Machine learning based closed loop adaptive image transfer system and method

Abstract

The application relates to the technical field of industrial printing and surface treatment, and provides a closed-loop self-adaptive image transfer system and method based on machine learning. The method comprises the following steps: obtaining a target digital image; extracting color distribution, texture features and pigment density features of the target digital image; generating initial printing parameters through a machine learning prediction model according to material optical properties, ink absorption properties and equipment nozzle configuration, ink properties and operation parameters; performing an image transfer operation on the surface of the material according to the initial printing parameters, realizing integrated and seamless transfer of six surfaces and internal hole positions of the stone; collecting actual image results after the transfer is completed; comparing the actual image results with the target digital image, calculating color difference and structural similarity, and generating error data; automatically adjusting the printing parameters according to the error data, generating corrected printing parameters; inputting the error data and the corrected printing parameters into the machine learning prediction model, and updating parameters of the machine learning prediction model.

Description

Technical Field

[0001] This application relates to the field of industrial printing and surface treatment technology, and in particular to a closed-loop adaptive image transfer system and method based on machine learning. Background Technology

[0002] In the fields of industrial printing and surface treatment, digital image transfer technology is widely used in industries such as architectural decoration and home building materials to reproduce designed patterns on the surfaces of natural stone, quartz, artificial stone, ceramics, and composite materials. Currently, the industry generally uses a combination of static ICC color profiles and manual color adjustment to achieve image transfer: technicians pre-create standard color curves for specific materials and printing equipment, convert the digital image into color data that the printing equipment can recognize, and then manually observe the differences between the printed result and the target image, adjusting the printing parameters to correct the deviation.

[0003] The aforementioned existing technology has the following insurmountable drawbacks: First, static color curves can only reflect the color mapping relationship between materials and equipment under specific standard conditions. They cannot adapt to the inherent differences in ink absorption, reflectivity, surface texture, etc. of different batches of materials, nor can they compensate for output fluctuations caused by equipment aging and changes in ambient temperature and humidity, resulting in a continuous deviation between digital images and actual transfer results.

[0004] Secondly, manual color matching relies heavily on the experience and subjective judgment of operators, resulting in long cycles, low efficiency, and significant differences in results among different operators, making it difficult to guarantee consistency in mass production. Furthermore, manual color matching cannot quantify color and structural deviations, making precise local corrections impossible.

[0005] Third, existing technologies employ an open-loop control mode, where printing parameters remain unchanged throughout the production process once determined, lacking an automatic feedback correction mechanism based on actual transfer results. When deviations occur, production must be paused for manual adjustments, severely impacting production efficiency.

[0006] Fourth, existing technologies lack self-learning capabilities and cannot accumulate transfer printing experience under different materials, equipment, and environmental conditions. Each time materials or equipment are changed, color curves need to be remade and colors need to be manually adjusted, which increases production and time costs.

[0007] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0008] This application provides a closed-loop adaptive image transfer system and method based on machine learning, which aims to solve the problems that static color curves can only reflect the color mapping relationship between materials and equipment under specific standard conditions, and cannot adapt to the inherent differences in ink absorption, reflectivity, surface texture, etc. of different batches of materials, nor can they compensate for the output fluctuations caused by equipment aging and changes in ambient temperature and humidity, resulting in a continuous deviation between digital images and actual transfer results.

[0009] In a first aspect, embodiments of this application provide a closed-loop adaptive image transfer method based on machine learning, the method comprising: Acquire the target digital image; extract the color distribution, texture features, and pigment density features of the target digital image; Based on the preset material optical properties, ink absorption properties, as well as the equipment printhead configuration, ink properties, and operating parameters, the initial printing parameters are generated through a machine learning prediction model; an image transfer operation is performed on the material surface according to the initial printing parameters; and the actual image result after the transfer is completed is collected. The actual image result is compared with the target digital image to calculate color difference and structural similarity, generating error data; the printing parameters are automatically adjusted based on the error data to generate corrected printing parameters; the error data and corrected printing parameters are input into the machine learning prediction model to update the parameters of the machine learning prediction model.

