Edge computing-based inkjet printer real-time image calibration system and method

Abstract

The present application relates to the technical field of image calibration, and discloses a real-time image calibration system and method for a code jet printer based on edge computing, which comprises the following steps: firstly, a composite parameter matrix is constructed, no-load error analysis is completed, a compensation matrix and a dynamic weight distribution function are generated; secondly, a multi-scale gradient feature pyramid is constructed according to the composite parameter matrix, a feature correlation degree heat map is generated by fusing the features, and a model is established after correction to predict the ink diffusion trajectory; finally, a control signal is generated by combining the compensation matrix and the prediction result to control the motion pose of the nozzle; the method realizes accurate control of the code jet process by comprehensively considering multiple factors, significantly improves the precision and quality of the code jet pattern, effectively reduces the waste rate, and improves the production efficiency.

Description

Technical Field

[0001] This invention relates to the field of image calibration technology, and more specifically, to a real-time image calibration system and method for inkjet printers based on edge computing. Background Technology

[0002] In the industrial production sector, inkjet printing technology, as a key means of product identification and information traceability, is widely used in numerous industries such as food, pharmaceuticals, and tobacco. The quality of the inkjet-printed characters directly affects the readability, traceability, and brand image of the product. In the context of cigarette box inkjet printing, the quality of the character edges is crucial to the anti-counterfeiting and traceability effectiveness in the tobacco industry. If the edges of the inkjet-printed characters are blurry, it will cause great difficulties for QR code recognition and other tasks, thereby affecting the entire production process and the market circulation of the product.

[0003] In the prior art, the Chinese patent application with publication number CN119459130A proposes an automatic printhead detection method and device based on machine vision for digital inkjet printers. This method primarily detects and calibrates the inkjet quality of the printhead by acquiring images of the printhead needles and analyzing the inkjet state. However, this prior art focuses solely on detecting the inkjet performance of the printhead itself. It lacks in-depth research and targeted treatment of ink diffusion behavior, which causes blurred character edges during the coding process. It cannot predict the diffusion trajectory based on ink properties (such as viscosity and surface tension), nor can it effectively distinguish between true edges and diffusion artifacts. Therefore, it struggles to accurately compensate for blurred edges of coded characters and cannot provide accurate data support for subsequent image calibration when dealing with the "burr" problem at the edges of coded characters on cigarette boxes.

[0004] Chinese patent CN117119115B discloses a calibration method, device, electronic device, and storage medium based on machine vision. It compensates for the encoding cycle by acquiring marker point images to determine offset data and achieve calibration. However, this prior art does not fully consider the impact of ink diffusion on the edges of the printed characters, cannot predict diffusion based on ink characteristics, and lacks specific measures for sub-pixel level calibration. Therefore, it cannot meet the high requirements for character edge clarity in cigarette box printing and struggles to achieve ideal calibration accuracy when dealing with blurred edges in printed characters.

[0005] In summary, existing inkjet printer calibration technologies are insufficient in dealing with the problem of blurred edges of inkjet characters. They cannot accurately distinguish between real edges and diffusion artifacts, and it is difficult to achieve sub-pixel level calibration accuracy. Therefore, they cannot meet the stringent requirements of the tobacco industry and other industries for the quality of inkjet character edges. Summary of the Invention

[0006] To overcome the aforementioned shortcomings of existing technologies, this invention provides a real-time image calibration system and method for inkjet printers based on edge computing. By constructing a composite parameter matrix that comprehensively considers factors such as ink diffusion, substrate material adsorption, and printhead movement, it provides comprehensive data support for image calibration. Based on this, no-load error analysis is performed to generate a compensation matrix and a dynamic weight allocation function, enabling adaptive adjustment of the compensation strategy. Simultaneously, a multi-scale gradient feature pyramid is constructed, features are fused to generate a heatmap, and a dynamic prediction model for ink diffusion behavior is established. This model can accurately predict the ink diffusion trajectory, and then, combined with the prediction results, a control signal is generated to dynamically control the printhead's motion posture. This series of operations effectively solves the problem of blurred edges in inkjet characters, achieves sub-pixel-level calibration, significantly improves the accuracy and quality of inkjet patterns, greatly increases the recognition rate of QR codes and other inkjet information, enhances the adaptability of inkjet printers to different working conditions, meets the high-quality inkjet printing requirements of various industries, and brings higher efficiency and lower costs to industrial production.

[0007] This invention is applicable to various industrial production scenarios with high requirements for coding quality, such as batch coding in food and beverages, model identification coding in electronic products, and cigarette box coding in the tobacco industry. In these scenarios, the accuracy, clarity, and stability of the coding are crucial; any coding defects may lead to product quality problems or difficulties in information traceability.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A real-time image calibration method for inkjet printers based on edge computing includes:

[0010] Construct a composite parameter matrix, perform no-load error analysis, generate a no-load artifact offset error compensation matrix, establish a nonlinear mapping relationship between each element of the no-load artifact offset error compensation matrix and the composite parameter matrix, and generate a dynamic weight allocation function for the compensation matrix.

[0011] Based on the composite parameter matrix, a multi-scale gradient feature pyramid is constructed, which includes a bottom layer, a middle layer, and a top layer. The bottom, middle, and top features of the multi-scale gradient feature pyramid are fused using tensors to generate a feature correlation heatmap covering different abstraction levels. The feature correlation heatmap is corrected by combining the empty artifact offset error compensation matrix, and a dynamic prediction model for ink diffusion behavior is established to generate real-time diffusion trajectory prediction results.

[0012] Based on the dynamic weight allocation function of the compensation matrix and the real-time diffusion trajectory prediction results, an artifact compensation control signal is generated. This artifact compensation control signal is then coupled to the nozzle drive circuit to perform dynamic feedback control on the nozzle's motion and pose.

[0013] Furthermore, the construction of the composite parameter matrix includes:

[0014] Obtain the surface viscosity coefficient and potential energy gradient values ​​to form a sub-matrix of fluid diffusion field parameters;

[0015] The adsorption decay characteristic curves of ink on different material surfaces were measured, and the adsorption decay time constant and saturation adsorption concentration threshold were extracted to form a sub-matrix of substrate material adsorption characteristic data.

[0016] The nozzle movement process is tracked to obtain the nozzle vibration displacement signal. The spatial curvature change features of the nozzle movement trajectory are extracted from the nozzle vibration displacement signal to form a spatial feature sub-matrix of the nozzle movement trajectory.

[0017] A composite parameter matrix is ​​constructed based on the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic data sub-matrix, and the nozzle motion trajectory spatial characteristic sub-matrix.

[0018] Furthermore, the formation of the fluid diffusion field parameter sub-matrix includes: collecting dynamic contact angle change data of ink droplets on the substrate surface and extracting the surface viscosity coefficient; measuring the spatial difference of pressure distribution inside the ink droplets and calculating the potential energy gradient value; and quantifying the surface viscosity coefficient and potential energy gradient value into the fluid diffusion field parameter sub-matrix.

[0019] Furthermore, the generation of the empty artifact offset error compensation matrix includes:

[0020] Based on the fluid diffusion field parameter sub-matrix, the artifact width distribution data formed during the ink diffusion process is obtained, and the diffusion artifact width distribution spectrum is generated.

[0021] The diffusion artifact width distribution spectrum is used as a priori constraint condition for the printing start stage under no-load conditions. It is loaded into a lightweight reverse generative adversarial topology network in a pre-built edge computing node to obtain the diffusion artifact prior constraint.

[0022] Based on the fluid diffusion field parameter sub-matrix and the substrate material adsorption characteristic data sub-matrix, the relative contribution rate of the adsorption effect and the relative contribution rate of the ink potential energy driving effect are obtained.

[0023] By combining the prior constraints of diffusion artifacts, error analysis is performed on the inkjet printing image under no-load conditions to generate the no-load artifact offset error compensation matrix.

[0024] Furthermore, the relative contribution rate of the adsorption effect and the relative contribution rate of the ink potential energy driving effect are obtained as follows:

[0025] Based on the surface viscosity coefficient and potential energy gradient value in the fluid diffusion field parameter sub-matrix, a mathematical and physical model of the ink droplet diffusion process on the substrate surface is constructed. The finite element analysis method is used to numerically simulate the fluid dynamics characteristics inside the droplet, and a quantitative description function of the ink potential energy driving effect is obtained.

[0026] Based on the adsorption decay time constant and saturated adsorption concentration threshold in the adsorption characteristic data sub-matrix of the substrate material, an adsorption kinetic model between ink droplets and the substrate surface is constructed. Molecular dynamics simulation is used to quantitatively analyze the adsorption process and obtain a quantitative description function of the adsorption effect.

[0027] The quantitative description functions of ink potential energy driving effect and adsorption effect are normalized to obtain the relative contribution rate of adsorption effect and the relative contribution rate of ink potential energy driving effect.

[0028] Furthermore, the step of combining diffusion artifact prior constraints to perform error analysis on the inkjet printing image under no-load conditions and generating an no-load artifact offset error compensation matrix includes:

[0029] By combining diffusion artifact prior constraints, a preset reference pattern is printed using an inkjet printer under no-load conditions, and the actual inkjet image under no-load conditions is collected.

[0030] Feature point matching is performed between the acquired actual inkjet image and the reference pattern to calculate the geometric offset of the inkjet outline;

[0031] The geometric offset is represented by a matrix to generate an empty artifact offset error compensation matrix.

[0032] Furthermore, the calculation of the geometric offset of the inkjet printing profile includes:

[0033] Extract key feature points from the actual inkjet image and the reference pattern, and construct a feature point matching matrix;

[0034] Based on the feature point matching matrix, the geometric offset function of the inkjet printing profile is fitted to obtain the local geometric offset and the global geometric offset.

[0035] The local geometric offset and the global geometric offset are combined to form the geometric offset of the inkjet printing profile.

[0036] Furthermore, the step of representing the geometric offset as a matrix to generate the empty artifact offset error compensation matrix includes:

[0037] Based on the relative contribution rate of the adsorption effect and the relative contribution rate of the ink potential energy driving effect, the local geometric offset and the overall geometric offset are weighted and combined to obtain the comprehensive geometric offset matrix.

[0038] The elements in the comprehensive geometric offset matrix are normalized and discretized to generate the empty artifact offset error compensation matrix.

[0039] Furthermore, the construction of the multi-scale gradient feature pyramid includes:

[0040] Based on the fluid diffusion field parameter sub-matrix, multi-scale gradient invariant feature clusters formed during the diffusion of ink on the substrate surface are extracted, and the bottom layer of the multi-scale gradient feature pyramid is constructed; the multi-scale gradient invariant feature clusters include a 64-dimensional directional gradient histogram.

[0041] Based on the adsorption characteristic data sub-matrix of the substrate material, the micro porosity distribution mapping of the substrate material surface is captured, and the middle layer of the multi-scale gradient feature pyramid is constructed.

[0042] Based on the spatial feature submatrix of the nozzle motion trajectory, the fluctuation characteristics of the second derivative of the nozzle motion trajectory are analyzed, and the top layer of the multi-scale gradient feature pyramid is constructed.

[0043] A real-time image calibration system for inkjet printers based on edge computing is provided to implement the aforementioned real-time image calibration method for inkjet printers based on edge computing. The system includes:

[0044] The compensation matrix generation module is used to construct a composite parameter matrix, perform no-load error analysis, generate a no-load artifact offset error compensation matrix, establish a nonlinear mapping relationship between each element of the no-load artifact offset error compensation matrix and the composite parameter matrix, and generate a dynamic weight allocation function for the compensation matrix.

