script, which synthesizes the quality score according to the set weight of the sharpness score and the texture intensity score, and outputs a comprehensive quality score value, which is used to judge whether the printed product image meets the preset quality standard.

[0052] Step S145: The initial quality of the printed product image is divided into 2200 regions by grid division, and printed product image quality grid region data is obtained;

[0053] In the embodiment of the present application, in this step, the initial quality of the printed product image is divided into grid regions. The image is divided into 2200 uniform regions, each with a size of about 4.8x4.8mm, suitable for standard images of A4 paper. The division process is performed by setting a fixed size grid template, and the entire image is scanned and cut in sequence to ensure that each region is the same size and covers every part of the image. Through this operation, the quality information of the image is refined to each region. After the division is completed, for each grid region, according to the previous evaluation results of sharpness and texture intensity, a quality score is assigned to each region. In this way, the obtained grid region data provides detailed local quality information for subsequent high-frequency component loss area detection.

[0054] Step S146: Detect the high-frequency component loss area of the printed product image according to the printed product image quality grid region data;

[0055] In the embodiment of the present application, the printed product image quality grid region data obtained through step S145 is used to detect the regions where high-frequency components are lost in the image. Fourier transform is used to perform frequency domain analysis on each grid region, and the image is converted into a frequency domain image. Fourier transform converts the spatial information of the image into frequency information, which facilitates the analysis of the high-frequency components (such as details, textures, etc.) of the image. For each region, the energy distribution of the high-frequency part of its frequency spectrum is calculated. If the high-frequency components of some regions are obviously lost, it indicates that there is blur, noise or printing defects in the region. In this process, Fast Fourier Transform (FFT) algorithm is used to quickly complete the frequency domain conversion, and Python is used to realize automatic frequency analysis. The determination standard of the high-frequency component loss region is that the frequency energy of the region is less than a set threshold.

[0056] Step S147: performing image optimization processing on the high-frequency component loss region of the printed product image to obtain printed product image optimization data.

[0057] In the embodiment of the present application, the high-frequency component loss region is subjected to image optimization processing, and a high-pass filtering method is used to restore the lost image details. Specifically, for each high-frequency component loss region, local image enhancement processing is performed, and a Laplacian filter is applied to enhance the detail part of the image. The Laplacian filter is a second derivative operator that can highlight the edges and high-frequency information of the image. By adjusting the parameters of the filter, precise enhancement of the high-frequency part is realized. The enhanced image region is merged with part of the original image to generate an optimized image. This process is realized through the filtering function in OpenCV, and ensures that the optimized image can restore the high-frequency details without generating artifacts. The optimized image is used as the printed product image optimization data, which is ready for the subsequent defect detection and model training process.

[0058] Preferably, the determination of the distortion abnormality state of the printed pattern in step S2 comprises:

[0059] According to the printed product image optimization data, the external contour condition of the printed product pattern is collected;

[0060] In the embodiment of the present application, the image is collected by a high-resolution camera or a scanner to ensure that the image has sufficient details and resolution. After the image is collected, an edge detection algorithm (such as Canny edge detection) is used to process the image. The edge detection algorithm can highlight the contours of objects in the image by analyzing the gray level changes in the image to identify the edges in the image. The Canny function in the OpenCV library is used to perform gray scale conversion on the image, and then the edges of the image are extracted through Canny edge detection. The outer contour in the image is the edge information obtained by the algorithm. The key operation of this step is to efficiently identify the outer contour data of the printed product pattern and prepare the data for the next proportion analysis.

[0061] According to the printed product image optimization data, the overall proportion change of the printed product pattern is measured;

[0062] In the embodiment of the present application, the overall proportion change of the printed product pattern is evaluated using a geometric measurement method, and the obtained contour data is converted into geometric feature representation. Through the contour data, the total area of the pattern and its aspect ratio (the ratio between length and width) are calculated. According to the geometric size information of the image, the overall shape change of the printed pattern is measured. The aspect ratio of the pattern is calculated using the bounding box in the image, and the aspect ratio formula is the ratio of the length of the long side to the length of the short side. By comparing the proportion of the original design drawing, the proportion change data of the pattern is obtained by calculating the difference between the aspect ratio of the current pattern and the design drawing. This data will be used as the basis for subsequent distortion detection, reflecting whether the pattern has stretched or compressed deformation.

[0063] The printed product pattern contour condition and the overall proportion change of the printed product pattern are integrated to obtain printed product pattern feature data;

[0064] In the embodiment of the present application, the obtained contour data and the obtained proportion change data are integrated, and the contour data and the proportion change data are combined into a group of feature vectors. In the merging process, the contour condition includes the shape features of the contour (such as area, perimeter, edge smoothness, etc.), and the proportion change reflects the shift of the aspect ratio. By summarizing these data into feature data in a unified format, it is more convenient for the next step of analysis and processing. These integrated data are used for further identification of pattern features and subsequent image registration process.

[0065] Obtain printed image structure design data, and perform image registration processing on the printed product pattern feature data according to the printed image structure design data to obtain printed product image registration data;

[0066] In the embodiments of the present application, the printing image structure data at the design stage is obtained. These data are usually from the design drawings or CAD files of the printing machine, recording the ideal size, position and scale of the pattern and other information. After obtaining these design data, the integrated printing finished pattern feature data is processed based on image registration technology. Image registration uses an algorithm based on corner detection, such as SIFT (Scale-Invariant Feature Transform) or ORB (Oriented FAST and Rotated BRIEF) algorithm, to extract the feature points in the pattern that match the design data, and then match these feature points with the corresponding points in the design image through the registration algorithm. In the image registration process, the printing image is adjusted by using affine transformation or perspective transformation to ensure its alignment with the design image, and the obtained printing finished image registration data eliminates the deviation of the pattern caused by the shooting angle or deformation, making it more consistent with the design image, and providing accurate data for subsequent node offset detection.

[0067] Based on the printing finished image registration data, the printing pattern node offset situation is measured;

[0068] In the embodiments of the present application, the obtained registration data is used to measure the node offset. The pattern node usually refers to the feature points at the key positions in the image, such as the corner points or geometric center points of the pattern. Image analysis software (such as OpenCV or Matlab) is used to detect the nodes in the registered image, and locate the important nodes in the pattern. Then, the deviation between the actual node position of the pattern and the node position in the design image is calculated. The deviation is quantified by calculating the Euclidean distance of the node position, and a node offset data set is generated. If the value of the node offset is greater than the set threshold, it indicates that the pattern has a large deformation. Through this step, the local offset situation of the pattern can be quantified, and the basis for subsequent image deformation analysis is provided.

[0069] According to the printing pattern node offset situation, the image aspect ratio abnormal offset situation is evaluated;

[0070] In the embodiments of the present application, based on the obtained node offset data, the abnormal offset of the image aspect ratio is further evaluated. By comparing the aspect ratio of the printing finished pattern with that of the design pattern, if the offset amplitude exceeds the predetermined tolerance range, it is considered that the pattern has an aspect ratio abnormality. In specific implementation, by comparing the aspect ratio in the design with the actual aspect ratio of the registered pattern, the percentage deviation of the aspect ratio is calculated. If the deviation exceeds the set threshold, the image is considered to have a serious aspect ratio abnormality. This step helps to determine whether the pattern is affected by mechanical error or design deviation in the production process, and further determines the deformation degree of the image.