[0010] In some embodiments, acquiring the target digital image includes: receiving a digital image file in a preset image format; converting the received digital image file to a format to generate a target digital image in a uniform format; and performing noise removal and resolution unification processing on the target digital image.

[0011] In some embodiments, the extracted color distribution, texture features, and pigment density features of the target digital image include: extracting the red, green, and blue channel values ​​of each pixel in the target digital image; converting them to generate color distribution data for each pixel; extracting texture feature data of the target digital image; performing region segmentation on the target digital image; calculating the pigment deposition density of each segmented region; and forming pigment density feature data.

[0012] In some embodiments, generating initial printing parameters using a machine learning prediction model based on preset material optical properties, ink absorption properties, and device printhead configuration, ink properties, and operating parameters includes: retrieving material parameters for the corresponding material from a pre-stored material database, the material parameters including optical properties, ink absorption properties, and surface treatment parameters; retrieving device parameters for the corresponding device from a pre-stored device database, the device parameters including printhead configuration, ink properties, operating speed, operating temperature, and operating pressure; inputting the extracted image features and the retrieved material and device parameters into the machine learning prediction model; and generating an initial color map, initial printing speed, initial printing temperature, and initial printing pressure using the machine learning prediction model.

[0013] In some embodiments, performing the image transfer operation on the material surface according to the initial printing parameters includes: converting the color data of the target digital image according to the initial color map; controlling the operation of the printing equipment according to the initial printing speed, initial printing temperature, and initial printing pressure; and depositing ink layer by layer on the material surface to complete the image transfer.

[0014] In some embodiments, acquiring the actual image result after the transfer is completed includes: taking a picture of the material surface after the transfer is completed to obtain visible light data of the actual image; scanning the material surface after the transfer is completed to obtain spectral data of the actual image; and integrating the visible light data and spectral data into the actual image result.

[0015] In some embodiments, comparing the actual image result with the target digital image, calculating color difference and structural similarity, and generating error data includes: aligning the actual image result with the target digital image pixel by pixel; calculating the color difference value of the corresponding pixel; calculating the brightness similarity, contrast similarity, and structural similarity between the actual image result and the target digital image; generating difference distribution data based on the color difference value and structural similarity, and marking areas where the difference exceeds a preset threshold.

[0016] In some embodiments, the step of automatically adjusting printing parameters based on error data to generate corrected printing parameters includes: determining the printing parameter deviation corresponding to each difference region based on difference distribution data; adjusting the corresponding value of the color map table for each difference region; adjusting the printing speed, printing temperature, and printing pressure of the corresponding region; and integrating all adjusted parameters to generate corrected printing parameters.

[0017] In some embodiments, the step of inputting error data and corrected printing parameters into a machine learning prediction model to update the parameters of the machine learning prediction model includes: using error data and corrected printing parameters as training samples; inputting the training samples into the machine learning prediction model; adjusting the weights of the machine learning prediction model in conjunction with a pre-established physical model; and completing the parameter update of the machine learning prediction model.

[0018] Secondly, this application provides a closed-loop adaptive image transfer system based on machine learning, used to implement the method provided in any embodiment of this application, characterized in that it includes: The image acquisition unit is used to acquire the target digital image and extract the color distribution, texture features, and pigment density features of the target digital image. The result acquisition unit is used to generate initial printing parameters based on preset material optical properties, ink absorption properties, as well as equipment printhead configuration, ink properties, and operating parameters, through a machine learning prediction model; perform image transfer operation on the material surface according to the initial printing parameters; and acquire the actual image result after the transfer is completed. The model update unit is used to compare the actual image results with the target digital image, calculate the color difference and structural similarity, and generate error data; automatically adjust the printing parameters based on the error data to generate corrected printing parameters; and input the error data and corrected printing parameters into the machine learning prediction model to update the parameters of the machine learning prediction model.