[0045] Diffusion trajectory prediction module: Based on the composite parameter matrix, a multi-scale gradient feature pyramid is constructed, which includes a bottom layer, a middle layer, and a top layer; the bottom, middle, and top layer features of the multi-scale gradient feature pyramid are fused using tensors to generate a feature correlation heatmap covering different abstraction levels; combined with the empty artifact offset error compensation matrix, the feature correlation heatmap is corrected, and a dynamic prediction model for ink diffusion behavior is established to generate real-time diffusion trajectory prediction results;

[0046] Error compensation module: Based on the dynamic weight allocation function of the compensation matrix and the real-time diffusion trajectory prediction results, it generates an artifact compensation control signal, which is then coupled to the nozzle drive circuit to perform dynamic feedback control on the nozzle's motion posture.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] This invention integrates key factors such as ink diffusion, substrate material adsorption, and printhead motion by constructing a composite parameter matrix, providing comprehensive data support for subsequent analysis and compensation. Based on this, no-load error analysis is performed, generating a compensation matrix and a dynamic weight allocation function, which can adaptively adjust the compensation strategy according to actual conditions. A multi-scale gradient feature pyramid is constructed and features are fused to generate a heatmap. After correction with the no-load artifact offset error compensation matrix, a dynamic prediction model of ink diffusion behavior is established, which can accurately predict the ink diffusion trajectory. Finally, control signals are generated based on the compensation matrix and prediction results to control the printhead motion pose. This achieves precise optimization of the entire inkjet printing process, from data acquisition and analysis, error compensation to real-time control, greatly improving the accuracy and quality of the printed pattern, reducing the scrap rate caused by inkjet printing errors, lowering production costs, and enhancing the adaptability of the inkjet printer to different working conditions, significantly improving production efficiency. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating the principle of the real-time image calibration method for inkjet printers based on edge computing in this invention.

[0051] Figure 2 This is a flowchart of the method for constructing a composite parameter matrix in the real-time image calibration method for inkjet printers based on edge computing of the present invention;

[0052] Figure 3 This is a flowchart of the method for forming a fluid diffusion field parameter sub-matrix in the real-time image calibration method for inkjet printers based on edge computing of the present invention;

[0053] Figure 4 This is a flowchart of the method for obtaining the relative contribution rate of adsorption effect and the relative contribution rate of ink potential energy driving effect in the real-time image calibration method for inkjet printers based on edge computing of the present invention.

[0054] Figure 5 This is a flowchart of the method for calculating the geometric offset of the inkjet printer outline in the real-time image calibration method for inkjet printers based on edge computing of the present invention.

[0055] Figure 6 This is a flowchart of the method for constructing a multi-scale gradient feature pyramid in the real-time image calibration method for inkjet printers based on edge computing of the present invention;

[0056] Figure 7This is a functional block diagram of the real-time image calibration system for inkjet printers based on edge computing in this invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Example 1

[0059] Please see Figure 1 As shown, this embodiment provides a real-time image calibration method for inkjet printers based on edge computing, including:

[0060] Step S1000: Construct a composite parameter matrix, perform no-load error analysis, generate a no-load artifact offset error compensation matrix, establish a nonlinear mapping relationship between each element of the no-load artifact offset error compensation matrix and the composite parameter matrix, and generate a dynamic weight allocation function for the compensation matrix.

[0061] Further, step S1000 includes:

[0062] Step S1100: Construct the composite parameter matrix;

[0063] Furthermore, such as Figure 2 As shown, step S1100 includes:

[0064] Step S1110: Obtain the surface viscosity coefficient and potential energy gradient value to form a sub-matrix of fluid diffusion field parameters;

[0065] Furthermore, such as Figure 3 As shown, step S1110 includes:

[0066] Step S1111: Collect dynamic contact angle change data of ink droplets on the substrate surface and extract the surface viscosity coefficient;

[0067] Step S1112: Measure the spatial difference in pressure distribution inside the ink droplets and calculate the potential energy gradient value;

[0068] Step S1113: Quantize the surface viscosity coefficient and potential energy gradient value into a fluid diffusion field parameter submatrix.

[0069] Specifically, the dynamic contact angle variation data of ink droplets on the substrate surface is collected using a microfluidic sensor array to extract the surface viscosity coefficient. A microfluidic sensor array is a miniature sensor array fabricated using microelectromechanical systems (MEMS) technology, capable of accurately measuring various parameters of fluids at a microscale. During inkjet printing, the interaction between ink droplets and the substrate surface is influenced by multiple factors, one of which is the surface viscosity coefficient. The surface viscosity coefficient reflects the magnitude of the internal friction force experienced by the ink droplets as they flow on the substrate surface, determining the speed and uniformity of ink diffusion. For example, when the surface viscosity coefficient is large, the ink droplets have difficulty flowing on the substrate surface, resulting in slow diffusion and potentially unclear edges of the printed pattern; conversely, if the surface viscosity coefficient is small, the ink droplets diffuse too quickly, causing problems such as blurring and smudging in the printed pattern. By collecting dynamic contact angle variation data to extract the surface viscosity coefficient, the flow characteristics of ink on the substrate surface can be accurately obtained, providing fundamental data for subsequent analysis. This helps to assess the compatibility between the ink and the substrate material before printing, allowing for adjustments to printing parameters in advance and ensuring printing quality. Meanwhile, accurate surface viscosity coefficient data can provide accurate input parameters for the establishment of dynamic prediction models of ink diffusion behavior, improve the accuracy of model prediction, and thus achieve precise control of the inkjet printing process.

[0070] The potential energy gradient is calculated by measuring the spatial difference in pressure distribution within an ink droplet using a pressure sensor array. A pressure sensor array is a combination of sensors used to measure pressure distribution, acquiring pressure information at different locations within the ink droplet. The potential energy gradient represents the degree of change in potential energy per unit distance; during ink diffusion, it reflects the magnitude and direction of the driving force propelling the ink droplet. When a pressure difference exists within an ink droplet, a potential energy gradient is generated, causing the ink to diffuse from high-pressure areas to low-pressure areas. For example, if the internal pressure on one side of an ink droplet is greater than on the other, the potential energy gradient will push the droplet towards the side with lower pressure, affecting the shape and size of the printed pattern. Measuring the potential energy gradient helps understand the movement trend of ink within the droplet and analyze its impact on printing quality. By accurately determining the potential energy gradient, printhead design and printing processes can be optimized, and the spray pressure and angle of the printhead can be adjusted appropriately to ensure more uniform ink diffusion on the substrate surface, thereby improving the accuracy and quality stability of the printed pattern. Furthermore, the potential energy gradient data can also be used to study the kinetics of ink diffusion, providing a theoretical basis for improving ink formulations and developing ink products that are more suitable for inkjet printing processes.

[0071] The surface viscosity coefficient and potential energy gradient are quantized into sub-matrices of fluid diffusion field parameters. Quantization is the process of converting continuous physical quantities into discrete digital quantities for computer processing and analysis. By quantizing these two parameters into sub-matrix form, the key physical characteristics of the ink diffusion process can be presented in a structured data format. This structured data facilitates subsequent integration and analysis with other relevant data (such as substrate material adsorption characteristics data, printhead motion trajectory data, etc.). For example, when constructing a composite parameter matrix, the fluid diffusion field parameter sub-matrix, as one part, is interconnected with other sub-matrices, jointly providing comprehensive data support for inkjet image calibration. Moreover, the quantized sub-matrices can be directly used as input data for algorithms and models, facilitating efficient numerical calculations and analysis using computers. This enables the use of more complex and precise algorithms in studying ink diffusion behavior and optimizing inkjet printing processes, improving analysis efficiency and accuracy, thereby achieving refined control of the inkjet printing process and improving inkjet printing quality and production efficiency.

[0072] Step S1120: Measure the adsorption decay characteristic curves of ink on different material surfaces, extract the adsorption decay time constant and saturated adsorption concentration threshold, and form a sub-matrix of substrate material adsorption characteristic data.

[0073] Specifically, spectral analysis is used to measure the adsorption decay characteristic curves of ink on different material surfaces, and then the adsorption decay time constant and saturation adsorption concentration threshold are extracted to form a sub-matrix of substrate material adsorption characteristic data. Spectroscopic analysis is a technique that analyzes the composition and structure of a substance based on its absorption, emission, or scattering characteristics of light at different wavelengths. During inkjet printing, the adsorption of ink by the substrate material affects the ink diffusion and adhesion. The adsorption decay time constant reflects the rate at which ink is adsorbed onto the substrate material surface, while the saturation adsorption concentration threshold indicates the maximum amount of ink that the substrate material can adsorb. For example, when inkjet printing on paper packaging boxes, if the paper's adsorption decay time constant is small, it indicates that the ink is adsorbed quickly onto the paper surface, which may result in fast drying of the printed pattern, but may also lead to uneven color distribution. Conversely, a low saturation adsorption concentration threshold may result in insufficient ink adhesion to the paper surface, affecting the clarity and durability of the printed pattern. By measuring the adsorption decay characteristic curves and extracting relevant parameters to form a sub-matrix, a detailed understanding of the substrate material's adsorption characteristics of ink can be obtained. This helps in selecting suitable substrate materials and ink combinations, optimizing the coding process, such as adjusting the printhead's spray volume and spray frequency, to ensure that the ink can achieve the best adsorption and diffusion effect on the substrate surface, improve the quality and adhesion of the coding pattern, and at the same time reduce ink waste.

[0074] Step S1130: Track the nozzle movement process, acquire the nozzle vibration displacement signal, extract the spatial curvature change features of the nozzle movement trajectory from the nozzle vibration displacement signal, and form a spatial feature sub-matrix of the nozzle movement trajectory.

[0075] Specifically, a high-speed camera is used to track the printhead's movement at a high frame rate, acquiring the printhead's vibration and displacement signals. Then, Fourier transform is used to extract the spatial curvature variation features of the printhead's trajectory from these signals, quantizing them to generate a spatial feature sub-matrix of the printhead's trajectory. The high-speed camera can capture the details of the printhead's movement at a high frame rate, recording the vibration and displacement of the printhead during the coding process. Fourier transform is a mathematical method that converts time-domain signals into frequency-domain signals, facilitating the analysis of the signal's frequency components and characteristics. The spatial curvature variation features of the printhead's trajectory reflect the stability and accuracy of the printhead's movement. For example, during continuous coding, if the spatial curvature variation of the printhead's trajectory is large, it indicates instability in the printhead's movement, potentially leading to deviations in the coding pattern or uneven lines. By extracting this feature and quantizing it into a sub-matrix, the printhead's movement status can be monitored in real time. This is crucial for timely detection of printhead malfunctions or anomalies, facilitating early equipment maintenance and adjustments, and ensuring the stability of the coding quality. Meanwhile, the spatial feature submatrix of the printhead motion trajectory can also provide data support for optimizing the printhead motion control algorithm. By adjusting the printhead motion parameters, the printhead motion becomes smoother and more precise, thereby improving the accuracy and quality of the inkjet pattern.

[0076] Step S1140: Construct a composite parameter matrix based on the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic data sub-matrix, and the nozzle motion trajectory spatial characteristic sub-matrix.

[0077] Specifically, a composite parameter matrix is ​​constructed based on the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic data sub-matrix, and the printhead motion trajectory spatial characteristic sub-matrix. Integrating these three sub-matrices into a composite parameter matrix enables a unified description and analysis of multiple key factors in the inkjet printing process. The composite parameter matrix contains information on ink diffusion characteristics, substrate material adsorption characteristics, and printhead motion characteristics, providing a comprehensive data foundation for subsequent no-load error analysis, compensation matrix establishment, and ink diffusion behavior prediction. For example, when establishing a dynamic prediction model for ink diffusion behavior, the composite parameter matrix, as input data, can comprehensively consider the influence of multiple factors on ink diffusion, enabling the model to more accurately predict the diffusion trajectory and morphological changes of ink on the substrate surface. Furthermore, when performing inkjet printing error compensation, the composite parameter matrix provides a basis for determining compensation strategies and calculating compensation amounts. By analyzing the interrelationships between different sub-matrices, precise compensation for inkjet printing errors can be achieved, thereby improving the calibration accuracy of the inkjet printing image, ultimately enhancing the quality and production efficiency of inkjet-printed products, and reducing scrap rates and production costs.

[0078] Step S1200: Based on the fluid diffusion field parameter sub-matrix, obtain the artifact width distribution data formed during the ink diffusion process, and generate the diffusion artifact width distribution spectrum.

[0079] Specifically, step S1200 aims to obtain the artifact width distribution data formed during ink diffusion based on the fluid diffusion field parameter sub-matrix, and generate the diffusion artifact width distribution spectrum. This step plays a crucial role in subsequent analysis of error sources in the inkjet printing process and in achieving image calibration.

[0080] During inkjet printing, the diffusion behavior of ink on the substrate surface is influenced by various factors, with the surface viscosity coefficient and potential energy gradient value in the fluid diffusion field parameter sub-matrix being crucial in determining the ink diffusion characteristics. The surface viscosity coefficient reflects the magnitude of the internal friction force experienced by the ink as it flows on the substrate surface, affecting the speed and uniformity of ink diffusion; the potential energy gradient value represents the magnitude and direction of the driving force propelling the ink diffusion due to pressure differences within the ink. When ink diffuses, these factors can lead to uneven ink distribution on the substrate surface, resulting in artifacts. Artifacts refer to unwanted or distorted portions in the printed pattern that do not meet expectations, affecting the quality and accuracy of the printing.

