Digital printing machine image quality detection method and system

Abstract

The invention discloses a digital calico printing machine image quality detection method and system, and relates to the technical field of digital calico printing machine management.The method comprises the steps that first state information and second state information are determined; based on comparative analysis of the first state information and the second state information, change information is sent to the ink jet mechanism, and the change information is used for indicating the second state information; the trained prediction model is connected to a draft tracing mechanism, patterns are input, and the patterns of the digital printing machine are restored and output through regional analysis; according to the technical key points, an initial prediction model is established, a relation model of printing pressure, scraping printing speed, ink viscosity, scraping printing angle and ink transfer rate is established by utilizing a BP neural network, state estimation is combined, and the prediction model is trained and updated, so that real-time correction of an ink jet mechanism is realized, and the misjudgment rate is reduced; and the color features and the contour features are further subjected to targeted analysis, so that the quality of the output image is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital printing machine management, and in particular to a method and system for detecting image quality of a digital printing machine. Background Art

[0002] Digital printing is a comprehensive discipline that integrates pattern design, computer information technology, mechanical processing, and printing and dyeing. Its working principle is basically the same as that of an inkjet printer. Specifically, it introduces flat screen and rotary screen inkjet, laser imagesetters, and printing CAD systems on the basis of traditional printing technology. It uses inkjet nozzles for printing and dyeing, which shortens the printing cycle and improves the accuracy of printed colors. The digital printing process directly affects the quality and cost of printing. It is limited by the stability of the ink formula and the calibration accuracy of the nozzle, and is prone to color difference due to uneven ink droplet injection or differences in the ink absorbency of the substrate. In the traditional 3D digital printing process, printing pressure, scraping speed, scraping angle and ink viscosity have a great influence on the printing quality. The interaction between these factors makes it difficult to determine the printing quality. The existing algorithms and data processing capabilities are unable to cope with complex industrial scenarios. In addition, multi-dimensional manual visual inspection and simple threshold segmentation cannot quantitatively analyze color or contour features, resulting in a large number of test prints and a high ink waste rate. At the same time, the multi-dimensional static parameter settings of digital printing machine-related equipment (such as inkjet mechanisms) cannot adapt to material / environmental changes. The detection dimension is single, resulting in poor real-time performance and low intelligence level. Summary of the Invention

[0003] (1) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides a method and system for detecting image quality of a digital printing machine. By establishing an initial prediction model and using a BP neural network to establish a relationship model between printing pressure, scraping speed, ink viscosity, scraping angle and ink transfer rate, combined with state estimation, the prediction model is trained and updated to achieve real-time correction of the inkjet mechanism and reduce the misjudgment rate; further targeted analysis of color features and contour features is performed to improve the output image quality, thereby solving the problems raised in the background technology.

[0004] (2) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: In a first aspect, the present application provides a method for detecting image quality of a digital printing machine, which is applied to a digital printer, and the digital printer includes at least an inkjet mechanism, a drafting mechanism, and an image output mechanism, and the method includes: Determining first state information and second state information; wherein the first state information is obtained by prediction using a preset initial prediction model and includes at least an ink transfer rate, and the second state information is determined by state estimation of an inkjet pulse signal output by an inkjet mechanism and includes at least an ink droplet kinetic energy evaluation coefficient; Compare the difference measurement value of the first state information and the second state information, and import the difference measurement value into the preset difference threshold for judgment: If the difference metric value is greater than or equal to the difference threshold, the change information is sent to the inkjet mechanism, and the initial prediction model is updated based on the training of the change information to obtain a quality prediction model; If the difference measure value is less than the difference threshold, a maintenance message is sent to the inkjet mechanism, and the initial prediction model is used as the quality prediction model; The quality prediction model is connected to the drafting mechanism, the first pattern is scanned and input, the first pattern area is analyzed, the second pattern is restored, and the image quality detection result is obtained based on the first pattern and the second pattern.

[0005] Furthermore, the initial prediction model is constructed based on a BP neural network, and the initial prediction model includes optimizing weights and thresholds of the BP neural network using a particle swarm algorithm; The input parameters of the initial prediction model are printing process parameters, and the output parameters are ink transfer rate; wherein the printing process parameters include at least scraper pressure, scraper speed, scraper angle and ink viscosity.

[0006] Furthermore, the second state information is determined by performing state estimation on the inkjet pulse signal output by the inkjet mechanism, including: When receiving the inkjet pulse signal, the ink drop ejection direction and ejection speed of the inkjet mechanism are immediately obtained; The ejection direction is decomposed, and the nozzle axis is used as the baseline. The angle between the ejection direction of the ink droplet and the baseline is marked as the deflection inclination angle; the deflection inclination angle is combined with the ejection speed to obtain the kinetic energy evaluation coefficient of the ink droplet.

[0007] Furthermore, the kinetic energy evaluation coefficient of the ink droplet is obtained, including: When the deflection angle is greater than the preset deflection threshold, the environmental parameters at this time are collected, and machine learning is used to establish a mapping relationship between the environmental parameters and the angle compensation amount, and the kinetic energy evaluation coefficient is corrected; among them, the environmental parameters include at least temperature and humidity.

[0008] Furthermore, the initial prediction model is trained and updated based on the change information to obtain a quality prediction model, including: The initial prediction model is trained again based on the change information and the second state information to obtain a third prediction model; Determining first variation information of the initial prediction model and the second prediction model based on the initial prediction model and the second prediction model, and sending first change information to the inkjet mechanism if the first variation information is greater than or equal to a preset variation threshold, the first change information being used to indicate information of the second prediction model; wherein the information of the second prediction model includes: variation information between the second prediction model and the initial prediction model; Determining second variation information of the initial prediction model and the third prediction model based on the initial prediction model and the third prediction model, and sending second change information to the inkjet mechanism if the second variation information is greater than or equal to a variation threshold, the second change information being used to indicate information of the third prediction model; wherein the information of the third prediction model includes variation information between the third prediction model and the initial prediction model; If the inkjet mechanism receives any one of the maintenance information, the first change information or the second change information, the corresponding initial prediction model, the second prediction model or the third prediction model is selected as the prediction model for real-time training, and the prediction model is obtained by training and updating.