[0019] This application achieves closed-loop adaptive control of the entire image transfer process. By automatically collecting transfer results, calculating errors, correcting parameters, and updating the model, deviation correction can be completed without manual intervention, significantly improving the accuracy of image transfer and the consistency of batch production. By employing a machine learning prediction model to generate printing parameters, it can comprehensively learn the influence of material optical properties, ink absorption characteristics, equipment configuration, and operating parameters on the transfer effect, effectively compensating for differences in materials from different batches and changes in equipment status, eliminating the need for frequent recalibration. The feedback learning mechanism based on actual transfer results allows the model to continuously accumulate experience and optimize parameters, with transfer accuracy continuously improving with increased usage, achieving system self-evolution. It completely eliminates reliance on operator experience, significantly shortens production preparation time, improves production efficiency, reduces production costs, and is suitable for large-scale industrial production.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic flowchart illustrating the steps of a closed-loop adaptive image transfer method based on machine learning, provided in an embodiment of this application. Figure 2 This is a schematic block diagram of a closed-loop adaptive image transfer system based on machine learning, provided in one embodiment of this application. Figure 3 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application.

[0023] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0026] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0027] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0028] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0029] In the fields of industrial printing and surface treatment, digital image transfer technology is widely used in industries such as architectural decoration and home building materials to reproduce designed patterns on the surfaces of natural stone, quartz, artificial stone, ceramics, and composite materials. Currently, the industry generally uses a combination of static ICC color profiles and manual color adjustment to achieve image transfer: technicians pre-create standard color curves for specific materials and printing equipment, convert the digital image into color data that the printing equipment can recognize, and then manually observe the differences between the printed result and the target image, adjusting the printing parameters to correct the deviation.

[0030] The aforementioned existing technology has the following insurmountable drawbacks: First, static color curves can only reflect the color mapping relationship between materials and equipment under specific standard conditions. They cannot adapt to the inherent differences in ink absorption, reflectivity, surface texture, etc. of different batches of materials, nor can they compensate for output fluctuations caused by equipment aging and changes in ambient temperature and humidity, resulting in a continuous deviation between digital images and actual transfer results.

[0031] Secondly, manual color matching relies heavily on the experience and subjective judgment of operators, resulting in long cycles, low efficiency, and significant differences in results among different operators, making it difficult to guarantee consistency in mass production. Furthermore, manual color matching cannot quantify color and structural deviations, making precise local corrections impossible.

[0032] Third, existing technologies employ an open-loop control mode, where printing parameters remain unchanged throughout the production process once determined, lacking an automatic feedback correction mechanism based on actual transfer results. When deviations occur, production must be paused for manual adjustments, severely impacting production efficiency.

[0033] Fourth, existing technologies lack self-learning capabilities and cannot accumulate transfer printing experience under different materials, equipment, and environmental conditions. Each time materials or equipment are changed, color curves need to be remade and colors need to be manually adjusted, which increases production and time costs.

[0034] Therefore, a method is urgently needed to solve at least one of the above problems.

[0035] To solve the above problem, please refer to Figure 1This application provides a closed-loop adaptive image transfer method based on machine learning, applied to computer equipment. The computer equipment can be deployed on a single server or server cluster. It can also be deployed on handheld terminals, laptops, wearable devices, or robots, etc. It should be noted that all information involved in the method provided in this application is extracted with the authorization of the relevant user and in accordance with relevant regulations, and will not infringe on user privacy.

[0036] The provided machine learning-based closed-loop adaptive image transfer method includes steps S101 to S103. Details are as follows: Step S101. Acquire the target digital image; extract the color distribution, texture features and pigment density features of the target digital image.

[0037] Specifically, the core function of this step is to convert the digital image to be transferred into standardized feature data that can be recognized by machine learning prediction models, providing input for the generation of subsequent printing parameters.

[0038] The target digital image acquisition process receives user-submitted digital image files for transfer and performs unified preprocessing on image files of different formats to generate target digital images that meet the system's processing requirements. The preprocessing eliminates image noise, standardizes image resolution, and ensures the accuracy and consistency of subsequent feature extraction.