[0081] By deeply analyzing the surface viscosity characteristics and potential energy distribution reflected in the fluid diffusion field parameter sub-matrix, the artifact width distribution data formed during ink diffusion can be extracted. For example, if the surface viscosity coefficient is small, the ink diffuses faster on the substrate surface, which may lead to an increase in artifact width; while a higher potential energy gradient may intensify ink diffusion in a specific direction, causing the artifact to extend in that direction. Organizing and analyzing this data generates a diffusion artifact width distribution spectrum. The diffusion artifact width distribution spectrum is a graph that presents artifact width data in a specific form. It can intuitively show the changes in artifact width at different locations, providing an intuitive and quantitative data foundation for subsequent error analysis and compensation.

[0082] This step yields significant benefits. First, generating the diffusion artifact width distribution spectrum helps to accurately locate the areas and extent of artifacts during the inkjet printing process. Analysis of the spectrum clearly identifies which areas suffer from severe artifacts, providing a clear direction for subsequent targeted error compensation. For example, when inkjet printing on packaging, if a large artifact width is found in a certain area, it may lead to unclear production dates or batch information. Locating this area using the diffusion artifact width distribution spectrum allows for focused adjustments and compensation in subsequent inkjet printing processes. Second, this step provides crucial data support for establishing a more accurate inkjet printing error model. Based on the diffusion artifact width distribution spectrum, the mechanism of artifact formation can be further studied, and its relationship with other factors (such as printhead movement and substrate material properties) can be analyzed, thereby establishing a more comprehensive inkjet printing error model and improving the ability to predict and control inkjet printing errors. Finally, generating the diffusion artifact width distribution spectrum is beneficial for optimizing inkjet printing process parameters. By comparing and analyzing the width distribution spectrum of diffusion artifacts generated under different process parameters, the optimal range of parameters such as surface viscosity coefficient and potential energy gradient can be determined. This allows for the adjustment of process parameters such as ink formulation and printhead injection pressure, thereby reducing artifact generation and improving coding quality.

[0083] Step S1300: The diffusion artifact width distribution spectrum is used as a priori constraint for the printing start stage under no-load conditions. It is loaded into the lightweight reverse generative adversarial topology network in the pre-built edge computing node to obtain the diffusion artifact priori constraint.

[0084] Specifically, step S1300 involves using the diffusion artifact width distribution spectrum as a priori constraint for the initial printing stage under no-load conditions, and loading it into a lightweight reverse generative adversarial topology network in a pre-built edge computing node to obtain the diffusion artifact priori constraint. This step, leveraging advanced network models and edge computing technology, helps to achieve accurate coding error compensation in the subsequent process.

[0085] Prior constraints refer to limitations set based on existing knowledge or experience before performing certain analyses or calculations. They can guide the model to train and predict in a direction that better reflects reality. In this step, the diffusion artifact width distribution spectrum serves as a prior constraint, providing initial information about artifacts during the inkjet printing process for the lightweight reverse generative adversarial topology network.

[0086] Lightweight reverse generative adversarial topology networks are network models specifically designed for edge devices. Edge computing is a distributed computing paradigm that brings computation and data storage closer to the data source or user, reducing data transmission latency and improving system response speed. This network model, through a carefully designed compact network structure and optimized computational operations, keeps the model size below 9.8MB, enabling real-time inference on resource-constrained edge devices. Real-time inference means the network can quickly process input data during the coding process and output prediction results promptly, meeting the real-time calibration requirements of coding machines.

[0087] By loading the diffusion artifact width distribution spectrum into the network, the network learns the features and patterns in this data to generate prior constraints on diffusion artifacts. For example, the network can learn information such as the changing trends of artifact width in different regions and the probability distribution of artifact occurrence from the diffusion artifact width distribution spectrum, thereby generating corresponding constraints. These constraints can be understood as the network's "expectation" of possible artifact situations during the coding process. Subsequently, when performing coding error analysis and compensation, these "expectations" can be used to more accurately judge the errors in the actual coding image and make corresponding adjustments.

[0088] The benefits of this step are multifaceted. From the perspective of error compensation accuracy, the prior constraint on diffusion artifacts provides a more reliable reference for subsequent error analysis. In actual inkjet printing, various factors can interfere with the printing image, often resulting in various errors. With prior constraints, it is possible to more accurately identify whether these errors are caused by artifacts or other factors, thereby improving the targeting and accuracy of error compensation. For example, when printing on cigarette packaging, the printing image may be affected by various factors such as uneven surface material of the outer shell and slight vibration of the printhead. Through the prior constraint on diffusion artifacts, it is possible to more accurately determine which errors are caused by ink diffusion artifacts, and then take more effective compensation measures. From a real-time perspective, the real-time inference capability of the lightweight reverse generative adversarial topology network on the edge device ensures that the inkjet printer can obtain prior constraint information in a timely manner during the printing process and adjust parameters such as printhead movement in real time based on this information, greatly improving the working efficiency of the inkjet printer. Furthermore, the lightweight design of this network model reduces the demand for edge device hardware resources, enabling the inkjet printer to achieve efficient image calibration functions with low-cost hardware configurations, demonstrating good economic efficiency and practicality.

[0089] Step S1400: Based on the fluid diffusion field parameter sub-matrix and the substrate material adsorption characteristic data sub-matrix, the relative contribution rate of the adsorption effect and the relative contribution rate of the ink potential energy driving effect are obtained.

[0090] Furthermore, such as Figure 4 As shown, step S1400 includes:

[0091] Step S1410: Based on the surface viscosity coefficient and potential energy gradient value in the fluid diffusion field parameter sub-matrix, a mathematical and physical model of the ink droplet diffusion process on the substrate surface is constructed. The finite element analysis method is used to numerically simulate the fluid dynamics characteristics inside the droplet to obtain a quantitative description function of the ink potential energy driving effect.

[0092] Step S1420: Based on the adsorption decay time constant and saturated adsorption concentration threshold in the adsorption characteristic data sub-matrix of the substrate material, an adsorption kinetic model between ink droplets and the substrate surface is constructed. Molecular dynamics simulation method is used to quantitatively analyze the adsorption process and obtain a quantitative description function of the adsorption effect.

[0093] Step S1430: Normalize the quantitative description functions of ink potential energy driving effect and adsorption effect to obtain the relative contribution rate of adsorption effect and the relative contribution rate of ink potential energy driving effect.

[0094] Specifically, the purpose of step S1400 is to obtain the relative contribution rate of the adsorption effect and the relative contribution rate of the ink potential energy driving effect based on the fluid diffusion field parameter sub-matrix and the substrate material adsorption characteristic data sub-matrix. This is crucial for a deeper understanding of the ink diffusion mechanism on the substrate surface and for achieving precise control of the coding process. In step S1410, the mathematical physics model is a tool for mathematically abstracting and simplifying actual physical phenomena. By establishing such a model, the diffusion process of ink droplets on the substrate surface can be described mathematically. The surface viscosity coefficient and potential energy gradient value, as key parameters of the model, directly affect the diffusion behavior of the ink droplets. The finite element analysis method is a numerical calculation method that discretizes a continuous solution domain into a finite number of elements. It can transform complex physical problems into a system of algebraic equations for solution. In this step, the finite element analysis method is used to simulate the hydrodynamic characteristics inside the ink droplets, enabling detailed analysis of the distribution of velocity fields, pressure fields, etc., inside the droplets, thereby obtaining a quantitative description function of the ink potential energy driving effect. For example, during the simulation, it is possible to observe how the fluid inside the ink droplet flows under different combinations of surface viscosity coefficients and potential energy gradients, and how this flow affects the diffusion range and velocity of the droplet on the substrate surface. Through such simulation and analysis, the resulting quantitative descriptive function can accurately express the relationship between the ink potential energy driving effect and related parameters, providing a quantitative basis for accurately assessing the impact of the ink potential energy driving effect on the coding process. This helps to adjust relevant parameters according to different needs during the coding process design stage, optimize the ink diffusion effect, and improve coding quality.

[0095] In step S1420, the adsorption kinetics model is used to describe the change in the adsorption process of ink droplets on the substrate surface over time. The adsorption decay time constant reflects the rate of ink adsorption on the substrate surface, while the saturation adsorption concentration threshold represents the maximum amount of ink that the substrate material can adsorb. Molecular dynamics simulation is a computer simulation technique that studies the behavior of matter at the atomic scale. By simulating the motion and interaction of molecules, it studies the microscopic mechanism of the adsorption process. In this step, using molecular dynamics simulation, the interaction process between ink molecules and substrate surface molecules can be observed in detail, and how this interaction affects the amount and rate of ink adsorption on the substrate surface. For example, simulation can reveal the differences in the distribution of adsorption sites and the formation process of the adsorption layer on the substrate surface under different adsorption decay time constants and saturation adsorption concentration thresholds. Through such simulation and analysis, the obtained quantitative description function of the adsorption effect can accurately reflect the relationship between the adsorption process and related parameters, providing a microscopic basis for a deeper understanding of the influence of the adsorption effect on the inkjet printing process. This is of great significance for selecting appropriate substrate materials and ink combinations, as well as optimizing parameters such as drying time in the coding process, which can improve the adhesion of ink to the substrate surface and the clarity of the coding pattern.

[0096] In step S1430, normalization is a method to transform data with different dimensions or value ranges to a unified scale. It eliminates dimensional differences between data, facilitating comparison and analysis. By normalizing the two quantitative descriptive functions, the ink potential energy driving effect and adsorption effect can be measured on the same scale, thus obtaining their relative importance in the entire ink diffusion process, i.e., their relative contribution rate. For example, if the relative contribution rate of the adsorption effect is found to be high after normalization, it indicates that under the current coding conditions, adsorption has a greater impact on ink diffusion. Therefore, when optimizing the coding process, more attention needs to be paid to the adsorption characteristics of the substrate material, selecting a substrate material with more suitable adsorption performance, or adjusting the drying conditions during the coding process to optimize the ink diffusion effect. Conversely, if the relative contribution rate of the ink potential energy driving effect is high, then parameters related to ink potential energy, such as the printhead's injection pressure, should be adjusted. By determining these two relative contribution rates, coding process parameters can be optimized more specifically, achieving precise control of the ink diffusion process, thereby improving the stability and reliability of coding quality.

[0097] Step S1500: Combine the prior constraints of diffusion artifacts to perform error analysis on the inkjet image under no-load conditions and generate the no-load artifact offset error compensation matrix.

[0098] Further, step S1500 includes:

[0099] Step S1510: Combining the prior constraints of diffusion artifacts, the inkjet printer prints a preset reference pattern under no-load conditions and collects the actual inkjet image under no-load conditions.

[0100] Specifically, the diffusion artifact prior constraint is obtained by loading the diffusion artifact width distribution spectrum into a lightweight inverse generative adversarial network. It reflects an expected information about potential diffusion artifacts under the current idle state of the inkjet printing system, including predictions of the artifact's shape, location, and size. In actual operation, the inkjet printer runs in an idle state, meaning it does not perform actual product coding; it only prints a reference pattern on a specific substrate material according to preset coding parameters. For inkjet printers used for coding cigarette packaging paper, the selected substrate material can be paper similar to cigarette packaging paper to simulate the actual coding environment as closely as possible. The preset reference pattern is usually carefully designed, containing various shapes and line combinations, such as patterns composed of rectangles, circles, and lines of different sizes. These patterns can cover various geometric features that may occur during the coding process, facilitating comprehensive analysis of the coding image later. After coding is completed, a high-precision image acquisition device, such as an industrial camera, is used to photograph the pattern printed on the substrate material. When acquiring images, it is necessary to ensure the stability of the acquisition environment and avoid factors such as changes in lighting and equipment vibration that may interfere with image quality in order to obtain clear and accurate actual inkjet printing images. For example, by adjusting parameters such as camera focal length, aperture, and exposure time, the details of the inkjet printing pattern in the image can be clearly presented.