[0009] Furthermore, the first pattern area is analyzed, including: Divide the first pattern into several areas, extract several pixels in each area and their color features and contour features, perform contour matching and color matching, obtain corresponding matching differences, and set a difference gradient range of ±wc%. When the matching difference is within the difference gradient range, send an inkjet pulse signal to perform inkjet operation; Otherwise, an adjustment analysis is performed until the matching differences are all smaller than the matching differences and are within the difference gradient range; wherein the adjustment analysis at least includes adjusting any one of the color feature and the contour feature.

[0010] Furthermore, the color characteristics are adjusted, including: Obtain the area where the matching difference corresponding to the color feature is greater than the difference gradient range, replace the CMYK channel with the CMYK+RGB channel, establish seven monochrome channels, and perform color adjustment; use the original color value as the X-axis and the adjusted color value as the Y-axis to draw the mapping change curve of any monochrome channel, and perform curve fitting on all the mapping change curves on the seven monochrome channels. Use the points where the derivative value on the fitting curve is 0 as the dividing points to divide the color value into various monotonic intervals; Obtain the range under each monotonic interval and obtain the first range mean; for any monotonic interval, extract the color value of the corresponding monochrome channel under the monotonic interval, calculate the second range mean under the corresponding monochrome channel, and select the monochrome channel corresponding to the maximum range mean as the dominant channel; When the second range mean is greater than or equal to the first range mean, the mapping change curve corresponding to the dominant channel is compressed; otherwise, the mapping change curve corresponding to the dominant channel is stretched and its complementary color channel is adjusted synchronously.

[0011] Furthermore, the contour features are adjusted, including: Based on the contour features, the areas where the contour is missing are extracted and filled with lines; Establish a rectangular coordinate system, with the ratio of the length of the patch line to the angle of the patch line as the X-axis and the product of the length of the patch line and the angle of the patch line as the Y-axis, and draw a contour adjustment curve; at the same time, draw a standard contour adjustment curve in the coordinate system, and divide it into the first zone, the second zone, and the third zone based on the standard contour curve; Compare and analyze the contour adjustment curve with the standard contour curve to obtain the situation where the contour adjustment curve is above the standard contour curve: In the first zone, give priority to adjusting the fill line angle; In the second zone, balance and adjust the length and angle of the patch line; In the third zone, priority is given to adjusting the length of the fill line; The first partition is smaller than the second partition and smaller than the third partition.

[0012] Furthermore, obtaining an image quality detection result according to the first pattern and the second pattern includes: preprocessing the first pattern and the second pattern, wherein the preprocessing at least includes geometric alignment, grayscale conversion, and local window division; The SSIM technology is used to calculate the SSIM values ​​of the first pattern and the second pattern window by window, and a difference map is obtained based on the SSIM values. The connected domain of the difference map is analyzed to obtain the similarity percentage and the number of different areas, which are used as the image quality detection results.

[0013] In a second aspect, the present application provides an image quality detection system for a digital printing machine, the system being applied to a digital printer, wherein the digital printer includes at least an inkjet mechanism, a drafting mechanism, and an image output mechanism, the system comprising: a state determination module configured to determine first state information and second state information; wherein the first state information is obtained by prediction using an initial prediction model and includes at least an ink transfer rate, and the second state information is determined by state estimation based on an inkjet pulse signal output by the inkjet mechanism and includes at least a kinetic energy evaluation coefficient; Judgment training module: compares the difference measurement value of the first state information and the second state information, and imports the difference measurement value into the preset difference threshold for judgment: If the difference metric value is greater than or equal to the difference threshold, the change information is sent to the inkjet mechanism, and the initial prediction model is updated based on the change information to obtain a quality prediction model; If the difference metric value is less than the difference threshold, a maintenance message is sent to the inkjet mechanism, and the initial prediction model is used as the quality prediction model; Image output module: used to connect the quality prediction model to the drafting mechanism, scan and input the first pattern and analyze the first pattern area, restore the second pattern, and obtain the image quality detection result based on the first pattern and the second pattern.

[0014] (3) Beneficial effects The present invention provides a method and system for detecting image quality of a digital printing machine, which has the following beneficial effects: 1. This invention establishes an initial prediction model and uses a BP neural network to establish a relationship model between printing pressure, squeegee speed, ink viscosity, squeegee angle, and ink transfer rate, thereby greatly improving the product development rate and automated production of printing machines, and fundamentally improving the quality of printing machine images. In this process, a particle swarm algorithm is used to optimize the weights and thresholds in the prediction model training process. 2. The present invention analyzes the ejection direction and direction of ink droplets through state estimation, combines environmental parameters to obtain a kinetic energy evaluation coefficient, and outputs second state information. The initial prediction model is trained in combination with the first state information to obtain an updated second prediction model. By obtaining the change amount information, difference measurement value and threshold comparison analysis between the two, the initial prediction model is updated to a third prediction model, thereby achieving real-time correction of the prediction model of the inkjet mechanism, improving the training speed of the prediction model, reducing the error rate, and realizing intelligent control; 3. The present invention conducts targeted analysis of color features and contour features. In this process, by replacing the CMYK channel with the CMYK+RGB channel, the color space of the printing machine is effectively expanded. The mapping change curve of any single-color channel is drawn, the dominant channel is adjusted, and the single-color channel ink is supplied. Without changing the tracing mechanism, the color reproduction is improved, the dependence on ink is reduced, and the precise control of the ink amount is achieved. By drawing the contour adjustment curve, the contour line of the printing machine can be quickly identified and targetedly optimized in different partitions, which can effectively improve the quality and efficiency of line filling. 4. The present invention uses the inkjet mechanism, the drafting mechanism and the image output mechanism to intelligently control the digital printing machine, thereby improving the image quality output by the digital printing machine. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 1 is a schematic diagram showing the steps of a method for detecting image quality of a digital printing machine according to an exemplary embodiment; Figure 2 is a schematic diagram of a process of optimizing a BP neural network using a particle swarm algorithm according to an exemplary embodiment; Figure 3 is a device block diagram of a digital printing machine according to an exemplary embodiment; Figure 4 The figure is a module diagram of a digital printing machine image quality detection system according to an exemplary embodiment. DETAILED DESCRIPTION