[0039] Image feature extraction involves performing multi-dimensional feature analysis on the preprocessed target digital image to extract three core features: Color distribution characteristics reflect the overall and local color composition information of an image, and are a key basis for ensuring color reproduction after transfer. Texture features reflect the surface texture, edge contours, and detail information of an image, and are used to guide parameter adjustments for different texture regions during the printing process; Pigment density characteristics reflect the pigment deposition requirements in different areas of the image and are used to control the amount of ink ejected and the deposition thickness during printing.

[0040] Step S102. Based on the preset material optical properties, ink absorption properties, and equipment printhead configuration, ink properties, and operating parameters, generate initial printing parameters through a machine learning prediction model; perform image transfer operation on the material surface according to the initial printing parameters to achieve seamless integrated transfer of the six sides of the stone and its internal holes; collect the actual image results after the transfer is completed; the initial transfer process parameters include: pattern output resolution 200-400 DPI, printing ink volume 50%-300%, vacuum heating furnace temperature 100-300℃, and transfer time 5-60 minutes.

[0041] Specifically, the core function of this step is to generate optimal initial printing parameters based on the extracted image features, combined with the inherent characteristics of the materials and equipment, through a machine learning prediction model, and control the printing equipment to complete the image transfer operation.

[0042] Initial printing parameters are generated by retrieving the characteristic parameters of the material to be transferred and the operating parameters of the printing equipment from a pre-built material database and equipment database, respectively. These parameters, along with the image features extracted in step S101, are then input into a pre-trained machine learning prediction model. The model automatically generates initial printing parameters adapted to the current image, material, and equipment by learning the non-linear mapping relationship between input parameters and output effects from a large amount of historical transfer data.

[0043] Image transfer is performed by sending the generated initial printing parameters to the printing equipment, which then controls the equipment to deposit ink on the material surface according to the preset parameters, completing the transfer process from digital image to material surface. During printing, key operating indicators such as printing speed, temperature, and pressure are strictly controlled according to the initial parameters to ensure the stability of the transfer process. The initial transfer process parameters include: pattern output resolution of 200-400 DPI, ink volume of 50%-300%, vacuum furnace heating temperature of 100-300℃, and transfer time of 5-60 minutes.

[0044] After the transfer is completed, a high-precision acquisition device is used to comprehensively collect the actual transfer results on the material surface, obtaining actual image data containing information such as color, texture, and structure, providing a basis for subsequent comparative analysis and parameter correction.

[0045] Step S103. Compare the actual image result with the target digital image, calculate the color difference and structural similarity, and generate error data; automatically adjust the printing parameters according to the error data to generate corrected printing parameters; input the error data and corrected printing parameters into the machine learning prediction model to update the parameters of the machine learning prediction model.

[0046] Specifically, this step is the core of achieving closed-loop adaptive control. By comparing the difference between the actual transfer result and the target image, the printing parameters are automatically corrected, and the correction experience is fed back to the machine learning prediction model to achieve continuous self-optimization of the system.

[0047] Results comparison and error analysis involve precisely aligning and comparing the acquired actual image results with the original target digital image in multiple dimensions. The color difference and structural similarity between the two are calculated to generate quantified error data. This error data visually reflects the degree and distribution of deviations in color and structure of the transfer results.

[0048] The automatic printing parameter correction system analyzes the causes of deviations based on the generated error data and determines the corresponding adjustment amount for each deviation area. For different deviation areas, it adjusts the color mapping relationship and equipment operating parameters separately to generate corrected printing parameters. These corrected parameters can be directly used for the next transfer of the same type of image, or for adjusting parameters on products in the current batch that have not yet completed the transfer process.

[0049] The machine learning prediction model is updated by using the error data generated during the current transfer process and the corrected printing parameters as new training samples, which are then input into the machine learning prediction model for incremental learning. The model adjusts its own weights in conjunction with a pre-established physical model, continuously optimizing the mapping relationship between input and output, and improving the accuracy of subsequent printing parameter generation. As the number of transfers increases, the model's prediction accuracy will continue to improve, and the system's adaptive capability will also be continuously enhanced.