[0101] This step offers several beneficial effects. From an error analysis perspective, the acquired actual inkjet images provide a direct data source for subsequently determining errors in the inkjet printing process. By comparing the actual inkjet images with reference patterns, deviations in shape, position, and size can be clearly observed. These deviations represent the errors present in the inkjet printing process, providing fundamental data for subsequent precise error compensation. In cigarette packaging inkjet printing, if a character in the actual inkjet image shifts position, the amount of shift can be determined through subsequent analysis, allowing for appropriate adjustments. From the perspective of improving inkjet printing quality, combining image acquisition with prior constraints on diffusion artifacts allows for more targeted attention to artifact-related errors. Because prior constraints provide information about artifacts, areas potentially affected by artifacts can be observed and recorded during image acquisition, facilitating in-depth analysis of the artifact impact mechanism on inkjet printing quality and providing a basis for optimizing the inkjet printing process. For example, if prior constraints indicate that artifacts due to ink diffusion may occur in a certain area, this area can be the focus during image acquisition. If blurring or distortion is found in the actual inkjet image in that area, further analysis can be conducted to identify the contributing factors and implement corrective measures. Furthermore, acquiring images under no-load conditions eliminates interference from the actual product content, providing a purer reflection of the printer's performance and error characteristics. This makes the subsequently generated no-load artifact offset error compensation matrix more targeted and accurate, ultimately improving inkjet quality and reducing product defects caused by inkjet errors.

[0102] Step S1520: Perform feature point matching between the acquired actual inkjet image and the reference pattern, and calculate the geometric offset of the inkjet outline.

[0103] Furthermore, such as Figure 5 As shown, step S1520 includes:

[0104] Step S1521: Extract key feature points from the actual inkjet image and the reference pattern, and construct a feature point matching matrix;

[0105] Step S1522: Based on the feature point matching matrix, fit the geometric offset function of the inkjet printing profile to obtain the local geometric offset and the global geometric offset.

[0106] Step S1523: Combine the local geometric offset and the overall geometric offset to form the geometric offset of the inkjet printing profile.

[0107] Specifically, the core task of step S1520 is to perform feature point matching between the acquired actual inkjet image and the reference pattern, calculate the geometric offset of the inkjet outline, determine the error in the inkjet process, and provide key data support for subsequent error compensation.

[0108] Step S1521 uses the Scale Invariant Feature Transform (SIFT) algorithm to extract key feature points from the actual inkjet image and the reference pattern, constructing a feature point matching matrix. The Scale Invariant Feature Transform (SIFT) algorithm is a feature extraction technique widely used in computer vision. Its principle is based on feature detection and description of images at different scales. Gaussian filtering is applied to the image at different scales to construct a Gaussian pyramid, and then differential Gaussian (DoG) operations are used to detect key points in the image. For each key point, the gradient orientation histogram of its surrounding neighborhood is calculated to form a feature descriptor. These feature descriptors possess scale invariance, rotation invariance, and a certain degree of illumination invariance, enabling them to stably represent the features of the image under different conditions. In inkjet image analysis, factors such as nozzle jitter, ink diffusion, and differences in substrate material surface properties may exist during the inkjet printing process, leading to differences in scale, rotation, and illumination between the actual inkjet image and the reference pattern. The SIFT algorithm can effectively address these variations and accurately extract key feature points from the image. For example, in cigarette packaging inkjet printing, the characters in the actual inkjet image may rotate and scale due to slight vibrations of the printhead. Additionally, the varying textures of the packaging paper surface can cause differences in light reflection. The SIFT algorithm can extract key feature points such as corner points and endpoints of the character edges from these complex variations. By extracting key feature points from the actual inkjet image and a reference pattern, a feature point matching matrix is ​​constructed. The elements in the matrix represent matching feature point pairs in the two images. The beneficial effect of this step is that the application of the SIFT algorithm significantly improves the accuracy and stability of feature point extraction. Accurate feature point extraction is the foundation for subsequent precise calculation of geometric offsets, reducing mismatches caused by image variations. By constructing the feature point matching matrix, a reliable data foundation is provided for subsequent calculation of the geometric offset of the inkjet outline, making subsequent error analysis and compensation more accurate, thus helping to improve inkjet quality and ensure the accuracy and consistency of the inkjet pattern.

[0109] In step S1522, fitting the geometric offset function involves mathematically processing the data in the feature point matching matrix to find a function that describes the geometric transformation relationship between the actual inkjet image and the reference pattern. Since errors during the inkjet printing process may differ at different positions on the inkjet outline, it is necessary to calculate both local and global geometric offsets separately. The local geometric offset reflects the offset of various local areas of the inkjet outline, accurately pinpointing the deviation at a specific location within the inkjet pattern. For example, in a brand logo printed on cigarette packaging, a stroke of a letter might shift due to momentary nozzle vibration; calculating the local geometric offset accurately determines the direction and distance of this shift. The global geometric offset describes the overall offset trend of the entire inkjet outline relative to the reference pattern, reflecting systematic deviations during the inkjet printing process. For example, if the entire brand logo shifts in a certain direction during printing, the global geometric offset reflects this overall deviation. By fitting the geometric offset function to obtain the local and global geometric offsets, a comprehensive understanding of the inkjet outline's offset can be achieved. This facilitates in-depth analysis of the sources and nature of inkjet printing errors. Local geometric offsets can help identify anomalies in localized areas during the coding process, such as localized deviations caused by individual printhead malfunctions, providing a basis for targeted printhead adjustments. Global geometric offsets, on the other hand, help evaluate the overall performance and stability of the coding machine, providing a reference for adjusting its parameters, thereby improving coding accuracy and stability, and reducing distortion and offset of the coding pattern.

[0110] Step S1523 integrates the results of the first two sub-steps, combining local and overall offset information to form a complete geometric offset data for the inkjet printing outline. The geometric offset of the inkjet printing outline comprehensively describes the differences in shape and position between the actual inkjet image and the reference pattern. It is a comprehensive indicator that includes the offset information of each part of the inkjet pattern. For example, in cigarette packaging inkjet printing, the geometric offset of the inkjet printing outline not only reflects the offset of individual characters but also the overall offset of the entire inkjet pattern (such as a pattern composed of brand logos, production dates, etc.) relative to the reference pattern. This complete geometric offset data is crucial for the subsequent generation of the empty artifact offset error compensation matrix, providing accurate error information for the compensation matrix. Accurate geometric offsets allow for more precise calculation of the compensation amount, making subsequent inkjet error compensation more accurate and effective. When generating the compensation matrix, corresponding compensation strategies can be formulated based on the specific values ​​of the geometric offsets for offsets of different positions and degrees, thereby effectively reducing the offset and deformation of the inkjet pattern, improving inkjet quality, ensuring that the inkjet pattern meets the expected design requirements, and enhancing the product's appearance quality and readability.

[0111] Step S1530: The geometric offset is represented in a matrix form to generate the empty artifact offset error compensation matrix.

[0112] Further, step S1530 includes:

[0113] Step S1531: Based on the relative contribution rate of the adsorption effect and the relative contribution rate of the ink potential energy driving effect, the local geometric offset and the overall geometric offset are weighted and combined to obtain the comprehensive geometric offset matrix.

[0114] Step S1532: Normalize and discretize the elements in the comprehensive geometric offset matrix to generate the empty artifact offset error compensation matrix.

[0115] Specifically, the main purpose of step S1530 is to represent the geometric offset of the inkjet pattern in a matrix form, generate an empty artifact offset error compensation matrix, and provide key data support for subsequent accurate error compensation of the nozzle motion pose.

[0116] In step S1531, the relative contribution rates of the adsorption effect and the ink potential energy driving effect are calculated through previous steps. They reflect the relative importance of adsorption and ink potential energy driving effects in the ink diffusion process, respectively. During inkjet printing, both the adsorption effect and the ink potential energy driving effect affect ink diffusion, leading to offsets in the printed pattern. For example, in cigarette packaging paper printing, different materials of packaging paper have different adsorption capacities for ink. A higher relative contribution rate of the adsorption effect indicates that the adsorption of ink by the packaging paper has a greater impact on the offset of the printed pattern. The ink potential energy driving effect is related to factors such as the internal pressure and surface tension of the ink. A higher relative contribution rate of the ink potential energy driving effect indicates that the potential energy driving effect of the ink itself has a more significant impact on the offset of the printed pattern. Local geometric offset and global geometric offset describe the offset of the printed outline from different perspectives. The weighted combination assigns different weights to the local and global geometric offsets based on the relative importance of the adsorption effect and the ink potential energy driving effect, and then performs a combined calculation. For example, if the relative contribution rate of the adsorption effect is 0.6 and the relative contribution rate of the ink potential energy driving effect is 0.4, when calculating the comprehensive geometric offset, the weight corresponding to the local geometric offset may be 0.6, and the weight corresponding to the overall geometric offset may be 0.4. The comprehensive geometric offset matrix is ​​obtained through this weighted calculation. The beneficial effect of this step is that it considers the differences in the contribution of different factors to the offset of the inkjet pattern. By integrating the local and overall geometric offsets through a weighted combination, the comprehensive geometric offset matrix can more accurately reflect the combined influence of various factors on the offset of the inkjet pattern. Compared to considering only a single factor or a simple combination without weighting, this method can more comprehensively consider the complexities of the inkjet printing process, providing a more reliable data foundation for subsequently generating a more accurate empty artifact offset error compensation matrix. This helps improve the accuracy of inkjet error compensation, thereby improving inkjet quality, reducing errors and defects in the inkjet pattern, and ensuring that the inkjet pattern is presented more accurately and clearly on cigarette packaging paper.

[0117] In step S1532, the normalization process maps the elements in the comprehensive geometric offset matrix to a specific interval, typically [0,1]. The purpose is to eliminate dimensional differences between elements, making them comparable. During the coding process, geometric offsets may have different numerical ranges due to different measurement units or calculation methods. Normalization unifies these different ranges into a standard interval, facilitating subsequent calculations and analysis. Discretization converts continuous values ​​into discrete values ​​for easier computer processing and storage. When generating the idle artifact offset error compensation matrix, its row and column count matches that of the inkjet printer nozzle array. This is because the compensation matrix needs to compensate for the coding error of each nozzle, and matching it to the nozzle array ensures the accuracy and specificity of the compensation. For example, assuming the elements in the comprehensive geometric offset matrix represent the offset distance of the coding pattern at different positions, where the unit of measurement might be pixels, normalization converts these offset distances into relative values, allowing comparisons of offsets at different positions on the same scale. After discretization, these relative values ​​are converted into discrete numerical values, such as integers, to facilitate subsequent calculations in the compensation algorithm. The generated no-load artifact offset error compensation matrix can be directly used to adjust the printhead movement or ink ejection parameters to compensate for errors in the coding process. For example, based on the element values ​​in the compensation matrix, the ejection time, ejection volume, or printhead trajectory of each nozzle can be adjusted to correct the offset of the coding pattern. The beneficial effect of this step is that normalization and discretization make the comprehensive geometric offset matrix more suitable for subsequent calculations and applications. The no-load artifact offset error compensation matrix generated through these processes can be used more effectively for coding error compensation, improving the accuracy and efficiency of compensation. The design, consistent with the number of rows and columns of the inkjet printer nozzle array, ensures that the compensation matrix can accurately correct the coding error of each nozzle, thereby significantly improving coding quality, reducing errors and defects in the coding pattern, and ensuring that the coding pattern on cigarette packaging paper is clear, accurate, and meets product quality requirements.

[0118] Step S1600: Establish the nonlinear mapping relationship between each element of the unloaded artifact offset error compensation matrix and the composite parameter matrix, and generate the dynamic weight allocation function of the compensation matrix.

[0119] Further, step S1600 includes:

[0120] Step S1610: The relative contribution rate of adsorption effect, the relative contribution rate of ink potential energy driving effect, the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic sub-matrix, and the printhead motion trajectory spatial feature sub-matrix are used as input parameters to construct a composite input vector.

[0121] Step S1620: Learn the nonlinear mapping relationship between the composite input vector and each element of the empty artifact offset error compensation matrix using a neural network method;

[0122] Step S1630: Generate a dynamic weight allocation function for the compensation matrix based on the learned nonlinear mapping relationship.

[0123] Specifically, step S1600 aims to establish a nonlinear mapping relationship between each element of the idle artifact offset error compensation matrix and the composite parameter matrix, and generate a dynamic weight allocation function for the compensation matrix, thereby achieving adaptive adjustment of the inkjet printing error compensation and improving inkjet printing accuracy.