[0016] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] Example 1: The embodiment of the present invention provides a method for detecting image quality of a digital printing machine; Figure 1 1 is a schematic diagram showing the steps of a method for detecting image quality of a digital printing machine according to an exemplary embodiment; Figure 2 is a schematic diagram of a process of optimizing a BP neural network using a particle swarm algorithm according to an exemplary embodiment; Figure 3 is a block diagram of a digital printing machine according to an exemplary embodiment; Figures 1 to 3 The method is applied to a digital printer, and the digital printer at least includes an inkjet mechanism, a drafting mechanism and an image output mechanism, and includes the following steps: S1. Determine first state information and second state information; wherein the first state information is obtained by prediction using a preset initial prediction model and includes at least an ink transfer rate; and the second state information is determined by state estimation of an inkjet pulse signal output by an inkjet mechanism and includes at least an ink droplet kinetic energy evaluation coefficient. The preset initial prediction model is constructed based on a BP neural network, and the initial prediction model includes optimizing the weights and thresholds of the BP neural network using a particle swarm algorithm. The input parameters of the initial prediction model are printing process parameters, and the output parameter is the ink transfer rate. The printing process parameters include but are not limited to scraper pressure, scraper speed, scraper angle, and ink viscosity, as well as the printing waiting time and ink supply mode. The following is an explanation of the design terms: Squeegee pressure: This refers to the vertical force applied by the squeegee to the substrate or screen during the printing process, which directly affects the efficiency and uniformity of ink penetration through the pores of the screen. A value that is too high may cause excessive ink penetration (blooming), substrate deformation, or increased squeegee / screen wear. A value that is too low may result in insufficient ink transfer, resulting in missing patterns or insufficient color saturation. Scraper speed: refers to the speed at which the scraper moves during the printing process, which determines the ink coating efficiency and dynamic shear force. If the speed is too fast, the ink will not fully penetrate, resulting in blurred edges or broken lines of the pattern. If the speed is too slow, the ink will accumulate, which may cause tailing or local over-thickness. Scraper angle: refers to the tilt angle between the scraper and the substrate or stencil (usually 45° to 75°), which affects the shear force and transfer uniformity of the ink. A large angle (e.g., close to vertical) is suitable for high-viscosity inks or scenes requiring thin coatings. A small angle (e.g., close to horizontal) is suitable for low-viscosity inks or 3D printing requiring thick coatings. Ink viscosity: refers to the quantitative index of ink fluidity, reflecting the deformation resistance of ink under shear force. When the viscosity is high, the fluidity is poor, and higher scraper pressure or slower speed is required to ensure transfer, which is suitable for three-dimensional effects or high covering power requirements. When the viscosity is low, the fluidity is good and it is easy to penetrate but may cause blooming, which is suitable for gradient or transparent effects. The scraper pressure is monitored in real time by a pressure sensor, and the ink viscosity is monitored in real time by a viscosity sensor. The above sensors are not shown in the figure and are adaptively installed at relevant positions on the digital printing machine. Printing waiting time: During the printing process of the digital textile printing machine, the media may be replaced (large spliced ​​images), the ink may be replaced, or different types of ink may be added, resulting in long pauses or too frequent pauses. The above operations may cause discontinuous images and color differences in the image output; Ink supply mode: Piezoelectric mode is usually selected. The ink supply pipeline is resistant to high temperatures and has good sealing properties to prevent the equipment (or mechanism) from leaking ink during operation or being affected by the working temperature, which may reduce the solubility of the various phases in the ink and cause separation, change the ink viscosity, and cause mold to form in the ink supply pipeline and block the ink channel. Among them, the BP neural network is a multi-layer feedforward neural network, the main features of which are signal forward transmission and error reverse transmission; in this embodiment, the initial prediction model is designed to have three layers, including an input layer, a hidden layer, and an output layer, and the transfer functions of the hidden layer and the output layer are both log-sigmoid functions to determine the neural network topology; assuming that the input printing process parameters only include scraper pressure, scraper speed, scraper angle, and ink viscosity, the input nodes are: n = 4, and the output node is: m = 1; assuming that according to the empirical formula: Determine the number of hidden layer neurons l to be 10 (force the adjustment constant a = 7.5, and round up the number of hidden layer neurons), then the designed BP neural network structure is: 4-10-1; there are 4*10+10*1=50 weights in total, and 10+1=11 thresholds; Based on the BP neural network structure, the particle swarm algorithm is used to initialize the population. The particle dimension is: 50+11=61 to determine the particle position. Each particle contains all the weights and thresholds in a network, and the initial weights and thresholds of the network are randomly selected in [-0.5, 0.5]. The optimal individual is obtained by updating the particle position and velocity, that is, the neural network weight and threshold corresponding to the minimum fitness value. The optimal individual obtained by the particle swarm algorithm is assigned to the weights and thresholds of the BP neural network to complete the training of the BP neural network. Among them, the current individual is used to generate the prediction result, the absolute value of the prediction error is obtained based on the prediction result, and the sum of the absolute values ​​of the prediction errors is used as the individual fitness value. By first establishing an initial prediction model for the digital printing machine, it is actually an indirect prediction of the output image quality of the digital printing machine, which can be attributed to the mathematical problem of fitting non-information functions. In actual production, the 3D additive printing process is the process in which ink adheres to the fabric surface through the gaps in the screen. The amount of ink adhered directly determines the quality of the product. Generally, the more ink adhered, the higher the image quality of the printed product. Therefore, the ink transfer rate is used as an indicator to evaluate the quality of the printed image. That is, the higher the ink transfer rate, the better the image quality of the digital printing machine product. Then, using the BP neural network to establish a relationship model between printing pressure, scraping speed, ink viscosity, scraping angle and ink transfer rate, it can greatly improve the product development speed and the automated production of the 3D additive printing process, and fundamentally improve the quality of the printing machine image. The second state information is determined by estimating the state of the inkjet pulse signal output by the inkjet mechanism, including: When receiving the inkjet pulse signal, the ink drop ejection direction and ejection speed of the inkjet mechanism are immediately obtained; The ejection direction is decomposed, with the nozzle axis as the baseline, and the angle between the ejection direction of the ink droplet and the baseline is marked as the deflection angle. The deflection angle is combined with the ejection velocity to set the formula to output the kinetic energy evaluation coefficient of the ink droplet: energy = ε * power * cos θ. In the formula, energy represents the kinetic energy evaluation coefficient, power represents the ejection velocity, θ represents the deflection angle, and ε represents the conversion coefficient, with ε>0. This formula is preset through a large amount of data measurement. Formula explanation: When the deflection angle θ is smaller, the value of cosθ is larger, which means the axis direction is more concentrated and the jet direction is more accurate. When the jet speed is higher, the kinetic energy evaluation coefficient is higher, the ink jet is more accurate, the ink adhesion is greater, and the corresponding printed product image quality is higher. In 3D digital printing machines, in most cases the jet direction is vertically downward, and the nozzle axis is the vertical axis. When the jet direction is not vertically downward, the nozzle axis is the extension line of the nozzle direction. When the deflection angle is greater than the preset deflection threshold, the environmental parameters at this time are collected, and machine learning is used to establish a mapping relationship between the environmental parameters and the angle compensation amount: θ comp =f(T, RH, air, P)+ξ to correct the kinetic energy evaluation coefficient; where θ comp Indicates the angle compensation amount. Environmental parameters include temperature, humidity, airflow, and air pressure. Applications are marked with T, RH, air, and P, respectively. ξ represents the deviation constant, and ξ>0. The preset deflection threshold is determined based on historical statistics, by collecting cases where the deflection angle is less than the preset deflection threshold, and calculating the average value and standard deviation thereof, and taking the average value and 2 times the standard deviation as the preset deflection threshold; By setting the angle compensation amount, the output angle is dynamically adjusted to reduce tracking errors, which is beneficial to improving the accuracy of output image quality.