[0071] According to the image aspect ratio abnormal offset situation, the printing pattern stretch-compression degree is detected;

[0072] In the embodiment of the present application, the stretching or compression of the printed pattern is detected according to the abnormal deviation of the aspect ratio. The stretching or compression degree of the pattern is quantified by calculating the deviation of the aspect ratio. If the aspect ratio of the pattern deviates greatly, it indicates that the pattern is stretched or compressed. This step uses the proportional calculation formula to obtain the degree of stretching or compression of the pattern by comparing the actual aspect ratio of the pattern with the aspect ratio of the design drawing. For example, if the actual aspect ratio is greater than the design aspect ratio, it indicates that the pattern is stretched; if the actual aspect ratio is less than the design aspect ratio, it indicates that the pattern is compressed. This detection helps to accurately locate the type of pattern deformation and provides data support for the judgment of the abnormal state of distortion.

[0073] The abnormal state of distortion of the printed pattern is determined based on the stretching-compression degree of the printed pattern and the abnormal deviation of the aspect ratio of the image.

[0074] In the embodiment of the present application, the abnormal state of distortion of the pattern is comprehensively judged based on the obtained stretching and compression degree and the abnormal deviation of the aspect ratio. Specifically, if the stretching and compression degree exceeds the predetermined tolerance range or the deviation of the aspect ratio exceeds the set threshold, it is considered that the pattern is distorted. At this time, the appearance of the pattern will change significantly in geometric shape, resulting in a printing quality problem. In actual operation, by writing a judgment rule, the system can automatically detect whether the pattern is distorted by combining the abnormal data of stretching, compression and aspect ratio, and output the corresponding abnormal state report of distortion. This step provides the image quality evaluation result and provides a decision basis for subsequent defect repair or adjustment.

[0075] Preferably, the printed pattern detail gradient blur condition detection in step S2 comprises:

[0076] The texture features of the printed product image are optimized according to the printed product image;

[0077] In the embodiment of the present application, the image data of the printed product is obtained using a high-resolution camera or a scanner to ensure the quality optimization of the image, and the image is preprocessed using an image processing tool to eliminate noise and blur. The preprocessing step of image processing includes color normalization, denoising, sharpening, etc. Then, the optimized image is processed using a texture analysis algorithm to extract the texture features of the printed product. In this embodiment, Gabor filter and wavelet transform are used to extract local texture features of the image. Gabor filter can capture texture details in different directions of the image, and wavelet transform helps to analyze the frequency components of the image, thereby obtaining local detail information of the image. The texture features include the change pattern of the gray value in the local area of the image, the texture complexity, the directionality, etc. These texture feature data will constitute the basic data for image analysis for subsequent operations.

[0078] Identify image local texture energy attenuation trend based on printed product image texture feature

[0079] In the embodiment of the present application, the extracted texture feature data is further used to analyze the attenuation trend of local texture energy in the printed product image. According to the gray level variation of the image texture feature, the image is divided into multiple sub-blocks (for example, 8x8 pixel blocks). The texture energy of each sub-block is calculated, specifically by calculating the gray level variance or the sum of square of gradient amplitude in the local region. The texture energy of each sub-block reflects the richness of texture details in the region. If the texture energy of a certain local region is significantly lower than that of the adjacent region, it indicates that the texture details of the region are lost or attenuated. Further, by comparing the texture energy values of each sub-block, the region with attenuation trend is identified. The strength and range of the attenuation trend indicate the severity of the loss of image details.

[0080] Evaluate the degree of loss of printed pattern details according to the image local texture energy attenuation trend

[0081] In the embodiment of the present application, after identifying the local texture energy attenuation region, the degree of loss of pattern details in these regions needs to be evaluated. By analyzing each attenuation region in detail, the texture feature and energy attenuation degree of the image are quantitatively analyzed. If the texture energy value of a certain local region is low and the details of the region are complex, it can be judged that the degree of loss of details in the region is high. The degree of loss of details can be quantified by the relative difference of texture energy value, and a threshold is set. When the difference exceeds the threshold, it is marked as severe loss of details. In this way, the overall loss of details of the printed pattern is comprehensively evaluated, and the specific degree of loss of details in the pattern is obtained.

[0082] Detect the continuity interruption of printed product pattern according to the degree of loss of printed pattern details and the image local texture energy attenuation trend

[0083] In the embodiment of the present application, after obtaining the degree of loss of pattern details, the relationship between the detail loss region and the image local texture energy attenuation trend is analyzed to detect the continuity interruption of the pattern. The continuity of the printed pattern refers to the smooth transition and continuity of the pattern in the entire image. When the loss of details in a local region reaches a certain degree, the pattern appears discontinuous or broken. By analyzing the distribution of the detail loss region and combining the attenuation trend of the local texture energy, it can be judged whether the pattern has been broken. Specifically, when the texture energy of certain regions attenuates significantly and the degree of loss of details in the region is high, it can be considered that the continuity of the pattern has been interrupted.

[0084] Identify image texture direction vector offset condition based on printed product image texture feature

[0085] In the embodiment of the present application, the texture features of the printed product image are extracted, and then the texture direction vector offset conditions in the image are analyzed according to the texture features. Specifically, the texture direction vector of each local region in the image is obtained by calculating the gradient direction of the region. By analyzing the texture direction vector of the image, the region with direction offset is identified. When the texture direction in the image has a large deviation from the expected design texture direction, it indicates that the texture of the region has been offset, which usually causes the visual effect of the pattern to be poor or irregular. The calculation of the texture direction vector can be realized by local gradient operation, for example, using the Sobel operator to calculate the gradient of the image to obtain the direction information of each pixel. For each texture direction offset region, further evaluation and processing are performed.

[0086] The printed product pattern detail weakening condition is determined based on the image texture direction vector offset condition and the printed pattern detail loss degree;

[0087] In the embodiment of the present application, the offset condition of the image texture direction is identified, and then the detail weakening condition of the printed pattern is further analyzed in combination with the obtained pattern detail loss degree. The detail weakening condition is evaluated by comprehensively considering the offset degree of the texture direction vector and the loss degree of the pattern detail. If the texture direction offset is large and the detail loss is serious, it can be considered that the detail of the pattern has been obviously weakened. At this time, by calculating the comprehensive influence of the detail loss and the direction offset of each region, the detail weakening degree of the printed product pattern can be obtained. The evaluation process is performed by weighted average or weighted scoring, so as to quantify the detail loss and the texture offset into a comprehensive value for subsequent defect detection.

[0088] The printed pattern detail gradient blur condition is detected based on the printed product pattern detail weakening condition and the printed product pattern continuity interruption condition.

[0089] In the embodiment of the present application, the detail weakening condition of the printed product pattern is obtained, and then the detail gradient blur condition of the printed pattern is evaluated in combination with the detected pattern continuity interruption condition. Gradient blur refers to the transition of the printed pattern in the detail part becoming unclear, which is usually caused by the detail weakening and continuity interruption of the pattern. In this step, by comparing the detail weakening condition and the continuity interruption condition of each region, and combining gradient calculation (such as image gradient amplitude), it is judged whether the pattern has the phenomenon of detail gradient blur. If the pattern shows the phenomena of detail loss and continuity interruption in multiple regions, and the gradient information of these regions is weak, it indicates that the degree of detail gradient blur of the pattern is high. Through this process, the detail gradient blur condition of the printed pattern can be accurately detected, and further data support is provided for the printed defect detection model.