[0050] In some embodiments, acquiring the target digital image includes: receiving a digital image file in a preset image format; converting the received digital image file to a format to generate a target digital image in a uniform format; and performing noise removal and resolution unification processing on the target digital image.

[0051] This embodiment details the specific implementation of obtaining the target digital image in step S101, including the following steps: Image file reception: The computer device receives digital image files submitted by the user via a wired or wireless communication interface, supporting the reception of digital image files in preset image formats. For portable document format files, the system automatically extracts the image content contained therein and converts it into processable raster image data.

[0052] Unified Format Conversion: This function converts received digital image files of different formats into the system's internal standard format. The conversion process preserves the original color and resolution information of the image, avoiding image quality loss due to format conversion. For portable network graphics images with alpha channels, the system automatically fills the alpha channels with a white background to ensure consistency in subsequent processing.

[0053] Noise Removal Processing: The converted image undergoes noise removal processing. A Gaussian filtering algorithm is used to smooth the image and eliminate Gaussian noise generated during image acquisition. A median filtering algorithm is used to remove salt-and-pepper noise from the image. The size of the filtering window is automatically adjusted according to the original resolution of the image. A larger filtering window is used for high-resolution images, and a smaller filtering window is used for low-resolution images, so as to preserve the edge and detail information of the image to the maximum extent while removing noise.

[0054] Resolution unification: The noise-removed images are uniformly adjusted to a preset standard resolution. During the adjustment process, a bicubic interpolation algorithm is used for image scaling to ensure good smoothness and clarity in the scaled image. Images with resolutions higher than the standard resolution are downsampled; images with resolutions lower than the standard resolution are upsampled. The size of the image after resolution unification matches the maximum print size of the printing equipment, avoiding image stretching or cropping during subsequent printing.

[0055] In some embodiments, the extracted color distribution, texture features, and pigment density features of the target digital image include: extracting the red, green, and blue channel values ​​of each pixel in the target digital image; converting them to generate color distribution data for each pixel; extracting texture feature data of the target digital image; performing region segmentation on the target digital image; calculating the pigment deposition density of each segmented region; and forming pigment density feature data.

[0056] This embodiment details the specific implementation of step S101, which involves extracting the color distribution, texture features, and pigment density features of the target digital image, including the following steps: Color distribution feature extraction: Extract the red, green, and blue channel values ​​of each pixel in the target digital image, with each channel ranging from 0 to 255, forming an original color data matrix. Convert the original color data matrix into luminance, red-green, and yellow-blue channel color space data. Standard color space conversion formulas are used during the conversion process to ensure accuracy. Statistical analysis is performed on the converted luminance, red-green, and yellow-blue channel data, calculating the mean, variance, histogram distribution, and other statistical characteristics of each channel to form complete color distribution data.

[0057] Texture feature extraction: Edge information of the target digital image is extracted using an edge detection operator to generate a binary edge image, marking the position of all edge pixels in the image; the gradient magnitude and gradient direction of each pixel in the target digital image are calculated to generate a gradient magnitude image and a gradient direction image, reflecting the intensity and direction of gray-level changes in the image; the texture direction features of the image are extracted using the gray-level co-occurrence matrix method, the gray-level co-occurrence matrix in different directions is calculated, and texture statistical features such as contrast, correlation, energy, and entropy are extracted to form texture feature data.

[0058] Pigment density feature extraction: A cluster-based image segmentation algorithm is used to segment the target digital image into several regions with uniform gray levels. During segmentation, the number of clusters is automatically determined based on the image's gray-level distribution to ensure the accuracy of the segmentation results. The average gray-level value of each segmented region is calculated and mapped to the corresponding pigment deposition density. The lower the gray-level value, the higher the corresponding pigment deposition density; the higher the gray-level value, the lower the corresponding pigment deposition density. A correspondence table between segmented regions and pigment deposition densities is generated, forming pigment density feature data.