[0124] In step S1610, the relative contribution rates of the adsorption effect and the ink potential energy driving effect reflect the relative importance of adsorption and ink potential energy driving effects in the ink diffusion process. The fluid diffusion field parameter sub-matrix includes surface viscosity coefficient and potential energy gradient values, reflecting the hydrodynamic characteristics of ink diffusion. The substrate material adsorption characteristic sub-matrix describes the adsorption characteristics of the substrate material on ink through the adsorption decay time constant and saturation adsorption concentration threshold. The printhead motion trajectory spatial feature sub-matrix characterizes the stability and accuracy of printhead motion. In the actual application scenario of cigarette packaging paper inkjet printing, different packaging paper materials have different ink adsorption performance. For example, paper packaging and plastic film packaging have different adsorption decay time constants and saturation adsorption concentration thresholds, which affect the relative contribution rate of the adsorption effect. The vibration and displacement of the printhead during the inkjet printing process are reflected by the printhead motion trajectory spatial feature sub-matrix. Combining these parameters to construct a composite input vector can comprehensively reflect the key factors related to ink diffusion, substrate material characteristics, and printhead motion during the inkjet printing process. By constructing a composite input vector, multiple factors affecting inkjet printing error compensation are integrated into a single vector, providing comprehensive and structured data for subsequent neural network learning. This enables the neural network to simultaneously consider the combined impact of multiple factors on inkjet printing error compensation, which helps improve the accuracy of the neural network's learning of mapping relationships. Consequently, it lays the foundation for generating a more accurate dynamic weight allocation function for the compensation matrix, thereby enhancing the precision of inkjet printing error compensation.

[0125] In step S1620, the neural network is a method based on representation learning of data. It consists of a large number of nodes (neurons) and edges connecting these nodes, learning patterns and relationships in the data by adjusting the connection weights between nodes. In this step, the input of the neural network is a composite input vector, and the output is the elements of the empty artifact offset error compensation matrix. During training, the neural network continuously adjusts its weights to minimize the error between the predicted output and the actual elements of the empty artifact offset error compensation matrix. For example, in cigarette packaging inkjet printing, when the nozzle trajectory changes, the nozzle trajectory spatial feature submatrix in the composite input vector changes accordingly. The neural network learns how this change affects the inkjet printing error compensation, thereby adjusting the elements of the output compensation matrix. Because the relationships between various factors in the inkjet printing process are complex and exhibit nonlinear characteristics, traditional linear methods are difficult to accurately describe these relationships. However, the powerful nonlinear fitting ability of the neural network can effectively capture the complex nonlinear mapping relationship between the composite input vector and the elements of the empty artifact offset error compensation matrix. By learning this relationship, the neural network can accurately predict the corresponding compensation matrix elements based on changes in input parameters, providing an accurate basis for the subsequent generation of dynamic weight allocation functions. This enables precise control of inkjet printing error compensation, improves inkjet printing quality, and reduces errors and defects in inkjet printing patterns.

[0126] In step S1630, the dynamic weight allocation function of the compensation matrix is ​​a function that adaptively adjusts the no-load artifact offset error compensation matrix based on the real-time changes in fluid dynamic parameters and material property parameters during the printing process. During the inkjet printing process on cigarette packaging paper, the fluid dynamic parameters and material property parameters change in real time due to variations in the packaging paper material (such as changes in absorbency caused by differences in the water absorption of different batches of paper), or changes in the motion accuracy of the printhead after prolonged operation. At this time, the dynamic weight allocation function of the compensation matrix adjusts the weights of each element in the no-load artifact offset error compensation matrix according to these changes. For example, if the adsorption effect in a certain area is enhanced, the function will correspondingly increase the weights of the compensation matrix elements related to the adsorption effect, making the compensation more focused on correcting the printing error caused by adsorption. By generating the dynamic weight allocation function of the compensation matrix, the no-load artifact offset error compensation matrix can be dynamically adjusted under different printing conditions, ensuring the accuracy of printing error compensation. This helps improve the adaptability of the inkjet printer in various complex situations. No matter what kind of cigarette packaging paper it encounters, or what slight changes occur in the printhead, it can achieve accurate inkjet error compensation, thereby improving the stability of inkjet quality, reducing scrap rate, and lowering production costs.

[0127] Step S2000: Based on the composite parameter matrix, construct a multi-scale gradient feature pyramid, which includes a bottom layer, a middle layer, and a top layer; perform tensor fusion on the bottom, middle, and top layer features of the multi-scale gradient feature pyramid to generate a feature correlation heatmap covering different abstraction levels; combine the empty artifact offset error compensation matrix to correct the feature correlation heatmap, and establish a dynamic prediction model for ink diffusion behavior to generate real-time diffusion trajectory prediction results;

[0128] Further, step S2000 includes:

[0129] Step S2100: Construct a multi-scale gradient feature pyramid based on the composite parameter matrix. The multi-scale gradient feature pyramid includes a bottom layer, a middle layer, and a top layer.

[0130] Furthermore, such as Figure 6 As shown, step S2100 includes:

[0131] Step S2110: Based on the fluid diffusion field parameter sub-matrix, extract the multi-scale gradient invariant feature clusters formed during the diffusion of ink on the substrate surface, and construct the bottom layer of the multi-scale gradient feature pyramid; the multi-scale gradient invariant feature clusters include a 64-dimensional directional gradient histogram.

[0132] Step S2120: Based on the adsorption characteristic data sub-matrix of the substrate material, capture the micro-porosity distribution mapping of the substrate material surface and construct the middle layer of the multi-scale gradient feature pyramid.

[0133] Step S2130: Based on the spatial feature submatrix of the nozzle motion trajectory, analyze the second derivative fluctuation features of the nozzle motion trajectory and construct the top layer of the multi-scale gradient feature pyramid.

[0134] Specifically, the purpose of step S2100 is to construct a multi-scale gradient feature pyramid based on the composite parameter matrix, and extract features related to ink diffusion, substrate material properties and printhead motion from different levels, so as to provide a basis for subsequent analysis and prediction of ink diffusion behavior.

[0135] In step S2110, the 64-dimensional histogram of oriented gradients (OG) from the multi-scale gradient-invariant feature clusters is used to characterize the scale-space structural features of the ink diffusion profile. The OG histogram is generated by calculating the gradient direction and magnitude of each pixel in the image and statistically analyzing the gradient direction distribution within a certain region. Calculations at different scales yield multi-scale OG histograms, which reflect the structural features of the ink diffusion profile at different sizes and resolutions. Taking cigarette packaging inkjet printing as an example, in the early stages of ink diffusion, the small-scale OG histogram can capture subtle changes at the edges of ink droplets; as diffusion progresses, the large-scale OG histogram reflects the overall diffusion trend. By extracting these multi-scale gradient-invariant feature clusters to construct the base layer of the pyramid, fundamental feature information on ink diffusion is provided for subsequent analysis. These features reflect the structural changes in the early and ongoing processes of ink diffusion, contributing to a deeper understanding of the dynamic process of ink diffusion. Meanwhile, it provides underlying data support for subsequent feature fusion and analysis, enabling the study of ink diffusion behavior to start from the most basic diffusion profile structure and gradually combine other factors for comprehensive analysis, thereby improving the accuracy and comprehensiveness of ink diffusion behavior analysis.

[0136] In step S2120, the micropore distribution on the surface of the substrate material affects the ink adsorption behavior on its surface. By introducing a fractal dimension descriptor to establish a correlation model between the microstructure characteristics of the substrate material and the ink adsorption behavior, the discriminative ability of the middle layer features can be improved. The fractal dimension descriptor is used to quantify the complexity of the material's microstructure, and its value reflects the irregularity and self-similarity of the micropore distribution. In the application of cigarette packaging paper, different materials of packaging paper have different micropores and fractal dimensions. For example, paper packaging paper has a larger and more complex micropore distribution, resulting in a relatively high fractal dimension; while plastic film packaging paper has a smaller micropore and a lower fractal dimension. By capturing the micropore distribution mapping to construct the pyramid middle layer, the microstructure characteristics of the substrate material can be linked to the ink adsorption behavior. This helps to analyze the influence mechanism of different substrate materials on ink diffusion, such as how micropore affects the adsorption rate and amount of ink, and thus the ink diffusion process. At the same time, when the mid-layer features are integrated with the bottom and top-layer features, the interaction of multiple factors during the ink diffusion process can be reflected more comprehensively, improving the accuracy of the prediction of ink diffusion behavior.

[0137] In step S2130, the second derivative fluctuation characteristics of the printhead motion trajectory reflect the acceleration changes of the printhead motion, which can characterize the instability of the printhead motion. The time-frequency domain features of the printhead vibration displacement signal are extracted using short-time Fourier transform, and then the nonlinear dynamic behavior of the printhead motion trajectory is analyzed using chaos theory to construct a top-level feature map. Short-time Fourier transform can convert time-domain signals into time-frequency domain signals, showing the energy distribution of the signal at different times and frequencies. Chaos theory is used to study seemingly random but actually regular behavior in nonlinear systems. During cigarette packaging coding, the vibration and displacement of the printhead cause changes in the coding position and pattern quality. If the printhead motion is unstable, its second derivative fluctuation will be large. By analyzing these fluctuation characteristics and constructing a top-level feature map, the printhead motion status can be monitored in real time. For example, when the printhead experiences abnormal vibration during coding, the top-level features will change accordingly, which helps to detect printhead faults or abnormalities in a timely manner. After the top-layer features are integrated with the bottom and middle-layer features, multiple factors such as ink diffusion, substrate material adsorption, and printhead movement can be comprehensively considered to more accurately predict ink diffusion behavior, providing an important basis for real-time adjustment of coding parameters and optimization of coding quality.

[0138] Step S2200: Tensor fusion is performed on the bottom, middle and top features of the multi-scale gradient feature pyramid to generate a feature correlation heatmap covering different levels of abstraction.

[0139] Further, step S2200 includes:

[0140] Step S2210: The multi-scale gradient invariant feature clusters of the bottom layer are fused with the micro-porosity distribution mapping of the middle layer to obtain the local ink diffusion behavior characteristics.

[0141] Step S2220: The second derivative fluctuation characteristics of the top-level nozzle motion trajectory are fused with the local ink diffusion behavior characteristics to obtain the correlation matrix between ink distribution and nozzle motion.

[0142] Step S2230: Perform thermodynamic encoding on the correlation matrix between ink distribution and printhead motion to generate a feature correlation heatmap that reflects the correlation strength between the local feature distribution of the bottom layer and the global feature of the top layer.

[0143] Specifically, the purpose of step S2200 is to perform tensor fusion on the bottom, middle and top features of the multi-scale gradient feature pyramid to generate a feature correlation heatmap covering different levels of abstraction, so as to reveal the intrinsic relationship between ink diffusion, substrate material properties and printhead motion, and provide a basis for subsequent accurate analysis and prediction of ink diffusion behavior.

[0144] In step S2210, a recursive tensor product algorithm is used to fuse the multi-scale gradient-invariant feature cluster at the bottom layer with the micro-porosity distribution mapping at the middle layer to obtain local ink diffusion behavior features. The recursive tensor product algorithm is a method for handling tensor operations; by progressively calculating the product between tensors, it can effectively fuse information from different dimensions. The multi-scale gradient-invariant feature cluster contains a 64-dimensional directional gradient histogram, characterizing the scale-space structure of the ink diffusion profile and reflecting the profile changes at different scales during ink diffusion. The micro-porosity distribution mapping at the middle layer reflects the influence of the substrate material's surface microstructure on ink adsorption. Taking cigarette packaging paper inkjet printing as an example, during the printing process, the ink diffuses on the packaging paper surface. The bottom-layer features can capture the detailed changes in the ink diffusion profile, such as the serrated structure of the ink droplet edges at different scales. The middle-layer features reflect how the micro-porosity of the packaging paper affects ink adsorption, thus affecting the initiation and continuation of the diffusion process. By fusing these two layers of features using the recursive tensor product algorithm, the diffusion characteristics of the ink itself and the influence of the substrate material can be comprehensively considered to obtain local ink diffusion behavior features. This fused feature provides a more comprehensive description of ink diffusion behavior within a local area, offering richer information for subsequent analysis. It helps to accurately analyze differences in ink diffusion across different substrate materials. For example, in different parts of cigarette packaging paper, the ink diffusion rate and morphology vary due to differences in micropore size. These differences can be accurately captured through local ink diffusion behavior characteristics, providing a basis for precise control of ink diffusion and improving the quality and stability of the coding pattern.