[0018] S2. Compare the difference measurement value of the first state information and the second state information, and import the difference measurement value into a preset difference threshold for determination: If the difference metric value is greater than or equal to the difference threshold, a change message is sent to the inkjet mechanism, and the initial prediction model is trained and updated according to the instruction message to obtain a quality prediction model; If the difference measure value is less than the difference threshold, a maintenance message is sent to the inkjet mechanism, and the initial prediction model is used as the quality prediction model. The maintenance message is used to indicate that the prediction result of the initial prediction model is accurate and no training update is required. The first state information may include relevant information described as ink transfer rate, and the second state information may include relevant information described as kinetic energy evaluation coefficient; Dissimilarity metric: quantifies the degree of difference between the first state information and the second state information. A larger value indicates a more significant difference. In addition, the dissimilarity metric can be measured using the Euclidean distance. Assuming that only the ink transfer rate and kinetic energy evaluation coefficient are considered: first, the ink transfer rate and kinetic energy evaluation coefficient are dimensionless; then, the difference between the processed ink transfer rate and kinetic energy evaluation coefficient is calculated using Euclidean distance; finally, the difference is used as the difference measure; If the difference measure between the first state information and the second state information is greater than or equal to the difference threshold, it means that the difference measure between the first state information obtained by the prediction model and the actually calculated second state information is large, that is, when the deviation between the two is large, the initial prediction model can be updated to avoid affecting the image quality output by the printing machine. The feedback change information can also be used for online training and updating of the prediction model to achieve continuous optimization of the performance of the prediction model. If the difference measure between the first state information and the second state information is less than the difference threshold, it means that the difference measure between the first state information of the prediction model and the actually calculated second state information is small, that is, the deviation between the two is small, and the initial prediction model may not be updated; The value of the difference threshold is not limited and is determined by the statistical distribution characteristics of historical data, including at least one of the following statistical distribution models: If it is a parametric distribution model: including normal distribution, lognormal distribution, Poisson distribution or Weibull distribution; If it is a non-parametric distribution model: including quantile regression model; The value of the difference threshold is dynamically generated according to the quantile or probability density function of the statistical distribution model; For example: Assuming that historical data conforms to a normal distribution, obtain the mean μ and standard deviation σ of the historical data, and the difference threshold is: μ+k*σ; where k is an adjustable coefficient and satisfies 0<k≤3; The lognormal distribution, Poisson distribution, and Weibull distribution can all be considered as deformations of the normal distribution, and the calculation steps for their thresholds are: Get the quantiles and calculate the threshold based on the quantiles; The lognormal distribution is transformed into a normal distribution by logarithmic transformation, and the quantile z is used a1 Calculate the threshold, then the difference threshold is: exp(μ+z a1 *σ); where 0<z a1 <1; The Poisson distribution is set up by directly using the quantile function or normal approximation, according to the calculation method of the normal distribution. Assuming the form of the normal approximation is: X~N(λ,λ), select the quantile z a2 Perform threshold calculation; the difference threshold is: The form of Poisson distribution is: X~Passion(λ), λ represents the scale parameter, X represents the random variable; where 0<z a2 <1; The Weibull distribution is calculated by calculating the quantile z based on the shape parameter xz and the scale parameter cd. a3 , using the quantile z a3 Calculate the threshold, then the difference threshold is: cd(-ln(1-z a3 )) 1 / xz ; Among them, 0<z a3 <1; Assuming the quantile regression model is used, the difference threshold is: Q p1 ;Q p1 It is the p1th percentile of the historical data and satisfies 0<p1<100. The specific calculation process is not described in detail. Training the initial prediction model to be updated based on the second state information to obtain a second prediction model; wherein the initial value of the initial prediction model to be updated is the knowledge code of the initial prediction model; updating the initial prediction model to be updated to the second prediction model; Specifically, the initial value of the prediction model to be updated is the knowledge encoding of the initial prediction model. The prediction model to be updated is the prediction model currently deployed in the system, that is, the model to be used for calculation. The updated prediction model refers to the one obtained by correcting and training the initial prediction model through state estimation. Assuming that the prediction model to be updated is the initial prediction model, the updated prediction model obtained at this time is the second prediction model. Knowledge encoding includes but is not limited to weights, feature representations, or rules; Leverage the knowledge encoding of an existing model (initial prediction model) to accelerate the training of a new model (secondary prediction model) or improve its generalization capabilities. Updating the initial prediction model to the secondary prediction model typically involves knowledge transfer and iterative model upgrades. The general implementation steps are: Knowledge extraction: Extract the knowledge encoding of the initial prediction model; for example, directly save the weight file, derive the rule / feature importance, and generate soft labels. Configure the new model: directly load the old knowledge encoding (e.g. weights), then align the reusable layers and add the adaptation layers; Optimize training in stages: Add several blank layers to the training layer to adjust the training, for example: freeze the old layers, train only the new layers, gradually unfreeze, and jointly train with soft labels and true labels; Performing secondary training on the initial prediction model to be updated based on the change information and the second state information to obtain a third prediction model; specifically, performing secondary training on the initial prediction model to be updated based on the first change information and the second state information, the steps of which are similar to the above steps and are not described in detail here; Based on the initial prediction model and the second prediction model, determining that the amount of change between the second prediction model and the initial prediction model is greater than or equal to a change threshold, sending first change information to the inkjet mechanism, and the first change information is used to indicate information of the second prediction model; wherein the information of the second prediction model includes: the amount of change between the second prediction model and the initial prediction model; Specifically, the first change information includes information such as model parameters, model structure, cascade relationship, and activation function of the second prediction model, and also includes information on the amount of change between information such as model parameters, model structure, cascade relationship, and activation function of the initial prediction model; in addition, the second prediction model can be determined based on the first change information, and the first change information can also include computer code describing the second prediction model. The system can compile the computer code to obtain the second prediction model; when the amount of change between the second prediction model and the initial prediction model is less than a change threshold, that is, when the amount of change between the updated prediction model and the prediction model last indicated to the digital printing machine is less than the change threshold, the updated prediction model is not indicated to the inkjet mechanism, thereby avoiding frequent indication of the updated prediction model to the inkjet mechanism and improving system efficiency; Based on the initial prediction model and the third prediction model, if it is determined that the amount of change between the initial prediction model and the third prediction model is greater than or equal to a change threshold, second change information is sent to the inkjet mechanism, and the second change information is used to indicate information of the third prediction model; wherein the information of the third prediction model includes: the amount of change between the third prediction model and the initial prediction model; It should be noted that the value of the change threshold is similar to that of the difference threshold and the difference threshold, and will not be described in detail here; The following is an explanation of the nouns involved: Variation information: This refers to the degree of difference between different models in terms of parameters, prediction behavior, or feature importance. Quantifying and analyzing this difference is crucial for scenarios such as model optimization, version iteration, and system robustness evaluation. In terms of parameter space changes: compare the numerical differences of model parameters (such as linear regression coefficients, neural network weight matrices) and the changes in the number of parameters or connection methods caused by different model architectures (such as the difference in the number of layers between ResNet and VGG); for example, cosine similarity can be used to measure the similarity between the weight matrices or coefficients corresponding to two models; In terms of predictive behavior: compare the distribution differences of the prediction results of the models for the same input data; for example, the KL divergence can be used to measure the distribution differences of the output probabilities of two models; In terms of feature importance: compare the model's dependence on input features; use the Rank-BiasedOverlap (RBO) algorithm to measure the similarity of two ranked lists; The above processes are all routine data processing and will not be described in detail here; By obtaining the corresponding degree of difference in the three aspects of parameter space change, prediction behavior or feature importance, and after normalization, the corresponding difference values ​​are obtained respectively, and weighted summation is performed to obtain the variation information. By quantifying the variation information between the two prediction models, the model iteration risk can be systematically evaluated, the integration strategy can be optimized, and the system interpretability and robustness can be improved. In this embodiment, the specific method for updating the initial prediction model to the second prediction model and the initial prediction model to the third prediction model is not limited, for example, it can be a back propagation gradient update algorithm, etc. When the inkjet mechanism receives any one of the maintenance information, the first change information or the second change information, the corresponding initial prediction model, the second prediction model or the third prediction model is selected as the prediction model for real-time training; That is, when only the maintenance information is received, the initial prediction model is selected as the prediction model for real-time training; when only the first change information is received, the second prediction model is selected as the prediction model for real-time training; when only the second change information is received, the third prediction model is selected as the prediction model for real-time training; If the inkjet mechanism receives at least two of the maintenance information, the first change information, or the second change information, the inkjet mechanism includes: When the maintenance information and the first change information are received, the second prediction model is selected as the prediction model for real-time training; when the maintenance information and the second change information are received, the third prediction model is selected as the prediction model for real-time training; when the second change information and the third change information are received, the third prediction model is selected as the prediction model for implementation training; when the maintenance information, the first change information, and the second change information are received, the third prediction model is selected as the prediction model for real-time training; In this embodiment, the change information may also be used to indicate any of the following information: The format information of the prediction model input and output, for example, indicating the input dimension and output dimension of the corresponding prediction model; Evaluation function of the prediction model: a function used to calculate the difference measure between the first state information and the second state information; for example, the Euclidean distance function; a function used to calculate the predicted behavior, for example, the KL divergence related function; a function used to calculate the feature importance, for example, the JS divergence, etc. Threshold information: including difference threshold and change threshold; Feedback information: feedback on whether the prediction model should be trained and updated; for example, if the difference measure between the first state information and the second state information is less than the difference threshold, it can be said that the result output by the initial prediction model is accurate and the initial prediction model does not need to be trained and updated; otherwise, the feedback needs to be updated; The inkjet mechanism of a digital printing machine is the key to digital printing, and the normal and efficient operation of the equipment (mechanism) is the key to maintaining the output and quantity of printing; the initial prediction model is pre-configured for the inkjet mechanism, and the initial prediction model is the same as the prediction model in the inkjet mechanism; by configuring the same prediction model in the inkjet mechanism and the system, it is possible to determine whether the prediction model configured in the inkjet mechanism is accurate based on the kinetic energy evaluation coefficient actually measured by the system. When it is inaccurate, the initial prediction model can be trained and updated through the second state information and change information, and the prediction model of the inkjet mechanism can be corrected in real time to obtain a quality prediction model.