[0090] Preferably, the printed pattern visual performance degradation trend evaluation in step S2 comprises:

[0091] detecting a structure distortion condition of the printed pattern according to the gradient blur condition of the printed pattern details and the distortion abnormality condition of the printed pattern;

[0092] In the embodiment, the gradient information of the printed pattern is obtained by image gradient calculation, and then the gradient blur condition of the details is determined. The gradient blur of the details refers to the blurred transition of the edge of the details in the pattern, which is usually caused by texture loss, pattern continuity interruption or image resolution reduction. The gradient amplitude of each region of the image is calculated to obtain the detail definition of each region, and then it is determined whether the gradient blur occurs in the pattern. When the gradient amplitude of some regions of the pattern is lower than a preset threshold, it indicates that the visual definition of the region is reduced and the details are blurred. In addition, the image distortion detection algorithm is used to identify the distortion abnormality condition of the pattern. The distortion abnormality can be detected by calculating the geometric distortion and uneven illumination of the image. For example, the Hough transform is used to detect whether the geometric shape of the pattern is distorted, or the deviation between the expected shape and the actual shape of the image is compared by the image correction algorithm. By combining the gradient blur condition and the distortion abnormality condition, it is detected whether the structure distortion of the printed pattern exists, and then the structural integrity of the pattern is evaluated.

[0093] determining a difficulty growth trend of the printed pattern structure recognition based on the structure distortion condition of the printed pattern;

[0094] In the embodiment, the structure distortion condition of the pattern is obtained by the distortion abnormality detection and the gradient blur detection of the details. Next, the difficulty of the printed pattern recognition is evaluated according to the structure distortion condition. The structure distortion of the printed pattern usually means that the geometric shape or the detail information of the pattern is damaged, which will directly increase the difficulty of the pattern recognition. In this embodiment, by analyzing the range and degree of the structure distortion, the geometric features of the pattern (such as the smoothness of the lines, the symmetry of the pattern, etc.) are combined to evaluate the trend of the pattern recognition difficulty. Specifically, for the pattern regions with high distortion degree, the recognition difficulty will be significantly increased. By calculating the area of the structure distortion region and the boundary deformation condition, a structure recognition difficulty index is obtained, which reflects the recognition difficulty of the entire printed pattern. When the structure distortion region is extensive or the distortion is serious, the structure recognition difficulty index will gradually increase, indicating that the recognition difficulty continues to increase.

[0095] predicting a sharp decay condition of the printed pattern decoding rate according to the difficulty growth trend of the printed pattern structure recognition;

[0096] In the embodiment of the present application, based on the calculated structural recognition difficulty index, the decoding rate of the printed pattern is predicted next. The decoding rate refers to the ability of the system to successfully analyze the printed pattern, which is affected by factors such as pattern details, structural distortion, and pattern clarity. When the structural distortion of the pattern is severe, the decoding rate of the pattern will decrease sharply. In this embodiment, according to the growth trend of the recognition difficulty index, historical data and statistical methods are used to predict the decay of the decoding rate. For example, based on existing sample data, a relationship model is established to model the correlation between the structural distortion degree of the pattern and the decoding rate, and the decoding rate change calculated according to the distortion degree of the pattern. In specific operation, a functional relationship between the distortion degree and the decoding rate is established using regression analysis and other methods. Through this method, it can be predicted how the decoding rate of the printed pattern decays under different distortion degrees, and the corresponding decay trend is obtained.

[0097] According to the sharp decay of the decoding rate of the printed pattern and the growth trend of the structural recognition difficulty of the printed pattern, the information reading failure condition of the printed pattern is detected;

[0098] In the embodiment of the present application, after predicting the trend of sharp decay of the decoding rate, the information reading failure condition needs to be evaluated next. The information reading of the printed pattern usually depends on the clarity of the pattern and the accuracy of the structural recognition. When the decoding rate decays sharply, it indicates that the recognizability of the pattern decreases, thereby causing the information reading failure. The detection of information reading failure is completed by comparing the error rate and the success rate in the decoding process. If the decoding error rate increases significantly and the success rate continues to decrease, it is judged that the information reading failure condition occurs. In this process, the image processing tool is used to decode the pattern for testing, the number of successful decoding and the number of failed decoding of the pattern are counted, and the occurrence of information reading failure is detected according to the growth trend of the number of failed decoding. Specifically, the image recognition algorithm (such as OCR, barcode decoding algorithm) is used to decode the pattern multiple times, and the success or failure of each decoding is tracked. If the number of reading failures exceeds a certain threshold, it indicates that there is a serious recognition problem with the pattern.

[0099] According to the information reading failure condition of the printed pattern and the sharp decay of the decoding rate of the printed pattern, the printed pattern trace chain breakage condition is predicted;

[0100] In the embodiment of the present application, based on the obtained information reading failure condition and the sharp decay condition of the decoding rate, the next step is to predict the breaking condition of the tracing chain of the printed pattern. The tracing chain refers to the process that the information obtained by decoding the pattern can be traced back to the source of the pattern. If the pattern decoding fails or the information reading fails, it will directly lead to the breaking of the tracing chain. In this step, by combining the success and failure ratio of decoding with the decoding rate decay trend of the pattern, the probability of breaking the tracing chain is judged. Specifically, if the pattern decoding rate is lower than a certain preset threshold, and the number of information reading failures increases sharply, it means that the tracing chain of the pattern will be broken. The prediction process obtains the risk assessment of the breaking of the tracing chain by analyzing the historical data of decoding success and failure, and the influence of pattern distortion on the tracing chain.

[0101] The visual performance degradation trend of the printed pattern is evaluated based on the breaking condition of the tracing chain of the printed pattern and the information reading failure condition of the printed pattern.

[0102] In the embodiment of the present application, the visual performance degradation trend of the printed pattern is comprehensively evaluated by combining the predicted breaking condition of the tracing chain and the information reading failure condition. The visual performance degradation trend reflects the overall visual quality decline of the pattern in actual use due to distortion, decoding problems, etc. In this embodiment, by combining the risk of breaking the tracing chain and the condition of information reading failure, a comprehensive evaluation model is constructed to output the visual performance degradation index of the printed pattern. The index is calculated according to the degree of information reading failure, the trend of decoding rate decay and the integrity of the tracing chain. When the visual performance degradation index exceeds a certain threshold, it means that the visual performance of the pattern has decreased significantly and cannot meet the needs of actual application.

[0103] Preferably, step S3 comprises the following steps:

[0104] Step S31: determining the printed plate running abnormal condition based on the visual performance degradation trend of the printed pattern;

[0105] In the embodiment of the present application, the decreasing trend of the visual performance of the printed pattern is used as input, and the running abnormality of the printing plate is inferred by analyzing the visual performance degradation of the pattern. The decreasing trend of the visual performance of the pattern includes factors such as detail gradient blurring, distortion abnormality, and decoding rate attenuation, which reflect defects and abnormalities existing in the printing process. Based on the data analysis results, an evaluation model is constructed, which can infer whether the printing plate has abnormal running according to the distortion degree and recognition difficulty of the pattern. In the specific operation process, by comparing the pattern data in the normal state with the visual performance data of the current pattern, it is determined whether significant degradation or distortion has occurred. If the visual quality of the pattern significantly decreases, and the corresponding decoding rate or information reading failure probability significantly increases, it indicates that the printing plate has abnormal running. The structure and texture of the pattern are compared using image processing software, the visual degradation degree of each region is calculated, and the running abnormality of the printing plate is judged in combination with the known pattern running standard. Through this process, the abnormal state of the printing plate can be obtained, which provides a basis for subsequent defect positioning and degradation prediction.