[0059] In some embodiments, generating initial printing parameters using a machine learning prediction model based on preset material optical properties, ink absorption properties, and device printhead configuration, ink properties, and operating parameters includes: retrieving material parameters for the corresponding material from a pre-stored material database, the material parameters including optical properties, ink absorption properties, and surface treatment parameters; retrieving device parameters for the corresponding device from a pre-stored device database, the device parameters including printhead configuration, ink properties, operating speed, operating temperature, and operating pressure; inputting the extracted image features and the retrieved material and device parameters into the machine learning prediction model; and generating an initial color map, initial printing speed, initial printing temperature, and initial printing pressure using the machine learning prediction model.

[0060] This embodiment details the specific implementation of step S102, which involves generating initial printing parameters using a machine learning prediction model based on preset material optical properties, ink absorption properties, printhead configuration, ink properties, and operating parameters. The specific implementation includes: Material Parameter Retrieval: Retrieves the characteristic parameters of the material to be transferred from a pre-stored material database, including: Optical properties: reflectivity, transmittance, and gloss parameters of the material surface; Ink Absorption Properties: ink absorption speed, ink penetration depth, and ink diffusion coefficient parameters of the material; Surface Treatment Parameters: surface roughness, smoothness, and pretreatment process parameters of the material surface. The material database stores characteristic parameters of various commonly used materials, including natural stone, quartz, artificial stone, ceramics, and composite materials. Users can select the corresponding parameters according to the actual material type being used.

[0061] Device parameter retrieval: Retrieves the printing device's operating parameters from a pre-stored device database, including: printhead configuration: number of printheads, nozzle diameter, nozzle spacing, and jet frequency parameters; ink characteristics: ink color, viscosity, surface tension, and drying speed parameters; operating parameters: the device's default printing speed, printing temperature, and printing pressure range. The device database stores parameter information for various printing device models, and the system can automatically identify the connected printing device model and retrieve the corresponding parameters.

[0062] The color distribution features, texture features, and pigment density features extracted in step S101 are standardized along with the retrieved material parameters and equipment parameters, converting them into input vectors recognizable by the machine learning prediction model. During standardization, all parameters are mapped to a numerical range of 0 to 1, eliminating dimensional differences between different parameters. Initial printing parameter generation: The standardized input vector is input into the pre-trained machine learning prediction model. The model calculates and outputs initial printing parameters through forward propagation, including: initial color map: establishing the mapping relationship between the target image color space and the printing equipment color space; initial printing speed: controlling the movement speed of the print head; initial printing temperature: controlling the heating temperature of the print head and the preheating temperature of the material surface; initial printing pressure: controlling the ink jet pressure.

[0063] In some embodiments, performing the image transfer operation on the material surface according to the initial printing parameters includes: converting the color data of the target digital image according to the initial color map; controlling the operation of the printing equipment according to the initial printing speed, initial printing temperature, and initial printing pressure; and depositing ink layer by layer on the material surface to complete the image transfer.

[0064] This embodiment details the specific implementation of step S102, which involves performing an image transfer operation on the material surface based on initial printing parameters. This includes: converting the color value of each pixel of the target digital image into ink jet volume data recognizable by the printing device based on the generated initial color map. During the conversion process, the ink absorption characteristics of the material and the ink layering effect are considered, and the ink jet volume for different colors is adjusted to ensure the accuracy of the transferred color. The initial printing speed, initial printing temperature, and initial printing pressure are sent to the controller of the printing device, which adjusts the device's operating status according to the received parameters. After setting, the system performs a self-check to confirm that all parameters are within the normal operating range. The material to be transferred is placed on the working platform of the printing device, and the material is precisely positioned using a positioning device to ensure that the material's position is consistent with the printing coordinate system. After positioning, the material is fixed to the working platform using vacuum adsorption or mechanical clamping to prevent movement during printing. The printing device deposits ink layer by layer on the material surface according to the converted ink jet volume data and the set operating parameters. For images that require multi-layer printing, the system automatically controls the amount of ink deposited and the drying time of each layer, ensuring that the previous layer of ink is completely dry before printing the next layer, thus avoiding ink mixing and dripping.