[0145] Step S2220 employs a higher-order singular value decomposition (HSV) algorithm to fuse the second-order derivative fluctuation characteristics of the top-level printhead motion trajectory with the local ink diffusion behavior characteristics, obtaining an ink distribution-printhead motion correlation matrix. The HSV algorithm is an extension of traditional singular value decomposition, capable of processing multi-dimensional tensor data and extracting key information from the data. The second-order derivative fluctuation characteristics of the top-level printhead motion trajectory reflect the changes in printhead acceleration, characterizing the instability of printhead motion; the local ink diffusion behavior characteristics describe the diffusion characteristics of ink within a local area. In the inkjet printing process on cigarette packaging paper, the instability of printhead motion leads to changes in the ink spray position and quantity, thus affecting the ink distribution on the packaging paper. For example, when the printhead motion acceleration suddenly changes, the ink spray direction and velocity will change, causing deviations in the ink distribution on the packaging paper. By fusing these two characteristics using the HSV algorithm, a correlation mapping between the global ink distribution pattern and the printhead motion pattern can be established, resulting in an ink distribution-printhead motion correlation matrix. This matrix clearly demonstrates how printhead motion affects ink distribution, providing a powerful tool for analyzing error sources in the coding process. By analyzing the correlation matrix, the impact of abnormal printhead motion on ink distribution can be identified in a timely manner, allowing for adjustments to printhead motion parameters, optimization of the coding process, reduction of coding quality issues caused by printhead instability, and improvement of coding accuracy and consistency.

[0146] Step S2230 performs thermodynamic encoding on the correlation matrix between ink distribution and printhead movement, generating a feature correlation heatmap reflecting the correlation strength between the bottom-level local feature distribution and the top-level global feature distribution. The feature correlation heatmap employs an adaptive color mapping scheme, dynamically adjusting color saturation and brightness gradients to enhance the visual prominence of key ink diffusion areas. Thermodynamic encoding is a method for converting data into an intuitively representable form. By encoding the correlation matrix between ink distribution and printhead movement, the data information in the matrix can be transformed into a visual heatmap. In the scenario of inkjet printing on cigarette packaging, the heatmap can intuitively show the correlation strength between ink distribution and printhead movement in different areas. For example, on the heatmap, the darker or more vibrant the color, the greater the influence of printhead movement on the ink distribution within that area. The adaptive color mapping scheme dynamically adjusts the color saturation and brightness gradients according to the characteristics of the data, making key ink diffusion areas (such as ink concentration areas and diffusion edges) more prominent in the heatmap.

[0147] This visualized feature correlation heatmap offers several beneficial effects. Firstly, it provides operators with an intuitive analytical tool. By observing the heatmap, they can quickly understand the relationship between ink distribution and printhead movement during the coding process, identifying potential problem areas, such as the correlation between abnormal ink distribution and unstable printhead movement in certain areas. Secondly, for subsequent ink diffusion behavior analysis and prediction, the heatmap provides a clear data visualization foundation, helping researchers to analyze the interactions between different factors more deeply, providing strong support for optimizing the coding process and improving coding quality. Simultaneously, the heatmap facilitates comparison and combined analysis with other data, further enhancing the understanding and control of the coding process.

[0148] In step S2300, the feature correlation heatmap is corrected by combining the empty artifact offset error compensation matrix, and a dynamic prediction model for ink diffusion behavior is established to generate real-time diffusion trajectory prediction results.

[0149] Step S2300 aims to combine the empty artifact offset error compensation matrix to correct the feature correlation heatmap, establish a dynamic prediction model for ink diffusion behavior, and generate real-time diffusion trajectory prediction results for fine control of inkjet printing contours, so as to achieve precise control and quality improvement of the inkjet printing process.

[0150] Further, step S2300 includes:

[0151] Step S2310: Construct a dynamic evolution model of ink diffusion behavior during the coding process, introduce the empty artifact offset error compensation matrix into the dynamic evolution model of ink diffusion behavior, establish an ink diffusion offset correction equation, perform offset correction on the feature correlation heatmap, and obtain the corrected feature correlation heatmap sequence.

[0152] Specifically, the kinetic evolution model, based on the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic data sub-matrix, and the printhead motion trajectory spatial characteristic sub-matrix, uses the finite element method (FEM) to numerically simulate the ink diffusion process. The FEM is a numerical calculation method that discretizes a continuous solution domain into a finite number of elements, effectively handling complex physical problems. In cigarette packaging inkjet printing, this model comprehensively considers the hydrodynamic properties of the ink, the adsorption effect of the substrate material on the ink, and the influence of printhead motion on ink ejection. The idle artifact offset error compensation matrix is ​​a data matrix obtained in previous steps used to compensate for printing errors. By introducing it into the kinetic evolution model, an ink diffusion offset correction equation is established. The least squares method is used to fit the mapping relationship between the idle artifact offset error compensation matrix and the feature correlation heatmap, thereby adaptively adjusting the position and shape of the ink diffusion region in the heatmap. For example, if the ink diffusion region shifts during cigarette packaging inkjet printing due to printhead motion or ink diffusion characteristics, the ink diffusion offset correction equation can accurately adjust the ink diffusion region in the heatmap based on the information in the compensation matrix. This correction improves the accuracy of the feature correlation heatmap, providing a more reliable data foundation for subsequent analysis and prediction. The corrected heatmap sequence more accurately reflects the actual situation of ink diffusion, helping to accurately analyze ink diffusion behavior, promptly identify potential coding quality problems, provide accurate basis for optimizing coding processes, and thus improve the stability and consistency of coding quality.

[0153] Step S2320: The viscosity response curve of ink droplets on the substrate surface is acquired in real time. The relative contribution rate of adsorption effect and the relative contribution rate of ink potential energy driving effect are combined to adaptively modify the dynamic evolution model of ink diffusion behavior, and the modified dynamic evolution model is obtained.

[0154] Specifically, by deploying a microfluidic sensor array, the viscosity response curves of ink droplets on the substrate surface are acquired in real time, enabling real-time monitoring of viscosity changes in the ink droplets on the substrate surface. The relative contribution rates of adsorption and ink potential energy driving effects reflect the relative importance of adsorption and ink potential energy driving effects in the ink diffusion process. During the inkjet printing process on cigarette packaging paper, the viscosity of the ink may change due to environmental factors or interactions with the substrate material. For example, when the humidity of the packaging paper changes, the viscosity of the ink will change, thus affecting ink diffusion. The contact angle boundary conditions of the solid-liquid interface in the kinetic evolution model are dynamically updated based on the relative contribution rate of the adsorption effect, the volumetric force load distribution is corrected based on the relative contribution rate of the ink potential energy driving effect, and the adsorption-diffusion coupled field theory is introduced to establish a multi-scale multi-field collaborative analysis framework. This allows the kinetic evolution model to more accurately describe the ink diffusion process, taking into account the real-time changes of various factors. By adaptively correcting the model, the diffusion state of ink at different times can be predicted more accurately, providing a more accurate model basis for subsequent diffusion trajectory prediction. This helps to improve the control accuracy of ink diffusion during the coding process, reduce errors in the coding pattern, and improve coding quality.

[0155] Step S2330: Based on the modified kinetic evolution model, construct a dynamic prediction model for ink diffusion behavior, input the corrected feature correlation heat map sequence into the dynamic prediction model for ink diffusion behavior, and predict the probability distribution cloud map of ink diffusion profile within a predetermined time in the future.

[0156] Specifically, the dynamic prediction model for ink diffusion behavior employs a long short-term memory (LSTM) neural network to spatiotemporally model the heatmap evolution process. By designing a reasonable network structure and loss function, it extracts the spatiotemporal correlation patterns inherent in the ink diffusion process, enabling the prediction of future ink diffusion trends. The LSTM neural network can effectively process time-series data and remember long-term dependencies. In cigarette packaging inkjet printing, by inputting a corrected feature correlation heatmap sequence, the model can learn the temporal and spatial variations in ink diffusion. For example, the model can learn the speed, direction, and changes in ink diffusion influenced by printhead movement and substrate material at different times. Training with the previously obtained feature correlation heatmap sequence allows the model to better fit the actual ink diffusion situation. The model's output is a probability distribution cloud map of the ink diffusion profile within a predetermined future timeframe, such as 3ms. This cloud map shows the possible diffusion range and probability distribution of the ink within a short period. By accurately predicting the probability distribution cloud map of ink diffusion profile, we can understand the trend of ink diffusion in advance, providing a basis for real-time adjustments during the coding process, such as adjusting the movement trajectory of the printhead or the amount of ink ejected in advance, to ensure the accuracy of the coding pattern, reduce coding errors, and improve coding quality and production efficiency.

[0157] Step S2340: Perform morphological analysis on the probability distribution cloud map of ink diffusion contour, extract key deformation feature points of ink diffusion edge, and construct an ink diffusion edge deformation prediction grid containing n1 nodes.

[0158] Specifically, a multi-scale curvature analysis algorithm is used to perform morphological analysis on the probability distribution cloud map of ink diffusion contours, extracting key deformation feature points at the ink diffusion edge and adaptively determining the number and distribution of these key feature points. Simultaneously, a B-spline curve fitting method is introduced to generate a predictive mesh using these key feature points as control vertices, and the mesh topology is optimized, implementing adaptive mesh refinement in areas of rapid ink diffusion. The multi-scale curvature analysis algorithm, by calculating the curvature of curves at different scales, can accurately capture local and global shape changes of curves, thereby extracting key deformation feature points at the ink diffusion edge. In cigarette packaging inkjet printing, the shape changes of the ink diffusion edge reflect the dynamic process of diffusion; by extracting key deformation feature points, the boundary changes of ink diffusion can be better described. The B-spline curve fitting method can generate smooth curves using these key feature points as control vertices to construct a predictive mesh. Adaptive mesh refinement increases the mesh density in areas of rapid ink diffusion based on the speed and trend of ink diffusion, to more accurately describe the diffusion situation in that area. For example, on cigarette packaging, if the ink spreads rapidly in a certain area, mesh refinement can more accurately track the leading edge of ink diffusion, improving the prediction accuracy of the diffusion profile. This method helps to more accurately predict the boundary and shape changes of ink diffusion, providing precise boundary information for subsequently generating high-precision real-time diffusion trajectory prediction results. It enhances the ability to finely control the coding profile, ensuring that the edges of the coding pattern are clear and accurate, meeting the requirements for fine-grained control of coding quality.

[0159] Step S2350: The ink diffusion profile probability distribution cloud map and the ink diffusion edge deformation prediction grid are fused to generate real-time diffusion trajectory prediction results for fine control of inkjet printing profile.

[0160] Specifically, a graph cut algorithm is used to adaptively threshold-segment the ink diffusion contour probability distribution cloud map to extract the main ink diffusion contour. Based on this, the ink diffusion edge deformation prediction mesh is mapped to the main ink diffusion contour. An affine transformation is used to synchronize mesh deformation with contour deformation, resulting in a high-precision real-time diffusion trajectory prediction. The spatial resolution of the prediction result is optimized to 0.01 pixels, meeting the requirements for fine-grained control of inkjet printing quality. The graph cut algorithm is an image segmentation technique that separates the target region from the background by finding the optimal segmentation boundary in the image, thus extracting the main ink diffusion contour. In cigarette packaging inkjet printing, accurately extracting the main ink diffusion contour is crucial for controlling the shape of the inkjet pattern. Mapping the previously generated prediction mesh to the main contour and synchronizing mesh deformation with contour deformation through an affine transformation accurately simulates the actual trajectory of ink diffusion. An affine transformation is a linear transformation that preserves the "straightness" and "parallelism" of the image. By adjusting the transformation parameters, the mesh can accurately deform as the ink diffusion contour changes. For example, when the ink diffusion contour changes, the mesh can adjust synchronously, accurately reflecting the real-time situation of ink diffusion. The high-precision real-time diffusion trajectory prediction results obtained through these operations, with spatial resolution optimized to 0.01 pixels, can provide precise control information for the inkjet printer. This allows the inkjet printer to adjust the movement of the printhead and the ink ejection in real time based on the prediction results, ensuring the accuracy and quality of the inkjet pattern. This meets the stringent requirements for fine pattern control in cigarette packaging inkjet printing, reduces the deformation and error of the inkjet pattern, and improves the appearance quality and readability of the product.