[0106] Step S32: printing plate defect positioning processing is performed using the printing plate running abnormality condition to obtain printing plate defect positioning data;

[0107] In the embodiment of the present application, based on the printing plate running abnormality condition determined in step S31, printing plate defect positioning processing is performed next. In this step, the defect areas appearing in the pattern are analyzed according to the decreasing condition of the visual performance of the printed pattern. For example, image processing techniques such as edge detection, texture analysis, morphological operation, etc. are used to further identify specific defects existing in the pattern. These defects are manifested as uneven printing, pattern overlap, color difference, spots, line breakage, etc., and the specific manifestations depend on the condition of the printing plate. Through multiple filtering processing, sharpening and noise reduction, etc. of the pattern image, the printing defect area is highlighted, and the accurate defect position is obtained. In this process, edge detection algorithms such as Sobel and Canny, or image segmentation techniques are used to divide the pattern into regions, and the spatial position of each defect area is further determined. The obtained defect positioning data contains information such as the coordinates, area, shape, etc. of the defects. This data provides an accurate basis for the degradation trend prediction and structure aging evaluation in the subsequent steps.

[0108] Step S33: predicting the structure degradation trend of the printing plate defect area based on the printing plate defect positioning data;

[0109] In the embodiment of the present application, after obtaining the printing plate defect positioning data, the data is used to predict the structural degradation trend of the defect area. The degradation trend reflects the gradual damage of the structure of some areas of the printing plate due to factors such as repeated printing pressure, temperature changes, or physical wear during use. In this step, the positioning data needs to be analyzed to identify the type and degree of the defect area. By establishing a degradation trend analysis model, the model can simulate and predict the degradation behavior of the defect area over time. In specific operation, historical data and field test data are combined to analyze the degradation mode of the defect area using data regression analysis or empirical rules. For example, for cracks appearing on the surface of the printing plate, the crack propagation trend is predicted according to the crack development speed, depth, width, and other parameters. By calculating the degradation speed and defect area change trend, the overall degradation of the printing plate is further calculated. This process combines the regular scanning and image recording of the defect area by the data acquisition system to establish a degradation model, thereby obtaining the degradation prediction value of each defect area, generating a degradation trend curve, and clearly determining the degradation degree and future development direction of each defect area.

[0110] Step S34: Evaluate the structural aging degree of the printing plate according to the structural degradation trend of the printing plate defect area and the printing plate defect positioning data.

[0111] In the embodiment of the present application, after obtaining the degradation trend in step S33, the overall structural aging degree of the printing plate is further evaluated in combination with the defect positioning data. Structural aging refers to the physical performance decline of the printing plate due to long-term use, including surface hardness reduction, material fatigue, and printing accuracy reduction. In this step, according to the degradation trend of the defect area, combined with factors such as printing plate usage time and work load, a printing plate aging evaluation model is established. The model calculates the aging degree of the printing plate by combining the degradation trend and positioning data. For example, for a region with cracks, the aging degree of the region is calculated according to the crack propagation trend and the crack propagation area. If the crack area is large and the propagation trend is obvious, it indicates that the aging speed of the region is fast. Through this method, the overall aging condition of the printing plate is comprehensively evaluated. The specific evaluation result is presented in a numerical form, obtaining a printing plate structural aging index, reflecting the service life and current aging degree of the entire printing plate. The index can be used to guide subsequent printing plate maintenance, replacement, and improvement work.

[0112] Preferably, step S33 comprises the following steps:

[0113] Step S331: Measure the printing plate defect size data according to the printing plate defect positioning data;

[0114] In the embodiment of the present application, the specific size of each defect is further calculated based on the printing plate defect positioning data obtained in step S32. This process is achieved through image processing techniques, in which image segmentation techniques are used to identify each defect area. For each defect area, edge detection algorithms such as Canny edge detection or Sobel operator are used to accurately delineate the boundaries of the defect. Then, region growing algorithms are used to further determine the area, shape, boundary profile and other key features of the defect area. By calculating the area and geometric features (such as maximum length, minimum width, depth, etc.) of the defect area, the size data of the printing plate defect is obtained. These defect size data provide a basis for subsequent structural wear analysis and degradation trend prediction. During the measurement process, high-resolution scanners or digital imaging systems are used to obtain image data of the printing plate, and ensure that the image is clear and free of noise, avoiding measurement errors caused by poor image quality. In addition, in order to improve the measurement accuracy, multiple measurements and data averaging method are used to further optimize the accuracy of the defect size data.

[0115] Step S332: Detect the structural wear of the printing plate based on the printing plate defect positioning data;

[0116] In the embodiment of the present application, based on the defect size data obtained in step S331, the structural wear of the printing plate is detected. In this step, it is necessary to analyze whether the defects are concentrated in a specific area of the printing plate according to the distribution of the defects, and to calculate the distribution density of the defect area. By calculating the morphological changes (such as shape irregularity, crack propagation direction, etc.) of each defect area, the wear of the printing plate in these areas is inferred. Specifically, for the defects such as cracks or pits found, combined with the physical properties of the printing plate material (such as hardness, wear resistance, etc.), stress analysis methods are used to quantitatively analyze the wear of the defect area. For example, by using finite element analysis (FEA) technology, the stress distribution of each area is calculated by simulating the stress condition of the printing plate in the working environment, and the wear degree is inferred combined with the defect position. In this process, the stress analysis model used will take into account the material properties of the printing plate surface and interior, further inferring the wear condition of the printing plate. Through this process, the spatial distribution characteristics of the printing plate wear are obtained, providing data support for the stress growth trend prediction in the subsequent step.

[0117] Step S333: Test the structural stress growth trend of the printing plate based on the structural wear of the printing plate;

[0118] In the embodiment of the present application, based on the wear information obtained in step S332, the stress growth trend test of the printing plate structure is carried out next. This step analyzes the stress accumulation and growth of the printing plate during use by combining the stress-strain theory of material science. In specific operation, the stress distribution of each defect area and the overall printing plate is evaluated by simulating the working load and external environmental factors (such as temperature change, mechanical vibration, etc.) encountered by the printing plate during long-term use through finite element analysis (FEA). According to the structure wear condition, the stress value of each area is dynamically adjusted by the model to reflect the stress change caused by wear. Using the stress growth model, the stress evolution curve of the overall printing plate and each local area is obtained. These curves reflect how the printing plate material gradually accumulates stress during the wear process under different working conditions. Through this analysis, the stress growth of the printing plate in a specific area at a future time point is predicted, and the related stress increase trend data is obtained. These data provide a scientific basis for evaluating the structural aging and crack propagation of the printing plate.

[0119] Step S334: predicting the crack propagation trend of the printing plate according to the stress growth trend of the printing plate structure and the defect size data of the printing plate;

[0120] In the embodiment of the present application, the stress growth trend data obtained in step S333 is combined with the defect size data in step S331 to predict the crack propagation trend. Crack propagation is one of the main manifestations of the structural aging of the printing plate, therefore, accurately predicting the crack propagation trend is crucial for evaluating the service life of the printing plate. In this step, a crack propagation model is used to calculate the crack propagation behavior under different stress conditions based on the theory of material mechanics. Specifically, for each defect area, the speed and direction of crack propagation are predicted based on the stress growth trend and defect size data. For example, by calculating the stress concentration degree and local stress distribution at the crack tip, the crack propagation trend under different external forces is evaluated. The model predicts the crack propagation path, propagation speed, and the influence of the crack on the overall printing plate structure according to the initial position of the crack, stress field, and geometric characteristics of the defect, and the prediction result shows the extent of crack propagation at a future time point, guiding the subsequent maintenance and replacement decisions.