[0065] After printing is complete, the system uses sensors to detect the ink drying status on the material surface. Once the ink is confirmed to be completely dry, the system automatically removes the material from the work platform, completing the image transfer operation.

[0066] In some embodiments, acquiring the actual image result after the transfer is completed includes: taking a picture of the material surface after the transfer is completed to obtain visible light data of the actual image; scanning the material surface after the transfer is completed to obtain spectral data of the actual image; and integrating the visible light data and spectral data into the actual image result.

[0067] This embodiment details the specific implementation of step S102, which involves acquiring the actual image result after the transfer is completed. This includes: calibrating the acquisition device before acquiring the transfer result. A standard color chart is used to calibrate the industrial camera to ensure the accuracy of the color data acquired by the camera; a standard white board is used to perform baseline calibration on the spectrometer to eliminate the influence of ambient light on the spectral measurement. The surface of the transferred material is photographed to obtain the visible light data of the actual image. The resolution of the industrial camera is not lower than the printing resolution of the printing equipment. Uniform light sources are used for illumination during the shooting process to avoid shadows and reflections. After shooting, distortion correction is performed on the acquired image to eliminate image distortion caused by camera lens distortion.

[0068] A spectrometer is used to scan the surface of the transferred material point by point to acquire spectral data of the actual image. The spectrometer's wavelength range covers the visible light band, with a wavelength resolution of no less than 5 nanometers. During the scanning process, the scanning step size of the spectrometer is matched with the pixel size of the industrial camera to ensure that each pixel location has corresponding spectral data.

[0069] The acquired visible light data and spectral data are spatially registered to ensure a one-to-one correspondence between visible light pixel data and spectral data at the same location. After registration, the two types of data are fused to generate an actual image result containing both color and spectral information. The fused actual image result can more comprehensively and accurately reflect the transfer effect.

[0070] In some embodiments, comparing the actual image result with the target digital image, calculating color difference and structural similarity, and generating error data includes: aligning the actual image result with the target digital image pixel by pixel; calculating the color difference value of the corresponding pixel; calculating the brightness similarity, contrast similarity, and structural similarity between the actual image result and the target digital image; generating difference distribution data based on the color difference value and structural similarity, and marking areas where the difference exceeds a preset threshold.

[0071] This embodiment details the specific implementation of step S103, which involves comparing the actual image result with the target digital image, calculating color difference and structural similarity, and generating error data. The method includes: employing an image registration algorithm based on feature point matching to align the actual image result with the target digital image pixel-by-pixel. First, feature points are extracted from both images. Then, a transformation relationship between the two images is established through feature point matching. Finally, a geometric transformation is performed on the actual image based on this transformation relationship to ensure complete spatial alignment with the target digital image.

[0072] After alignment, the color difference values ​​of corresponding pixels in the two images are calculated. The pixel values ​​of the two images are converted into a color space with luminance, red-green, and yellow-blue channels. The difference between each pixel in the three channels is calculated, and then the total color difference for each pixel is calculated using the color difference calculation formula. The larger the color difference value, the greater the color deviation at that pixel location.

[0073] The structural similarity between the actual image and the target digital image is calculated, comprising three parts: brightness similarity, contrast similarity, and structural similarity. Brightness similarity is calculated by comparing the average brightness values ​​of the two images; contrast similarity is calculated by comparing the standard deviations of the brightness values ​​of the two images; and structural similarity is calculated by comparing the normalized cross-covariance of the two images. The weighted average of these three similarities yields the overall structural similarity score. A structural similarity score closer to 1 indicates greater structural similarity between the two images.

[0074] Error data generation involves calculating the color difference of each pixel and the overall structural similarity to produce difference distribution data. This difference distribution data is presented as an image, with different colors representing different degrees of deviation. Simultaneously, the system automatically marks regions where the difference exceeds a preset threshold, generating a list of deviation regions to provide a basis for subsequent parameter adjustments.

[0075] In some embodiments, the step of automatically adjusting printing parameters based on error data to generate corrected printing parameters includes: determining the printing parameter deviation corresponding to each difference region based on difference distribution data; adjusting the corresponding value of the color map table for each difference region; adjusting the printing speed, printing temperature, and printing pressure of the corresponding region; and integrating all adjusted parameters to generate corrected printing parameters.