[0019] S3, connecting the quality prediction model to the drafting mechanism, scanning the first pattern and analyzing the first pattern area, restoring the second pattern, and obtaining an image quality detection result based on the first pattern and the second pattern; The steps of regional analysis include: Dividing the first pattern into a plurality of regions, extracting a plurality of pixel points in each region and their color features and contour features, performing contour matching and color matching, obtaining corresponding matching differences, and setting a difference gradient range of ±wc%. When the matching difference is within the difference gradient range, sending an inkjet pulse signal to perform inkjet operation; otherwise, performing adjustment analysis until the matching difference is less than the matching difference and is within the difference gradient range; wherein the adjustment analysis includes adjusting at least one of the color feature and the contour feature; Specifically, digital inkjet printing technology is to input the pattern into the computer in digital form, edit and process it through the computer printing color separation and tracing mechanism, and then the computer controls the micro-piezoelectric nozzle to split the ink fluid into uniform small ink droplets and directly spray them onto the substrate to form pixels. After color mixing, it is restored to the original pattern. Therefore, the constant pressure and stable supply of ink is an important factor affecting the quality of digital inkjet printing patterns. Through steps S1 and S2, the quality analysis is performed with the ink transfer rate and kinetic energy evaluation coefficient as analysis factors (for example: the larger the ink transfer rate value, the better the quality of the obtained printing pattern; the larger the kinetic energy evaluation coefficient value, the better the quality of the obtained printing pattern); in this step, the input pattern is directly analyzed; the steps of adjusting the analysis include: Adjust color features: Color features include color values; Obtain areas where the matching difference corresponding to the color feature is greater than the difference gradient range, replace the CMYK channel with the CMYK+RGB channel, establish seven monochrome channels, and perform color adjustment; use the original color value as the X-axis and the adjusted color value as the Y-axis to draw the mapping change curve of any monochrome channel, and perform curve fitting on all the mapping change curves on the seven monochrome channels. Use the points on the fitting curve with a derivative value of 0 as the dividing points to divide the color values ​​into monotonic intervals; The steps of curve fitting include: Ensure that the data lengths of the seven monochrome channels are consistent (for example, time series alignment and image pixel correspondence); Standardize the data of each monochrome channel; Extract the principal components of the multi-channel color values ​​corresponding to the corresponding area, and use the first principal component as the fitting curve: Y = PC1(Red(t), Green(t), Blue(t), Cyan(t), Magenta(t), Yellow(t), Black(t)); where Y is the fitting curve, PC1 is the first principal component, Red represents the red monochrome channel, Green represents the green monochrome channel, Blue represents the blue monochrome channel, Cyan represents the cyan monochrome channel, Magenta represents the magenta monochrome channel, Yellow represents the yellow monochrome channel, and Black represents the black monochrome channel, and t represents the time; the fitting curve can automatically capture the maximum variance direction of the multi-channel to facilitate the acquisition of the range; Obtain the range under each monotonic interval and obtain the mean of the first range; for any monotonic interval, extract the color value of the corresponding monochrome channel under the monotonic interval, calculate the second range mean under the corresponding monochrome channel, and select the monochrome channel corresponding to the largest range mean as the dominant channel; where the range represents the difference between the maximum and minimum values ​​of the data under any monotonic interval, which is used to indicate the degree of data dispersion; the range mean reflects the average intensity of the dynamic range change of each monotonic interval after color adjustment; when the second range mean is greater than or equal to the first range mean, it indicates that the channel may be overexposed or color overflow, and the mapping change curve corresponding to the dominant channel is compressed; The compressing of the mapping change curve corresponding to the dominant channel specifically reduces the slope of the mapping change curve corresponding to the dominant channel, that is, reduces the contrast or intensity of the channel; For example, assuming that the dominant channel is the red channel, the shape of the mapping change curve corresponding to the red channel is smoothed, such as forming a flat or inverse S-shaped low-slope curve; Select compression area: Select the channel area where the second range mean is greater than or equal to the first range mean, and reduce the slope. You can also adjust the complementary color channel later. The specific implementation is similar to the "Stretch the dominant channel mapping change curve and synchronously adjust its complementary color channel" step. The steps for "synchronously adjusting its complementary color channel" will be explained below and will not be explained here. On the contrary, the mapping change curve corresponding to the dominant channel is stretched, and its complementary color channel is adjusted synchronously to avoid color cast; The stretching of the mapping change curve corresponding to the dominant channel specifically increases the slope of the mapping change curve corresponding to the dominant channel, that is, enhances the contrast or intensity of the channel; For example: Based on the current color distribution, the dominant channel is determined. Assuming the dominant channel is the red channel (Red), it is stretched into a steep S-shaped or linear high-slope curve. The complementary color channel is defined as the cyan channel (Cyan), and the complementary channel is adjusted in the opposite direction to the dominant channel to maintain color balance, that is, the complementary color channel is compressed; the other non-complementary color channels (green channel (Green), blue channel (Blue), magenta channel (Magenta), yellow channel (Yellow), black channel (Black)) remain unchanged, or are fine-tuned as needed; for example: if the red channel (Red) and cyan channel (Cyan) are adjusted, the green channel (Green) and blue channel (Blue) can remain unchanged, but if the overall brightening is required, the black channel (Black) can be fine-tuned simultaneously; at the same time, the adjusted curve is smoothed and filtered (such as Gaussian filtering) to avoid color level breaks; the pseudo code is as follows: The use of CMYK 4-color separation (such as cyan, magenta, yellow, black, etc.) can cover most of the color gamut visible to the human eye, but it is insufficient in some special tones and color transitions. By rationally adding additional primary colors (such as orange, green, purple), the color space of the printing machine can be effectively expanded. Without changing the drafting mechanism, the CMYK-RGB 7-color separation can improve color reproduction and reduce dependence on ink. By analyzing the mapping change curve, the monochrome channel ink is fed and the ink amount is precisely controlled. Adjust the profile features: Based on the contour features, the areas where the contour is missing are extracted and filled with lines; Establish a rectangular coordinate system, with the ratio of the length of the patch line to the angle of the patch line as the X-axis and the product of the length of the patch line and the angle of the patch line as the Y-axis, and draw a contour adjustment curve; at the same time, draw a standard contour adjustment curve in the coordinate system, and divide it into the first zone, the second zone, and the third zone based on the standard contour curve; Compare and analyze the contour adjustment curve with the standard contour curve to determine if the contour adjustment curve is above the standard contour curve. In the first partition, prioritize adjusting the fill line angle. For example, the fill line angle determines the direction and amplitude of the printhead rotation to ensure that the ink droplet ejection direction is aligned with the original pattern to avoid misalignment. In the second zone, balance adjustment is performed for the length and angle of the patch line. For example, the balance adjustment problem is similar to a car driving on a path. The patch line length is similar to the distance the car needs to move (the longer the length, the farther the nozzle moves), and the patch line angle is similar to the steering wheel angle when the car turns (the larger the angle, the more rapid the change in path direction). The goal of balance adjustment is to ensure that the car reaches the specified location (completes the patch line length) while avoiding sharp turns that may cause loss of control (excessive patch line angle will cause edge burrs) or too slow (too small angle will cause inefficiency). In the third zone, the priority is to adjust the length of the patch line. For example, the longer the patch line is, the farther the print head needs to move, and the movement speed needs to be adjusted to avoid smearing. At the same time, the ink drop frequency needs to be increased to ensure continuous coverage. The inkjet mechanism is provided with a high-precision servo motor or linear motor to ensure the accuracy of the movement and rotation of the nozzle; in this embodiment, it is adaptively installed and the corresponding diagram is not drawn; The first partition is smaller than the second partition, which is smaller than the third partition. The first, second, and third partitions are divided based on quantiles. For example, if a three-digit number is used for partitioning and the partitioning is equal, the quantiles 33.33% and 66.67% are used. Using quantiles for partitioning ensures that the samples in each partition area are controllable, is insensitive to outliers, and dynamically adapts to the data distribution (e.g., directly reflects the data distribution characteristics), making the partitioning results stable. The following is an explanation of the nouns involved: Filling line length: The length of the line segment used to fill in the gaps, calculated using Euclidean distance. The length of the filling line is controlled by the printhead movement speed and ink droplet ejection frequency. Filling Angle: Consider the tangent direction of the contour at the fracture endpoint corresponding to the vacant area, so that the filling angle is naturally connected with the surrounding contour; the filling angle is controlled by the rotation angle of the printhead and the direction of ink droplet ejection; Patch length / patch angle: Indicates the length of the patch corresponding to a unit angle, reflecting the sensitivity of angle changes to length changes. When the value is too high, the length changes greatly under the same angle change; when the value is too low, the length is insensitive to the angle. Patch line length * patch line angle: indicates the "intensity of influence" of patch line length on the contour. The higher the value, the more significant the patch line's influence on the overall shape; the lower the value, the less influence the patch line has. When the contour adjustment curve is above the standard contour curve, the fill line length may be too long, the fill line angle may be too large, or the fill line length and fill line angle may deviate from the ideal value. By executing corresponding strategies in different partitions, the printhead movement speed, printhead rotation angle, and ink droplet ejection direction can be adjusted to further achieve precise control of ink supply. At the same time, the second pattern is restored and compared with the first pattern to determine whether the two patterns are consistent. The comparison method is not limited here. preprocessing the first pattern and the second pattern, wherein the preprocessing at least includes geometric alignment, grayscale conversion, and local window division; Geometric alignment: The two patterns to be detected (the first pattern and the second pattern) are resized to ensure that their height and width are consistent. Grayscale conversion: The first pattern and the second pattern are converted into grayscale images and noise interference is eliminated through Gaussian filtering. Local window division: The processed image is divided into overlapping local windows (xm, yn) (for example, the window size is 11×11 pixels). Calculate the structural similarity index (SSIM technique) for each local window and obtain a difference map. The value of each pixel in the difference map is: D(xm,yn) = 1-SSIM(xm,yn); where SSIM(xm,yn) indicates whether the first pattern and the second pattern are consistent within the local window (xm,yn), and the value range of SSIM(xm,yn) is [0,1], where 1 indicates that the two patterns are completely consistent within the local window (xm,yn), and 0 indicates that the two patterns are completely inconsistent within the local window (xm,yn). That is, the larger the D(xm,yn) value, the more significant the difference, and the smaller the D(xm,yn) value, the less significant the difference. Take the arithmetic average of the SSIM values ​​of all local windows, multiply the average result by the percentage, and mark it as the similarity percentage; When the difference map is greater than the preset difference threshold, it is marked as a difference area. Figure 2 The binary image is processed by marking the difference areas as white, and the number of independent white areas in the binary image is counted, which is the number of difference areas. The similarity percentage and the number of different areas are used to represent the image quality detection results; In addition, the image quality detection result can be obtained by comparing the differences in pixels, color features, and contour features of the two patterns and performing numerical normalization. The quantification of the differences can be based on Manhattan distance calculation. The calculation method is not limited, and the specific steps are conventional processing and will not be described in detail.