[0121] Step S335: evaluating the degree of decline in printing accuracy of the printing plate according to the crack propagation trend of the printing plate and the stress growth trend of the printing plate structure;

[0122] In the embodiment of the present application, according to the crack propagation trend predicted in step S334 and the stress growth trend in step S333, the degree of decline of the printing accuracy of the printing plate is further evaluated. When the printing plate has cracks, the cracks will affect the printing accuracy, causing phenomena such as pattern shift, loss or unclearness in the printing process. By establishing a printing accuracy evaluation model, the influence of the cracks on the printing quality is analyzed in combination with the prediction results of the crack propagation. In specific operation, according to the position and propagation trend of the cracks, the influence of the cracks on the pattern alignment and detail restoration in the printing process is evaluated. By simulating the printing process under different crack propagation degrees, the printing accuracy loss value under each crack state is obtained in combination with the printing accuracy standard. The accuracy loss value reflects the degree of decline of the printing quality caused by the cracks in the printing process. Through this process, the accuracy attenuation of the printing plate is clearly evaluated, and reference data is provided for subsequent process adjustment and maintenance.

[0123] Step S336: predicting the structural degradation trend of the defect area of the printing plate based on the degree of decline of the printing accuracy of the printing plate and the crack propagation trend of the printing plate.

[0124] In the embodiment of the present application, according to the degree of decline of the printing accuracy obtained in step S335 and the crack propagation trend in step S334, the structural degradation trend of the defect area of the printing plate is predicted. In this step, in combination with various factors of crack propagation and accuracy decline, the overall degradation trend of the printing plate is further predicted, the influence of the crack propagation of each defect area on the overall printing quality is evaluated, and the expansion speed and degradation degree of the degradation area are calculated in combination with the wear and stress distribution. Specifically, for the area that already has cracks, based on the crack propagation trend and stress growth, it is predicted how the physical properties of the area gradually decline, leading to the degradation of the overall structure of the printing plate. Through multi-factor analysis, the degradation trend of the entire printing plate is obtained, and prediction information about the degradation speed, area expansion, etc. is provided. These prediction information provides scientific decision support for subsequent maintenance, replacement and improvement of the printing plate.

[0125] Especially important is that step S335 includes the following steps:

[0126] Step S3351: detecting the change of the physical properties of the surface of the printing plate according to the crack propagation trend of the printing plate;

[0127] In the embodiment of the present application, according to the crack propagation trend of the printing plate, the crack propagation situation is detected by high-resolution image acquisition of the printing plate surface combined with microscope image analysis technology. The surface of the printing plate is scanned by using a high-resolution microscope camera to capture the surface detail image. By analyzing the crack propagation trend in the time sequence, the position, size and morphological change of the crack propagation are recorded, the crack features are extracted by using image processing software (such as OpenCV), and the specific area of crack propagation is identified. Combined with the position information of the crack propagation area, the physical properties of the area around the crack are measured by using precision instruments (such as surface roughness measuring instrument), and the parameters such as surface hardness, roughness and elastic modulus are particularly concerned. The changes of these physical properties are usually caused by the structural changes caused by crack propagation. Therefore, by measuring the changes of physical properties during crack propagation, the degradation of the surface material of the printing plate can be obtained, and data support for subsequent ink penetration prediction can be provided.

[0128] Step S3352: predicting the ink penetration of the printing plate based on the changes of the physical properties of the printing plate surface and the crack propagation trend of the printing plate;

[0129] In the embodiment of the present application, after detecting the changes of the physical properties of the printing plate surface, the ink penetration is analyzed combined with the crack propagation trend. By correlating the physical property data (such as surface roughness, hardness and elastic modulus) with the crack propagation trend, the ink penetration characteristics on the surface of the printing plate are derived. For the crack propagation area, the surface hardness of the area will usually decrease and the roughness will increase, and these changes will directly affect the penetration depth of the ink. The ink penetration rate is calculated by using a physical model, which considers the surface degradation effect caused by the crack. By establishing the relationship between the surface roughness and the ink penetration rate, the ink penetration situation of the crack propagation area is predicted. Combined with the surface property data of the printing plate and the crack propagation trend, a prediction model is constructed to obtain the spatial distribution of the ink penetration degree of each area, and further determine which areas have the risk of uneven ink distribution.

[0130] Step S3353: determining the uneven ink distribution on the surface of the printing plate according to the ink penetration of the printing plate;

[0131] In the embodiment of the present application, the spatial distribution data obtained according to the ink penetration is used to determine whether the ink distribution on the printing plate surface is uniform. By analyzing the depth and speed difference of ink penetration, combined with the influence of crack propagation area, the uniformity of the surface ink distribution is determined. In actual operation, high-precision ink sensors or ink measuring instruments are used to measure the ink content in different areas of the printing plate surface, and the measurement data is compared with the predicted data. If there is a region with large ink penetration depth, it will lead to uneven ink distribution. This problem affects the quality of the printed matter, especially in detail processing and color restoration. Therefore, through the evaluation of the ink penetration, it is clear whether the phenomenon of uneven ink distribution exists on the surface of the printing plate, and further provides the basis for the evaluation of the decline of printing accuracy.

[0132] Step S3354: estimating the deformation degree of the printing plate based on the stress growth trend of the printing plate structure;

[0133] In the embodiment of the present application, the deformation degree of the printing plate is estimated according to the stress growth trend of the printing plate. The structural stress refers to the cumulative effect of the mechanical load received by the printing plate during operation. In the printing process, long-term friction, pressure and thermal effects will cause deformation on the surface of the printing plate. The stress sensor is used to monitor the surface of the printing plate in real time, and the stress distribution data of the printing plate is collected. By analyzing the stress distribution change, combined with the stress-deformation theoretical model, the deformation degree of the printing plate is evaluated. Specifically, in actual operation, the stress analyzer (such as strain gauge) is used to detect the change of the stress on the surface of the printing plate. Through time series analysis of the stress data, the specific trend of the deformation of the printing plate can be revealed. This data is used as the basis for judging whether the printing plate has undergone significant deformation. The printing plate with large deformation degree will directly affect the accuracy and quality of the printing.

[0134] Step S3355: detecting the unevenness of the printing plate surface according to the deformation degree of the printing plate;

[0135] In the embodiment of the present application, based on the deformation data obtained in the previous step, the three-dimensional scanner and other high-precision surface detection equipment are used to detect the unevenness of the printing plate surface. By processing the three-dimensional point cloud data of the printing plate surface, it can be clearly identified whether the printing plate has surface unevenness, especially the deformation of the stress area. By analyzing the degree of surface unevenness, combined with the previously obtained deformation data, the actual state of the printing plate is further confirmed. For example, the deformation of some areas causes the surface to be concave or convex, which will affect the uniform distribution of ink and the printing effect. Therefore, the detection of surface unevenness provides an important basis for the subsequent evaluation of printing accuracy.

[0136] Step S336: evaluating the degree of decline of the printing accuracy of the printing plate based on the unevenness of the printing plate surface and the uneven distribution of the ink on the printing plate surface.

[0137] In the embodiment of the present application, the degree of decline of the printing accuracy of the printing plate is evaluated by comprehensively considering the surface unevenness and the uneven ink distribution, the data of the surface unevenness and the data of the uneven ink distribution are analyzed in association, and it is found out which uneven regions have the phenomenon of uneven ink penetration. These regions usually cause the quality of the printed matter to decline, especially in the printing process with high accuracy requirements. Then, according to the surface unevenness and the uneven ink distribution, the degree of decline of the printing accuracy of each region is calculated by using a weight algorithm, and the percentage of decline of the printing accuracy of the whole printing plate is obtained, which further provides a guidance basis for subsequent optimization or replacement of the printing plate.