[0161] Step S3000: Based on the dynamic weight allocation function of the compensation matrix and the real-time diffusion trajectory prediction results, an artifact compensation control signal is generated. The artifact compensation control signal is coupled to the nozzle drive circuit to perform dynamic feedback control on the motion posture of the nozzle.

[0162] Further, step S3000 includes:

[0163] Step S3100: Based on the real-time diffusion trajectory prediction results and the dynamic weight allocation function of the compensation matrix, the estimated value of the inkjet printing error is obtained.

[0164] Further, step S3100 includes:

[0165] Step S3110: Based on the real-time diffusion trajectory prediction results, extract the curvature data of the ink diffusion edge curve and construct the ink diffusion contour morphology description matrix.

[0166] Step S3120: Perform feature matching between the ink diffusion profile morphology description matrix and the preset printing pattern, calculate the Euclidean distance between the ink diffusion morphology and the geometric profile of the printing pattern, and obtain the morphology matching index.

[0167] Step S3130: Input the morphological matching index into the dynamic weight allocation function of the compensation matrix, superimpose the weight coefficients corresponding to the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic data sub-matrix, and the nozzle motion trajectory spatial feature sub-matrix, calculate the spatial geometric offset compensation amount, and generate the inkjet printing error prediction value.

[0168] Specifically, in step S3110, the real-time diffusion trajectory prediction result includes real-time information on the diffusion of ink on cigarette packaging paper. This information is obtained through previous steps, such as constructing a multi-scale gradient feature pyramid, performing feature fusion, and establishing a dynamic prediction model. The curvature data of the ink diffusion edge curve reflects the degree of bending and the trend of change of the ink diffusion edge, which is crucial for describing the morphology of the ink diffusion profile. For example, during the inkjet printing of cigarette packaging paper, the ink diffusion edge may exhibit varying degrees of bending, and curvature data can quantify these bending conditions. By extracting this curvature data, an ink diffusion profile morphology description matrix is ​​constructed. The elements in the matrix can be curvature values ​​at different locations or parameters related to curvature. Such a matrix can present the morphological information of the ink diffusion profile in a structured data form, facilitating subsequent comparative analysis with a preset printing pattern. The beneficial effect of this step is that by accurately extracting curvature data to construct a description matrix, the morphological features of the ink diffusion profile can be captured more accurately. Compared to simple profile description methods, matrix description based on curvature data can reflect the changes in the ink diffusion edge in greater detail. When performing feature matching with the preset printing pattern in the subsequent process, it can provide more accurate information, improve the accuracy of the matching, and thus more accurately assess the coding error. This provides more reliable data support for subsequent compensation operations, helps improve the quality of the coding pattern, and reduces coding defects caused by ink diffusion patterns that do not match expectations.

[0169] In step S3120, feature matching is a commonly used technique in computer vision to compare the similarity between two objects or images. In this step, by performing feature matching between the ink diffusion contour morphology description matrix and the preset printing pattern, the degree of difference between the actual shape of the current ink diffusion and the ideal printing pattern can be determined. Euclidean distance is a commonly used method to measure the distance between two vectors. When calculating the Euclidean distance between the ink diffusion shape and the geometric contour of the printing pattern, the ink diffusion contour morphology description matrix and the geometric contour information of the printing pattern are converted into vector form, and the difference between the two is quantified by calculating the Euclidean distance between the vectors. For example, assuming the preset printing pattern is a regular rectangle, while the actual ink diffusion contour exhibits an irregular shape due to various factors (such as uneven ink diffusion, printhead movement deviation, etc.), a numerical value can be obtained by calculating the Euclidean distance between the two. The smaller the value, the more similar the ink diffusion shape is to the geometric contour of the printing pattern, i.e., the higher the morphology matching degree; conversely, the larger the value, the greater the difference. The obtained morphology matching degree index is a quantitative value used to intuitively evaluate the degree of conformity between the ink diffusion shape and the expected pattern during the coding process. The benefit of this step is that the shape matching index provides a quantitative standard for evaluating coding errors. This index clearly shows whether ink diffusion during the coding process meets expectations, thus judging the quality of the coding. In practical applications, when the shape matching index exceeds a certain threshold, problems in the coding process, such as printhead clogging or abnormal ink supply, can be detected promptly. This helps to quickly pinpoint the root cause of the problem, take corresponding measures for adjustment and improvement, ensure the stability of coding quality, reduce scrap rates, and improve production efficiency.

[0170] In step S3130, the dynamic weight allocation function of the compensation matrix is ​​established based on previous steps. It can adaptively adjust the compensation matrix according to the real-time changes in fluid dynamic parameters and material property parameters during the printing process. The fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic data sub-matrix, and the printhead motion trajectory spatial feature sub-matrix respectively contain key information affecting ink diffusion, the effect of the substrate material on the ink, and the printhead motion state. The morphological matching index is input into the dynamic weight allocation function of the compensation matrix, and the calculation is performed in combination with the weight coefficients corresponding to the above three sub-matrices. For example, when the morphological matching index is low, indicating a large difference between the ink diffusion morphology and the printed pattern, the adjustment range of the spatial geometric offset compensation will be increased according to the calculation rules of the dynamic weight allocation function of the compensation matrix. The spatial geometric offset compensation calculated in this way considers multiple factors affecting the coding error, and can more accurately reflect the amount of error that needs to be compensated in the actual coding process, thereby generating a coding error prediction value. The beneficial effect of this step is that the coding error prediction value calculated by comprehensively considering multiple factors is more accurate and reliable. By analyzing and integrating multiple key factors, the causes of errors during the coding process can be captured more comprehensively, thus allowing for more accurate prediction of the error amount. Accurate error prediction provides a precise basis for generating artifact compensation control signals, facilitating more precise coding error compensation, effectively improving coding quality, ensuring that the coding pattern on cigarette packaging meets design requirements, reducing product appearance defects caused by coding errors, and enhancing overall product quality and market competitiveness.

[0171] Step S3200: Based on the inkjet printing error prediction value and combined with the no-load artifact offset error compensation matrix, generate an artifact compensation control signal.

[0172] Specifically, step S3200 aims to generate an artifact compensation control signal based on the inkjet error prediction and the idle artifact offset error compensation matrix, providing control instructions for subsequent adjustment of the printhead motion pose to compensate for inkjet errors. First, the inkjet error prediction and the idle artifact offset error compensation matrix are fused to construct a composite error compensation tensor. The inkjet error prediction reflects the estimated deviation from the preset pattern caused by factors such as ink diffusion during the current inkjet printing process, while the idle artifact offset error compensation matrix is ​​a data set that compensates for artifact offset errors that may occur during inkjet printing in an idle state. The fusion of these two to construct the composite error compensation tensor can comprehensively consider error factors under different conditions and more comprehensively describe the inkjet error situation. For example, when inkjet printing on cigarette packaging paper, the inkjet error prediction may show that ink diffusion causes local blurring of the pattern, while the idle artifact offset error compensation matrix records the offset information that may be caused by the slight vibration of the printhead during idle conditions. Combining the two allows for a more accurate grasp of the overall error.

[0173] Next, based on the numerical values ​​of different elements in the composite error compensation tensor, the weight allocation scheme for each control node in the printhead drive circuit is determined, generating a weight allocation matrix. The elements in the composite error compensation tensor contain error information of different positions and types. By analyzing these elements, the contribution of each control node to the error compensation is determined, thereby assigning corresponding weights to each control node in the printhead drive circuit. For example, if at a certain position the coding error is mainly caused by the horizontal offset of the printhead, then in the weight allocation matrix, the nodes related to the horizontal movement control of the printhead will have relatively larger weights. This weight allocation scheme allows for targeted adjustment of the control signals of the printhead drive circuit according to the specific error conditions.

[0174] Finally, the standardized electronic control pulse signal is modulated using a weighting matrix to synthesize a composite pulse sequence with injected high-frequency micro-perturbations, forming the artifact compensation control signal. The standardized electronic control pulse signal is the basic control signal for the printhead drive circuit. By modulating it with the weighting matrix, the amplitude, frequency, and other parameters of the pulse signal can be adjusted according to the error compensation requirements. The composite pulse sequence with injected high-frequency micro-perturbations can more precisely control the printhead's movement. For example, when compensating for minor printhead offsets, the high-frequency micro-perturbations can allow the printhead to adjust its position more accurately, avoiding new errors caused by over- or under-compensation. The beneficial effect of this step is that the generated artifact compensation control signal can comprehensively consider multiple error factors and achieve precise control of the printhead movement by accurately adjusting the control signal of the printhead drive circuit. This precise control helps to more effectively compensate for coding errors, improving the accuracy and clarity of the coding pattern. In cigarette packaging coding, it can ensure that the printed brand logo, production date, and other information are clear and accurate, reducing product quality problems caused by coding errors, improving the product's appearance quality and readability, and enhancing the product's competitiveness in the market. Meanwhile, by taking into account various error factors for compensation, the inkjet printer's adaptability to different working conditions is improved. Even when faced with cigarette packaging paper of different materials or slight performance changes in the printhead, the stability of the inkjet printing quality can be guaranteed.

[0175] In step S3300, the artifact compensation control signal is coupled to the nozzle drive circuit to perform dynamic feedback control on the movement and posture of the nozzle.

[0176] Specifically, the purpose of step S3300 is to couple the artifact compensation control signal to the printhead drive circuit to perform dynamic feedback control on the printhead's motion and pose, thereby achieving real-time correction of printing errors and improving printing quality. The artifact compensation control signal is generated through previous steps, comprehensively considering the predicted printing error, the idle artifact offset error compensation matrix, and multiple factors affecting the printing process. This signal contains instructions for precisely adjusting the printhead's motion.

[0177] The printhead drive circuit is a key component controlling the printhead's movement. It receives artifact compensation control signals and converts them into actual driving actions, thereby adjusting the printhead's posture. In the inkjet printing process on cigarette packaging, the printhead's posture directly affects the ink ejection position and pattern formation. For example, even a slight deviation in the horizontal or vertical direction can lead to inaccurate positioning or shape distortion of the printed pattern. By coupling the artifact compensation control signal to the printhead drive circuit, the printhead's movement can be dynamically adjusted based on real-time error conditions.

[0178] Dynamic feedback control is a closed-loop control process that continuously adjusts the printhead movement based on the current coding error to reduce it. During coding, the actual situation of the coding pattern is continuously monitored (e.g., by acquiring real-time coding images through an image acquisition device), compared with a preset ideal pattern, and the error is calculated. An artifact compensation control signal is generated based on the error and fed back to the printhead drive circuit to adjust the printhead's motion posture. The process is repeated, monitoring the coding pattern again, to continuously optimize the printhead movement and gradually reduce the coding error. For example, if a portion of the coding pattern is found to be offset, dynamic feedback control will adjust the printhead's direction and speed based on the artifact compensation control signal, ensuring that subsequent ink jets accurately land at the expected position, thus correcting the pattern offset.

[0179] Step S3300 uses dynamic feedback to control the printhead's movement and posture, enabling real-time correction of errors during the coding process and significantly improving the accuracy and quality of the printed patterns. In cigarette packaging coding, this ensures the accuracy of printed text and patterns, enhancing the product's appearance and brand image. Simultaneously, dynamic feedback control improves the adaptability and stability of the coding machine. Even when encountering various interference factors during coding (such as minor printhead vibrations or slight changes in ink properties), the printhead movement can be adjusted promptly to ensure unaffected coding quality. Furthermore, this precise control method reduces scrap rates due to coding errors, lowers production costs, increases production efficiency, and brings better economic benefits to enterprises.

[0180] Example 2

[0181] This embodiment, based on Embodiment 1, provides a real-time image calibration system for inkjet printers based on edge computing, such as... Figure 7 As shown, it includes:

[0182] The compensation matrix generation module is used to construct a composite parameter matrix, perform no-load error analysis, generate a no-load artifact offset error compensation matrix, establish a nonlinear mapping relationship between each element of the no-load artifact offset error compensation matrix and the composite parameter matrix, and generate a dynamic weight allocation function for the compensation matrix.