[0020] Example 2: The embodiment of the present invention provides an image quality detection system for a digital printing machine; Figure 4 is a module diagram of a digital printing machine image quality detection system according to an exemplary embodiment; Figure 4The system is applied to a digital printer, and the digital printer includes at least an inkjet mechanism, a drafting mechanism, and an image output mechanism, and includes: a state determination module, a judgment training module, and an image output module, and the state determination module, the judgment training module, and the image output module are communicatively connected to each other; the state determination module is used to determine first state information and second state information; wherein the first state information is obtained by prediction using an initial prediction model and includes at least an ink transfer rate, and the second state information is determined by state estimation based on an inkjet pulse signal output by the inkjet mechanism and includes at least a kinetic energy evaluation coefficient; Judgment training module: compares the difference measurement value of the first state information and the second state information, and imports the difference measurement value into the preset difference threshold for judgment: If the difference metric value is greater than or equal to the difference threshold, the change information is sent to the inkjet mechanism, and the initial prediction model is updated based on the change information to obtain a quality prediction model; If the difference metric value is less than the difference threshold, a maintenance message is sent to the inkjet mechanism, and the initial prediction model is used as the quality prediction model; Image output module: used to connect the quality prediction model to the drafting mechanism, scan and input the first pattern and analyze the first pattern area, restore the second pattern, and obtain the image quality detection result based on the first pattern and the second pattern.