[0138] Especially important is that step S34 comprises the following steps:

[0139] Step S341: detecting the degree of growth of the friction of the printing plate according to the structural degradation trend of the defect region of the printing plate and the defect positioning data of the printing plate;

[0140] In the embodiment of the present application, the image data of the defect region of the printing plate surface is obtained by using high-precision image acquisition technology (such as a microscope, a scanning electron microscope, etc.). The image is analyzed by using image processing software (such as OpenCV), and the specific position, size and morphology of the defect region are extracted. Combined with these data, the structural degradation trend of the region is further analyzed. Specifically, by comparing the printing plate images collected at different times, the expansion trend of the defect region can be identified, and then the structural degradation degree of the region is deduced. After obtaining the positioning data of the defect region, real-time monitoring is performed in combination with a friction sensor (such as a surface stress sensor or a friction coefficient sensor). The friction sensor can measure the change of the friction force generated by the printing plate surface during operation, especially in the high-friction region. Through long-term data acquisition and recording, combined with the degradation trend of the defect region, the growth of the friction force of the printing plate in the specific region is analyzed. The speed and degree of friction growth are closely related to the degree of structural degradation, so the friction force data are used to infer the aging degree of the defect region. In this process, by continuously tracking the relationship between the friction force data and the structural degradation trend, the specific degree of friction growth is determined, and the friction growth speed of the printing plate is obtained. This process predicts the future trend of friction growth by setting a threshold or using a data fitting model.

[0141] Step S342: determining the degree of thermal expansion of the printing plate based on the degree of growth of the friction of the printing plate;

[0142] In the embodiment of the present application, the thermal expansion degree of the printing plate is further derived from the data of the friction growth degree. The increase of friction is usually accompanied by the rise of local temperature, especially in the printing process, the heat generated in the friction area of the printing plate will cause thermal expansion phenomenon. In order to accurately measure the thermal expansion degree, it is necessary to use high-precision temperature sensors (such as infrared thermal imager, thermocouple, etc.) to monitor the temperature change of the surface of the printing plate in real time. According to the relationship between the friction growth data and the surface temperature data, the thermal expansion degree of the printing plate is calculated through the thermal expansion theoretical model. The specific operation is: according to the rate of temperature rise of the surface of the printing plate and the friction data, the thermal expansion coefficient (determined by the characteristics of the printing plate material) is applied to estimate the thermal expansion change of the material. Generally, when the friction increases, the temperature of the local area will rise, causing the volume of the area to expand. By calculating the thermal expansion degree of the area, the influence of thermal effect on the deformation of the surface of the printing plate can be obtained. In this step, the temperature distribution map and the friction distribution map are combined for analysis, further verifying the relationship between thermal expansion and friction growth, and accurately calculating the deformation data caused by thermal expansion.

[0143] Step S343: According to the friction growth degree of the printing plate and the friction growth degree, the area gradient growth trend of the defect area is predicted.

[0144] In the embodiment of the present application, after obtaining the friction growth degree of the printing plate, the growth trend of the defect area is predicted in combination with the friction data of the defect area. By comparing the friction data and the structure degradation trend of different areas, a correlation model between friction and defect area growth is established, and the trend of friction change with time is obtained through long-time friction data monitoring. Then, the friction growth data and the expansion trend of the defect area are combined, and the gradient growth of the defect area is predicted by data fitting method. The prediction process is based on the physical correlation between friction and defect expansion. The increase of friction usually means that the surface damage of the local area is aggravated, which further leads to the expansion of the defect area. Therefore, by collecting friction data in real time, the future expansion range and expansion speed of these defect areas are predicted. Specifically, regression analysis, trend analysis and other data processing techniques are used to derive the growth trend of the defect area in combination with the known friction growth data.

[0145] Step S344: According to the area gradient growth trend of the defect area, the structure aging degree of the printing plate is estimated.

[0146] In the embodiments of the present application, the structural aging degree of the printing plate is further estimated according to the gradient growth trend of the defect area. Structural aging mainly reflects material degradation, surface damage and functional decline. By combining the gradient growth trend of the defect area, a relationship model between defect growth and structural aging is established. Specifically, the area growth rate of the defect area reflects the aging degree of the surface material of the printing plate. To achieve this, the surface of the printing plate is scanned in detail using a surface roughness instrument (such as an atomic force microscope) to obtain structural change data of the defect area. Then, by analyzing the correlation between the area change of the defect area and the physical properties such as surface hardness and elastic modulus, the aging degree of the printing plate is further calculated. By analyzing the area growth data of multiple defect areas, the gradient growth trend between the areas is constructed, and thus the overall aging of the printing plate is calculated. This process involves the comprehensive evaluation of multiple parameters, such as crack propagation speed, surface friction increase, temperature change, etc. Combined with these data, the structural aging degree of the printing plate can be accurately estimated.

[0147] Preferably, step S4 comprises the following steps:

[0148] Step S41: predicting the progressive error growth trend of the printing plate according to the structural aging degree of the printing plate;

[0149] In the embodiments of the present application, by analyzing the structural aging of the printing plate, the progressive error growth trend is predicted, and based on the structural wear and crack propagation data obtained in the previous steps (such as steps S33 and S332), the structural aging model is constructed by combining the use history of the printing plate, the working environment (such as temperature, humidity, mechanical load, etc.). Using these data, the stress distribution and gradual wear process of the printing plate in continuous use are simulated by stress analysis methods (such as finite element analysis). In the structural aging model, the wear rate of different regions, the crack propagation path, and the physical properties of the printing plate (such as hardness, elastic modulus, etc.) are considered, and the printing plate is comprehensively analyzed. In the specific operation, experimental data is collected, including multiple load tests under different conditions, and the parameters of the aging model are adjusted according to the experimental results. Through long-term aging experiments, the progressive error growth trend of different regions of the printing plate is obtained. For example, as the use time of the printing plate increases, the cracks will gradually expand, thereby affecting the printing accuracy and leading to the cumulative growth of the progressive error. The predicted error growth trend data can be used as a basis for judging the printing abnormalities in the subsequent steps.

[0150] Step S42: determining the printing abnormality of the printing plate according to the progressive error growth trend of the printing plate;

[0151] In the embodiment of the present application, after obtaining the progressive error growth trend of the printing plate, the printing abnormality is judged. This step mainly compares and analyzes the progressive error with the working state of the printing plate to detect whether the printing abnormality occurs. By measuring and recording the actual error value of each batch of printing and comparing it with the standard error in the initial state, the gradually increasing error value is obtained. If the error value exceeds the set tolerance range, it is considered that the printing plate has an abnormality. In order to accurately judge the abnormality, multiple error thresholds are set to classify and manage different error sizes. For example, if the error exceeds a certain threshold, it causes the printing pattern to be blurred or displaced, showing serious printing abnormality. By continuously tracking and recording the change of error trend, it is determined when the abnormality occurs and the severity of the abnormality is evaluated. This process needs to combine the continuous monitoring of printing quality in actual production, and use image processing tools (such as images taken by high-precision cameras) to compare the finished pattern of each printing in detail, so as to quantify the error and identify the abnormality in time.

[0152] Step S43: predicting the structural performance failure state of the printing plate based on the printing abnormality of the printing plate and the progressive error growth trend of the printing plate;

[0153] In the embodiment of the present application, this step further predicts the structural performance failure state of the printing plate by combining the progressive error growth trend of the printing plate and the printing abnormality. In the foregoing step, the error change data generated during the use of the printing plate and the gradually expanding crack situation have been obtained. Using these data, combined with the theory of material mechanics and fatigue analysis, a failure prediction model is established. Specifically, by calculating the stress distribution and defect expansion of the printing plate at different periods, the influence on the printing performance is analyzed. The known error growth trend is superimposed with the crack expansion trend to predict that at a certain time point, the printing plate will have a performance loss that cannot be restored due to structural aging and further expansion of cracks. In specific operation, by simulating the influence of various external stresses (such as impact, vibration, etc.) on the printing plate, combined with the gradually increasing error, the occurrence time and failure mode of the failure state are predicted. For example, the further expansion of the crack leads to the complete failure of the local area, causing the pattern to be unable to be accurately printed, and further affecting the printing quality of the entire production line. The purpose of this step is to predict the failure point of the structural performance by analyzing the gradual accumulation of aging and damage, and to provide data support for subsequent preventive measures.