[0183] Diffusion trajectory prediction module: Based on the composite parameter matrix, a multi-scale gradient feature pyramid is constructed, which includes a bottom layer, a middle layer, and a top layer; the bottom, middle, and top layer features of the multi-scale gradient feature pyramid are fused using tensors to generate a feature correlation heatmap covering different abstraction levels; combined with the empty artifact offset error compensation matrix, the feature correlation heatmap is corrected, and a dynamic prediction model for ink diffusion behavior is established to generate real-time diffusion trajectory prediction results;

[0184] Error compensation module: Based on the dynamic weight allocation function of the compensation matrix and the real-time diffusion trajectory prediction results, it generates an artifact compensation control signal, which is then coupled to the nozzle drive circuit to perform dynamic feedback control on the nozzle's motion posture.

[0185] In the compensation matrix generation module, constructing the composite parameter matrix includes:

[0186] Step S1110: Obtain the surface viscosity coefficient and potential energy gradient value to form a sub-matrix of fluid diffusion field parameters;

[0187] Step S1120: Measure the adsorption decay characteristic curves of ink on different material surfaces, extract the adsorption decay time constant and saturated adsorption concentration threshold, and form a sub-matrix of substrate material adsorption characteristic data.

[0188] Step S1130: Track the nozzle movement process, acquire the nozzle vibration displacement signal, extract the spatial curvature change features of the nozzle movement trajectory from the nozzle vibration displacement signal, and form a spatial feature sub-matrix of the nozzle movement trajectory.

[0189] Step S1140: Construct a composite parameter matrix based on the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic data sub-matrix, and the nozzle motion trajectory spatial characteristic sub-matrix.

[0190] In the compensation matrix generation module, the dynamic weight allocation function for generating the compensation matrix includes:

[0191] Step S1610: The relative contribution rate of adsorption effect, the relative contribution rate of ink potential energy driving effect, the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic sub-matrix, and the printhead motion trajectory spatial feature sub-matrix are used as input parameters to construct a composite input vector.

[0192] Step S1620: Learn the nonlinear mapping relationship between the composite input vector and each element of the empty artifact offset error compensation matrix using a neural network method;

[0193] Step S1630: Generate a dynamic weight allocation function for the compensation matrix based on the learned nonlinear mapping relationship.

[0194] In the diffusion trajectory prediction module, the construction of the multi-scale gradient feature pyramid includes:

[0195] Step S2110: Based on the fluid diffusion field parameter sub-matrix, extract the multi-scale gradient invariant feature clusters formed during the diffusion of ink on the substrate surface, and construct the bottom layer of the multi-scale gradient feature pyramid; the multi-scale gradient invariant feature clusters include a 64-dimensional directional gradient histogram.

[0196] Step S2120: Based on the adsorption characteristic data sub-matrix of the substrate material, capture the micro-porosity distribution mapping of the substrate material surface and construct the middle layer of the multi-scale gradient feature pyramid.

[0197] Step S2130: Based on the spatial feature submatrix of the nozzle motion trajectory, analyze the second derivative fluctuation features of the nozzle motion trajectory and construct the top layer of the multi-scale gradient feature pyramid.

[0198] In the diffusion trajectory prediction module, generating feature correlation heatmaps covering different levels of abstraction includes:

[0199] Step S2210: The multi-scale gradient invariant feature clusters of the bottom layer are fused with the micro-porosity distribution mapping of the middle layer to obtain the local ink diffusion behavior characteristics.

[0200] Step S2220: The second derivative fluctuation characteristics of the top-level nozzle motion trajectory are fused with the local ink diffusion behavior characteristics to obtain the correlation matrix between ink distribution and nozzle motion.

[0201] Step S2230: Perform thermodynamic encoding on the correlation matrix between ink distribution and printhead motion to generate a feature correlation heatmap that reflects the correlation strength between the local feature distribution of the bottom layer and the global feature of the top layer.

[0202] In the error compensation module, the artifact compensation control signal includes:

[0203] Based on the real-time diffusion trajectory prediction results and the dynamic weight allocation function of the compensation matrix, the inkjet printing error prediction value is obtained; based on the inkjet printing error prediction value, combined with the no-load artifact offset error compensation matrix, an artifact compensation control signal is generated.

[0204] The estimated value of the inkjet printing error includes:

[0205] Step S3110: Based on the real-time diffusion trajectory prediction results, extract the curvature data of the ink diffusion edge curve and construct the ink diffusion contour morphology description matrix.

[0206] Step S3120: Perform feature matching between the ink diffusion profile morphology description matrix and the preset printing pattern, calculate the Euclidean distance between the ink diffusion morphology and the geometric profile of the printing pattern, and obtain the morphology matching index.

[0207] Step S3130: Input the morphological matching index into the dynamic weight allocation function of the compensation matrix, superimpose the weight coefficients corresponding to the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic data sub-matrix, and the nozzle motion trajectory spatial feature sub-matrix, calculate the spatial geometric offset compensation amount, and generate the inkjet printing error prediction value.

[0208] The methods and systems of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated.

[0209] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0210] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A real-time image calibration method for inkjet printers based on edge computing, characterized in that, The method includes: Construct a composite parameter matrix, perform no-load error analysis, generate a no-load artifact offset error compensation matrix, establish a nonlinear mapping relationship between each element of the no-load artifact offset error compensation matrix and the composite parameter matrix, and generate a dynamic weight allocation function for the compensation matrix. Based on the composite parameter matrix, a multi-scale gradient feature pyramid is constructed, which includes a bottom layer, a middle layer, and a top layer. The bottom, middle, and top features of the multi-scale gradient feature pyramid are fused using tensors to generate a feature correlation heatmap covering different abstraction levels. The feature correlation heatmap is corrected by combining the empty artifact offset error compensation matrix, and a dynamic prediction model for ink diffusion behavior is established to generate real-time diffusion trajectory prediction results. Based on the dynamic weight allocation function of the compensation matrix and the real-time diffusion trajectory prediction results, an artifact compensation control signal is generated. This artifact compensation control signal is then coupled to the nozzle drive circuit to perform dynamic feedback control on the nozzle's motion and pose.

2. The real-time image calibration method for inkjet printers based on edge computing according to claim 1, characterized in that, The construction of the composite parameter matrix includes: Obtain the surface viscosity coefficient and potential energy gradient values ​​to form a sub-matrix of fluid diffusion field parameters; The adsorption decay characteristic curves of ink on different material surfaces were measured, and the adsorption decay time constant and saturation adsorption concentration threshold were extracted to form a sub-matrix of substrate material adsorption characteristic data. The nozzle movement process is tracked to obtain the nozzle vibration displacement signal. The spatial curvature change features of the nozzle movement trajectory are extracted from the nozzle vibration displacement signal to form a spatial feature sub-matrix of the nozzle movement trajectory. A composite parameter matrix is ​​constructed based on the fluid diffusion field parameter sub-matrix, the substrate material adsorption characteristic data sub-matrix, and the nozzle motion trajectory spatial characteristic sub-matrix.

3. The real-time image calibration method for inkjet printers based on edge computing according to claim 2, characterized in that, The formation of the fluid diffusion field parameter sub-matrix includes: collecting dynamic contact angle change data of ink droplets on the substrate surface and extracting the surface viscosity coefficient; measuring the spatial difference of pressure distribution inside the ink droplets and calculating the potential energy gradient value; and quantifying the surface viscosity coefficient and potential energy gradient value into a fluid diffusion field parameter sub-matrix.

4. The real-time image calibration method for inkjet printers based on edge computing according to claim 3, characterized in that, The generated idle artifact offset error compensation matrix includes: Based on the fluid diffusion field parameter sub-matrix, the artifact width distribution data formed during the ink diffusion process is obtained, and the diffusion artifact width distribution spectrum is generated. The diffusion artifact width distribution spectrum is used as a priori constraint condition for the printing start stage under no-load conditions. It is loaded into a lightweight reverse generative adversarial topology network in a pre-built edge computing node to obtain the diffusion artifact prior constraint. Based on the fluid diffusion field parameter sub-matrix and the substrate material adsorption characteristic data sub-matrix, the relative contribution rate of the adsorption effect and the relative contribution rate of the ink potential energy driving effect are obtained. By combining the prior constraints of diffusion artifacts, error analysis is performed on the inkjet printing image under no-load conditions to generate the no-load artifact offset error compensation matrix.

5. The real-time image calibration method for inkjet printers based on edge computing according to claim 4, characterized in that, The step of combining diffusion artifact prior constraints to perform error analysis on the inkjet printing image under no-load conditions and generating an no-load artifact offset error compensation matrix includes: By combining diffusion artifact prior constraints, a preset reference pattern is printed using an inkjet printer under no-load conditions, and the actual inkjet image under no-load conditions is collected. Feature point matching is performed between the acquired actual inkjet image and the reference pattern to calculate the geometric offset of the inkjet outline; The geometric offset is represented by a matrix to generate an empty artifact offset error compensation matrix.

6. The real-time image calibration method for inkjet printers based on edge computing according to claim 5, characterized in that, The calculation of the geometric offset of the inkjet printing profile includes: Extract key feature points from the actual inkjet image and the reference pattern, and construct a feature point matching matrix; Based on the feature point matching matrix, the geometric offset function of the inkjet printing profile is fitted to obtain the local geometric offset and the global geometric offset. The local geometric offset and the global geometric offset are combined to form the geometric offset of the inkjet printing profile.

7. The real-time image calibration method for inkjet printers based on edge computing according to claim 6, characterized in that, The step of representing the geometric offset in a matrix form to generate the empty artifact offset error compensation matrix includes: Based on the relative contribution rate of the adsorption effect and the relative contribution rate of the ink potential energy driving effect, the local geometric offset and the overall geometric offset are weighted and combined to obtain the comprehensive geometric offset matrix. The elements in the comprehensive geometric offset matrix are normalized and discretized to generate the empty artifact offset error compensation matrix.

8. The real-time image calibration method for inkjet printers based on edge computing according to claim 2, characterized in that, The construction of the multi-scale gradient feature pyramid includes: Based on the fluid diffusion field parameter sub-matrix, multi-scale gradient invariant feature clusters formed during the diffusion of ink on the substrate surface are extracted, and the bottom layer of the multi-scale gradient feature pyramid is constructed; the multi-scale gradient invariant feature clusters include a 64-dimensional directional gradient histogram. Based on the adsorption characteristic data sub-matrix of the substrate material, the micro porosity distribution mapping of the substrate material surface is captured, and the middle layer of the multi-scale gradient feature pyramid is constructed. Based on the spatial feature submatrix of the nozzle motion trajectory, the fluctuation characteristics of the second derivative of the nozzle motion trajectory are analyzed, and the top layer of the multi-scale gradient feature pyramid is constructed.

9. The real-time image calibration method for inkjet printers based on edge computing according to claim 8, characterized in that, The generation of feature correlation heatmaps covering different levels of abstraction includes: By fusing the multi-scale gradient invariant feature clusters of the bottom layer with the micro-porosity distribution mapping of the middle layer, local ink diffusion behavior characteristics are obtained. By fusing the second derivative fluctuation characteristics of the top-level nozzle motion trajectory with the local ink diffusion behavior characteristics, the correlation matrix between ink distribution and nozzle motion is obtained. The correlation matrix between ink distribution and printhead motion is thermodynamically encoded to generate a feature correlation heatmap that reflects the correlation strength between the local feature distribution of the bottom layer and the global feature of the top layer.

10. A real-time image calibration system for inkjet printers based on edge computing, used to implement the real-time image calibration method for inkjet printers based on edge computing as described in any one of claims 1-9, characterized in that, The system includes: The compensation matrix generation module is used to construct a composite parameter matrix, perform no-load error analysis, generate a no-load artifact offset error compensation matrix, establish a nonlinear mapping relationship between each element of the no-load artifact offset error compensation matrix and the composite parameter matrix, and generate a dynamic weight allocation function for the compensation matrix. Diffusion trajectory prediction module: Based on the composite parameter matrix, a multi-scale gradient feature pyramid is constructed, which includes a bottom layer, a middle layer, and a top layer; the bottom, middle, and top layer features of the multi-scale gradient feature pyramid are fused using tensors to generate a feature correlation heatmap covering different abstraction levels; combined with the empty artifact offset error compensation matrix, the feature correlation heatmap is corrected, and a dynamic prediction model for ink diffusion behavior is established to generate real-time diffusion trajectory prediction results; Error compensation module: Based on the dynamic weight allocation function of the compensation matrix and the real-time diffusion trajectory prediction results, it generates an artifact compensation control signal, which is then coupled to the nozzle drive circuit to perform dynamic feedback control on the nozzle's motion posture.

Images