[0021] In the application, the several formulas involved are all calculated by taking their numerical values ​​after removing the dimensions, and the formula is a formula of the most recent real situation obtained by collecting a large amount of data and performing software simulation. The formula is set by technical personnel in this field according to actual conditions.

[0022] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0023] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0024] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for detecting image quality of a digital printing machine, the method being applied to a digital printer, wherein the digital printer comprises at least an inkjet mechanism, a drafting mechanism, and an image output mechanism, and is characterized in that: The method comprises: Determining first state information and second state information; wherein the first state information is obtained by prediction using a preset initial prediction model and includes at least an ink transfer rate, and the second state information is determined by state estimation of an inkjet pulse signal output by an inkjet mechanism and includes at least an ink droplet kinetic energy evaluation coefficient; Compare the difference measurement values ​​of the first state information and the second state information, and import the difference measurement values ​​into a preset difference threshold for determination: If the difference metric value is greater than or equal to the difference threshold, sending change information to the inkjet mechanism, and training and updating the initial prediction model based on the change information to obtain a quality prediction model; If the difference metric value is less than the difference threshold, sending a maintenance message to the inkjet mechanism and using the initial prediction model as a quality prediction model; The quality prediction model is connected to the drafting mechanism, a first pattern is scanned and input, and the first pattern area is analyzed to restore the second pattern, and an image quality detection result is obtained based on the first pattern and the second pattern.

2. A digital printing machine image quality detection method according to claim 1, characterized in that: The initial prediction model is constructed based on a BP neural network, and the initial prediction model includes optimizing the weights and thresholds of the BP neural network using a particle swarm algorithm; The input parameters of the initial prediction model are printing process parameters, and the output parameters are ink transfer rate; wherein the printing process parameters include at least scraper pressure, scraping speed, scraper angle and ink viscosity.