[0154] Step S44: determining the printing abnormality defect situation according to the structural performance failure state of the printing plate and the visual performance decreasing state of the printing pattern;

[0155] In the embodiment of the present application, after obtaining the structural performance failure state of the printing plate in step S43, the specific situation of the printing abnormal defect is determined by further combining the visual performance degradation state of the printing pattern, obtaining a high-definition image of the printed matter through image acquisition technology, and performing detailed analysis to record the visual effect changes of the pattern in each printing process. By comparing the deviation of the actual pattern and the standard pattern, the error types in the printing process are quantified, such as position deviation, blur, color difference, etc. In particular, as the printing plate ages and the structural performance fails, the visual characteristics of the pattern such as clarity and alignment will gradually deteriorate. Therefore, for each printing process, an image is obtained through a high-precision camera system or a scanning system, and compared with a standard pattern, and an image recognition technology (such as edge detection, feature matching, etc.) is used to detect the deviation degree of the pattern. In this step, historical data and aging models also need to be combined to analyze the trend of the visual performance degradation of the pattern, and the visual effect changes in each printing process are evaluated, so as to accurately identify the printing abnormal defect. This analysis result will help to identify and classify abnormal defects and ensure accurate positioning of printing quality problems.

[0156] Step S45: using the printing abnormal defect situation to perform model training optimization on the printing defect detection model to obtain a printing defect detection optimized training model.

[0157] In the embodiment of the present application, after identifying the printing abnormal defect in step S44, the defect data is used to optimize and train the printing defect detection model. In this step, the abnormal defect situation and pattern deviation data detected in the foregoing steps are input into the defect detection model as a training set. The model is a defect detection algorithm based on image recognition technology (such as deep learning, convolutional neural network, etc.), which is specially used for identifying and classifying printing defects. By inputting a large amount of abnormal data, the model performs back propagation and parameter adjustment to optimize its detection accuracy. In the training process, data including different types of defect patterns (such as cracks, blur, displacement, etc.) are used, and are labeled and classified according to the severity of the defects. Through continuous optimization, a detection model capable of accurately detecting and classifying various printing defects is obtained. The optimized model can detect printing defects in real time in future production and provide timely feedback to ensure printing quality. Data enhancement techniques are also introduced in the model training process, and through operations such as rotation, cropping, transformation, etc. on the defect patterns, the robustness and accuracy of the model are further improved.

[0158] The present application also provides a training system applied to a printing defect detection model, which is used to execute the training method applied to the printing defect detection model as described above, and the training system applied to the printing defect detection model comprises:

[0159] The model construction module is configured to acquire appearance data of a printed product; perform image acquisition on the appearance data of the printed product to obtain printed product image data; perform image optimization processing on the printed product image data to obtain printed product image optimization data; and construct a printing defect detection model according to the printed product image optimization data to obtain the printing defect detection model.

[0160] The visual performance degradation trend evaluation module is configured to determine a distortion abnormality state of a printed pattern according to the printed product image optimization data; detect a gradient blurring condition of the printed pattern details according to the printed product image optimization data; and evaluate a visual performance degradation trend of the printed pattern based on the gradient blurring condition of the printed pattern details and the distortion abnormality state of the printed pattern.

[0161] The structure aging degree evaluation module is configured to determine a printing plate running abnormality condition based on the visual performance degradation trend of the printed pattern; perform printing plate defect positioning processing using the printing plate running abnormality condition to obtain printing plate defect positioning data; and evaluate a structure aging degree of the printing plate according to the printing plate defect positioning data.

[0162] The model training optimization module is configured to predict a printing plate progressive error growth trend according to the structure aging degree of the printing plate; determine a printing abnormality defect condition according to the printing plate progressive error growth trend and the visual performance degradation trend of the printed pattern; and perform model training optimization on the printing defect detection model using the printing abnormality defect condition to obtain a printing defect detection optimized training model.

[0163] The above description is merely a specific implementation of the present application, which enables a person skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application shall not be limited to the embodiments shown herein, but shall conform to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A training method for a printing defect detection model, characterized in that, Includes the following steps: Step S1: Obtain the appearance data of the printed product; collect images of the printed product from the appearance data of the printed product to obtain the image data of the printed product; Image optimization processing is performed on the printed product image data to obtain the printed product image optimization data; A printing defect detection model is constructed based on optimized data of printed product images. The model involves extracting edge distribution maps from each image and constructing an image texture feature matrix by combining the mean and standard deviation of grayscale distribution. Using image block sliding window analysis technology, the entire image area is scanned in a 10×10 pixel sliding window manner to statistically analyze the degree of color shift within local blocks, boundary breakage features, and local texture discontinuities, forming a corresponding defect label coordinate set. The feature data and image coordinates are labeled and organized into a structured training sample set in matrix form. A discrimination method based on feature rule set is used to construct the model, using defect type as the output label and the degree of feature anomaly of the corresponding image area as the input judgment factor to construct a defect detection model in the form of a decision tree. Step S2: Determine the abnormal state of the printed pattern distortion based on the optimized image data of the printed finished product; detect the gradient blur of the printed pattern details based on the optimized image data of the printed finished product; evaluate the declining trend of the visual performance of the printed pattern based on the gradient blur of the printed pattern details and the abnormal state of the printed pattern distortion. Step S3: Determine the abnormal operating conditions of the printing plate based on the declining visual performance of the printed pattern; use the abnormal operating conditions of the printing plate to perform defect localization processing and obtain printing plate defect localization data; evaluate the degree of aging of the printing plate structure based on the printing plate defect localization data. Step S4: Predict the progressive error growth trend of the printing plate based on the aging degree of the printing plate structure; determine the printing abnormality defects based on the progressive error growth trend of the printing plate and the decreasing visual performance of the printed pattern; use the printing abnormality defects to train and optimize the printing defect detection model to obtain an optimized training model for printing defect detection. The model training and optimization includes: using the detected abnormality defects as a training set and inputting them into the defect detection model; by inputting a large amount of abnormal data, the model performs backpropagation and parameter adjustment to optimize its detection accuracy. During the training process, the data used includes different types of defect patterns, and the data is classified according to the severity of the defects. Through continuous optimization, a detection model that can accurately detect and classify various printing defects is obtained.

2. The training method for a printing defect detection model according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Set the resolution of the visible light camera to 1920×1080 pixels, the minimum light sensitivity to 0.01 lux, and the frame rate to 30 FPS; Step S12: Obtain the appearance data of the printed finished product; Step S13: Acquire images of the printed finished product based on the appearance data of the printed finished product obtained by the visible light camera, and obtain image data of the printed finished product; Step S14: Perform image optimization processing on the printed product image data to obtain the printed product image optimization data; Step S15: Build a printing defect detection model based on the optimized data of the printed product image to obtain the printing defect detection model.

3. The training method for a printing defect detection model according to claim 2, characterized in that, Step S14 includes the following steps: Step S141: Perform gradient amplitude distribution analysis on the printed product image data to obtain the gradient amplitude distribution of the printed product image; Step S142: Evaluate the sharpness of the printed image based on the gradient amplitude distribution of the printed image. Specifically, for the gradient amplitude of all pixels, the distribution is statistically analyzed to obtain the overall sharpness and clarity of the image. When the average gradient amplitude of the image is greater than 60, the image is considered to have a high level of clarity. Step S143: Based on the printed product image data, perform local texture intensity recognition on the printed product image to obtain local texture intensity data of the image; Step S144: Evaluate the initial quality of the printed product image based on the local texture intensity data of the image and the sharpness of the printed product image; Step S145: Divide the initial quality of the printed finished image into 2200 grid regions to obtain the grid region data of the printed finished image quality; Step S146: Detect the high-frequency component loss area in the finished product image based on the grid area data of the printed finished product image quality; Step S147: Perform image optimization processing on the areas where high-frequency components are missing in the finished product image to obtain optimized data for the printed finished product image.

4. The training method for a printing defect detection model according to claim 1, characterized in that, The determination of the abnormal state of printed pattern distortion in step S2 includes: Optimize the data acquisition of the external outline of the printed pattern based on the image of the printed product; The overall proportion changes of the printed pattern are measured based on the optimized data of the printed finished product image. By integrating the external outline of the printed pattern and the overall proportion changes of the printed pattern, the characteristic data of the printed pattern are obtained. Obtain the printed image structure design data, and perform image registration processing on the printed finished product pattern feature data based on the printed image structure design data to obtain the printed finished product image registration data; Measurement of printed pattern node offset based on registration data of printed finished product images; Evaluate the image aspect ratio anomaly based on the offset of printed pattern nodes; Detect the degree of stretching-compression of printed patterns based on abnormal aspect ratio shifts in images; The abnormal state of printed pattern distortion is determined based on the degree of stretching-compression of the printed pattern and the abnormal offset of the image aspect ratio.

5. The training method for a printing defect detection model according to claim 1, characterized in that, Step S2, the detection of gradient blurring of printed pattern details, includes: Based on the optimized data acquisition of printed finished product images, texture features of printed finished product images are obtained. Identify the local texture energy decay trend of printed product images based on texture features; The degree of detail loss in printed patterns is assessed based on the local texture energy decay trend of the image. The continuity of the printed pattern is detected based on the degree of loss of printed pattern details and the trend of local texture energy decay in the image. Identify the texture direction vector offset of printed finished product images based on their texture features; The degree of weakening of pattern details in the printed product is determined based on the image texture direction vector offset and the degree of loss of printed pattern details. The gradient blurring of printed pattern details is detected based on the weakening of details and the interruption of the continuity of the printed pattern.

6. The training method for a printing defect detection model according to claim 1, characterized in that, The assessment of the declining visual performance of the printed pattern in step S2 includes: Detect the structural distortion of printed patterns based on the gradient blurring of printed pattern details and the abnormal distortion state of printed patterns. The increasing difficulty of recognizing printed pattern structures is determined based on the distortion of printed pattern structures. Based on the increasing trend of difficulty in recognizing printed pattern structures, predict the sharp decline in the decoding rate of printed patterns. Detect printing pattern information reading failures based on the sharp decline in the decoding rate of printed patterns and the increasing difficulty in recognizing printed pattern structures. Predict the breakage of the printing pattern traceability chain based on the failure to read printing pattern information and the sharp decline in printing pattern decoding rate; The decline in the visual performance of printed patterns is assessed based on the breakage of the traceability chain and the failure to read printed pattern information.

7. The training method for a printing defect detection model according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Determine the abnormal operating conditions of the printing plate based on the declining visual performance of the printed pattern; Step S32: Utilize the abnormal operation status of the printing plate to perform printing plate defect location processing and obtain printing plate defect location data; Step S33: Predict the structural degradation trend of the defective area of ​​the printing plate based on the defect location data; Step S34: Assess the degree of structural aging of the printing plate based on the structural degradation trend of the defective areas and the location data of the defects.

8. The training method for a printing defect detection model according to claim 7, characterized in that, Step S33 includes the following steps: Step S331: Measure the size of the printing plate defects based on the printing plate defect location data; Step S332: Detect the wear condition of the printing plate structure based on the printing plate defect location data; Step S333: Test the stress growth trend of the printing plate structure based on the wear condition of the printing plate structure; Step S334: Predict the crack propagation trend of the printing plate based on the stress growth trend of the printing plate structure and the size data of printing plate defects; Step S335: Assess the degree of decline in printing accuracy of the printing plate based on the crack propagation trend and the stress growth trend of the printing plate structure; Step S336: Predict the structural degradation trend of the defective area of ​​the printing plate based on the degree of decline in printing accuracy and the crack propagation trend of the printing plate.

9. The training method for a printing defect detection model according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Predict the growth trend of progressive error of printing plate based on the aging degree of printing plate structure; Step S42: Determine the printing plate printing anomalies based on the increasing trend of progressive error in the printing plate; Step S43: Predict the failure state of the printing plate structure performance based on printing plate printing anomalies and the progressive error growth trend of the printing plate; Step S44: Determine the printing defects based on the failure state of the printing plate structure and the decreasing trend of the visual performance of the printed pattern. Step S45: Use the printing abnormality and defect situation to train and optimize the printing defect detection model to obtain the optimized training model for printing defect detection.

10. A training system for a printing defect detection model, characterized in that, For performing the training method for a printing defect detection model as described in claim 1, the training system for the printing defect detection model includes: The model building module is used to acquire the appearance data of printed products; to collect images of printed products from the appearance data of printed products, and to obtain image data of printed products; to perform image optimization processing on the image data of printed products, and to obtain optimized image data of printed products; and to build a printing defect detection model based on the optimized image data of printed products, thereby obtaining a printing defect detection model. In this model, the edge distribution map of each image is extracted, and an image texture feature matrix is ​​constructed by combining the mean and standard deviation of gray-level distribution. Using the image block sliding window analysis technique, the entire image area is scanned in a 10×10 pixel sliding window manner to statistically analyze the degree of color shift, boundary breakage features, and local texture discontinuity within local blocks, and to form a corresponding defect label coordinate set. The feature data and image coordinates are labeled and organized into a structured training sample set in matrix form. A discrimination method based on feature rule set is used for model construction, using the defect type as the output label and the feature anomaly degree of the corresponding image area as the input judgment factor to construct a defect detection model in the form of a decision tree. The visual performance degradation assessment module is used to determine the abnormal state of printed pattern distortion based on the optimized image data of the printed finished product; detect the gradient blur of printed pattern details based on the optimized image data of the printed finished product; and assess the degradation trend of printed pattern visual performance based on the gradient blur of printed pattern details and the abnormal state of printed pattern distortion. The structural aging assessment module is used to determine the abnormal operating conditions of the printing plate based on the declining visual performance of the printed pattern; to perform defect localization processing on the printing plate using the abnormal operating conditions, thereby obtaining defect localization data; and to assess the structural aging degree of the printing plate based on the defect localization data. The model training and optimization module is used to predict the progressive error growth trend of the printing plate based on the aging degree of the printing plate structure; to determine the printing abnormality defects based on the progressive error growth trend of the printing plate and the decreasing visual performance of the printed pattern; and to train and optimize the printing defect detection model using the printing abnormality defects to obtain an optimized training model for printing defect detection. In the model training and optimization process, the detected abnormal defects are used as a training set and input into the defect detection model. By inputting a large amount of abnormal data, the model performs backpropagation and parameter adjustment to optimize its detection accuracy. During the training process, the data used includes different types of defect patterns, and labels are classified according to the severity of the defects. Through continuous optimization, a detection model capable of accurately detecting and classifying various printing defects is obtained.

Images