3. The method for detecting image quality of a digital printing machine according to claim 1, characterized in that: The second state information is determined by performing state estimation on the inkjet pulse signal output by the inkjet mechanism, including: When receiving the inkjet pulse signal, the ink drop ejection direction and ejection speed of the inkjet mechanism are obtained; The ejection direction is decomposed, and the nozzle axis is used as the baseline. The angle between the ejection direction of the ink droplet and the baseline is marked as the deflection inclination angle; the deflection inclination angle is combined with the ejection speed to obtain the kinetic energy evaluation coefficient of the ink droplet.

4. A digital printing machine image quality detection method according to claim 3, characterized in that: The step of obtaining the kinetic energy evaluation coefficient of the ink droplet comprises: When the deflection angle is greater than the preset deflection threshold, the environmental parameters at this time are collected, and machine learning is used to establish a mapping relationship between the environmental parameters and the angle compensation amount, and the kinetic energy evaluation coefficient is corrected; among them, the environmental parameters include at least temperature and humidity.

5. The method for detecting image quality of a digital printing machine according to claim 1, wherein: The training and updating of the initial prediction model according to the change information to obtain a quality prediction model includes: The initial prediction model is trained based on the second state information to obtain a second prediction model; wherein the initial value of the initial prediction model is the knowledge encoding of the initial prediction model; The initial prediction model is trained again based on the change information and the second state information to obtain a third prediction model; Determining first variation information of the initial prediction model and the second prediction model based on the initial prediction model and the second prediction model, and sending first change information to the inkjet mechanism if the first variation information is greater than or equal to a preset variation threshold, wherein the first change information is used to indicate information of the second prediction model; wherein the information of the second prediction model includes: variation information between the second prediction model and the initial prediction model; Determining second variation information of the initial prediction model and the third prediction model based on the initial prediction model and the third prediction model, and sending second change information to the inkjet mechanism if the second variation information is greater than or equal to a variation threshold, wherein the second change information is used to indicate information of the third prediction model; wherein the information of the third prediction model includes: variation information between the third prediction model and the initial prediction model; If the inkjet mechanism receives any one of the maintenance information, the first change information or the second change information, the corresponding initial prediction model, the second prediction model or the third prediction model is selected as the prediction model for real-time training, and the prediction model is obtained by training and updating.

6. The method for detecting image quality of a digital printing machine according to claim 1, characterized in that: The analyzing of the first pattern area includes: Divide the first pattern into several areas, extract several pixels in each area and their color features and contour features, perform contour matching and color matching, obtain corresponding matching differences, and set a difference gradient range of ±wc%. When the matching difference is within the difference gradient range, send an inkjet pulse signal to perform inkjet operation; Otherwise, the analysis is adjusted until the matching differences are all smaller than the matching differences and are within the difference gradient range; wherein the adjustment analysis at least includes adjusting any one of the color feature and the contour feature.

7. A digital printing machine image quality detection method according to claim 6, characterized in that: The adjusting of the color characteristics includes: Obtain the area where the matching difference corresponding to the color feature is greater than the difference gradient range, replace the CMYK channel with the CMYK+RGB channel, establish seven monochrome channels, and perform color adjustment; use the original color value as the X-axis and the adjusted color value as the Y-axis to draw the mapping change curve of any monochrome channel, and perform curve fitting on all the mapping change curves on the seven monochrome channels. Use the points where the derivative value on the fitting curve is 0 as the dividing points to divide the color value into various monotonic intervals; Obtain the range under each monotonic interval and obtain the first range mean; for any monotonic interval, extract the color value of the corresponding monochrome channel under the monotonic interval, calculate the second range mean under the corresponding monochrome channel, and select the monochrome channel corresponding to the maximum range mean as the dominant channel; When the second range mean is greater than or equal to the first range mean, the mapping change curve corresponding to the dominant channel is compressed; otherwise, the mapping change curve corresponding to the dominant channel is stretched and its complementary color channel is adjusted synchronously.

8. The method for detecting image quality of a digital printing machine according to claim 6, characterized in that: The adjusting of the contour features includes: Based on the contour features, the areas where the contour is missing are extracted and filled with lines; Establish a rectangular coordinate system, with the ratio of the length of the patch line to the angle of the patch line as the X-axis and the product of the length of the patch line and the angle of the patch line as the Y-axis, and draw a contour adjustment curve; at the same time, draw a standard contour adjustment curve in the coordinate system, and divide it into the first zone, the second zone, and the third zone based on the standard contour curve; Compare and analyze the contour adjustment curve with the standard contour curve to obtain a situation where the contour adjustment curve is above the standard contour curve: in the first partition, give priority to adjusting the fill line angle; In the second zone, balance and adjust the length and angle of the patch line; In the third zone, priority is given to adjusting the length of the fill line; The first partition is smaller than the second partition and smaller than the third partition.

9. The method for detecting image quality of a digital printing machine according to claim 1, characterized in that: The obtaining of the image quality detection result according to the first pattern and the second pattern includes: preprocessing the first pattern and the second pattern, wherein the preprocessing at least includes geometric alignment, grayscale conversion, and local window division; The SSIM technology is used to calculate the SSIM values ​​of the first pattern and the second pattern window by window, and a difference map is obtained based on the SSIM values. The connected domain of the difference map is analyzed to obtain the similarity percentage and the number of different areas, which are used as the image quality detection results.

10. A digital printing machine image quality detection system, which is applied to a digital printer, and the digital printer at least includes an inkjet mechanism, a drafting mechanism and an image output mechanism, characterized in that: The system comprises: A state determination module is configured to determine first state information and second state information; wherein the first state information is obtained by prediction using an initial prediction model and includes at least an ink transfer rate, and the second state information is determined by state estimation based on an inkjet pulse signal output by the inkjet mechanism and includes at least a kinetic energy evaluation coefficient; Judgment training module: compares the difference measurement value of the first state information and the second state information, and imports the difference measurement value into a preset difference threshold for judgment: If the difference metric value is greater than or equal to the difference threshold, sending change information to the inkjet mechanism, and training and updating the initial prediction model based on the change information to obtain a quality prediction model; If the difference metric value is less than the difference threshold, sending a maintenance message to the inkjet mechanism and using the initial prediction model as a quality prediction model; Image output module: used to connect the quality prediction model to the drafting mechanism, scan and input the first pattern and analyze the first pattern area, restore the second pattern, and obtain the image quality detection result based on the first pattern and the second pattern.

Images