[0076] This embodiment details the specific implementation of step S103, which involves automatically adjusting printing parameters based on error data to generate corrected printing parameters. This includes: parameter deviation analysis, which analyzes the deviation type and degree of each deviation region based on the generated difference distribution data and deviation region list. For regions with large color deviations, it determines which color channel deviation is causing the issue; for regions with large structural deviations, it determines whether the deviation is due to printing speed, temperature, or pressure.

[0077] Color map adjustments target areas of color deviation, adjusting the mapping relationship of corresponding color values ​​in the color map. If the red in a certain area is too dark, the ink ejection volume of the red channel in that area is reduced; if the blue in a certain area is too light, the ink ejection volume of the blue channel in that area is increased. Linear interpolation is used during the adjustment process to ensure a smooth transition in the color map and avoid color banding.

[0078] For areas with structural deviations, adjust the printing speed, temperature, and pressure accordingly. If the texture in a certain area is not clear enough, appropriately reduce the printing speed and increase the printing temperature and pressure to increase ink deposition and penetration depth. If ink diffusion occurs in a certain area, appropriately increase the printing speed and decrease the printing temperature and pressure to reduce ink diffusion.

[0079] The parameter integration process combines all adjusted color map parameters and device operating parameters to generate revised printing parameters. These revised parameters include global and local parameters. Global parameters apply to the entire image, while local parameters apply only to the corresponding deviation areas. The system automatically performs a validity check on the revised parameters to ensure that all parameters are within the normal operating range of the printing device.

[0080] In some embodiments, the step of inputting error data and corrected printing parameters into a machine learning prediction model to update the parameters of the machine learning prediction model includes: using error data and corrected printing parameters as training samples; inputting the training samples into the machine learning prediction model; adjusting the weights of the machine learning prediction model in conjunction with a pre-established physical model; and completing the parameter update of the machine learning prediction model.

[0081] This embodiment details the specific implementation of step S103, which involves inputting error data and corrected printing parameters into the machine learning prediction model to update the model's parameters. This includes: organizing the input data (image features, material parameters, equipment parameters), error data, and corrected printing parameters generated during the transfer process into a new training sample. The input portion of the training sample consists of image features, material parameters, and equipment parameters; the output portion consists of the corrected printing parameters; and the label portion consists of the error data.

[0082] The newly constructed training samples are standardized to ensure they conform to the format of the sample data used during model training. Simultaneously, the sample data is cleaned to remove outliers and noisy data, guaranteeing the quality of the training samples.

[0083] The preprocessed training samples are input into the machine learning prediction model for incremental learning. During training, the backpropagation algorithm is used to calculate the model's loss function, which consists of two parts: one part is the error between the model output and the corrected printed parameters, and the other part is the error between the predicted error corresponding to the model output and the actual error. The model's weight parameters are adjusted by minimizing the loss function.

[0084] During model training, a pre-established physical model is used to constrain the model's output. This physical model, based on the principles of ink diffusion, material ink absorption, and the working principles of printing equipment, can predict the transfer effect under given printing parameters. By inputting the model's output into the physical model, the error between the predicted and actual transfer effects is calculated, and this error is added to the loss function. This ensures the model's learning process conforms to physical laws, improving the model's generalization ability.

[0085] After training is complete, the updated machine learning prediction model parameters are saved. The system also retains the model parameters from previous versions, so that it can automatically roll back to a previous stable version when the performance of the new model degrades.

[0086] Please see Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of a machine learning-based closed-loop adaptive image transfer system 200 provided in this application embodiment. The machine learning-based closed-loop adaptive image transfer system 200 is used to execute the steps of the machine learning-based closed-loop adaptive image transfer method shown in the above embodiments. The machine learning-based closed-loop adaptive image transfer system 200 can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, a laptop computer, a wearable device, or a robot.

[0087] like Figure 2 As shown, the machine learning-based closed-loop adaptive image transfer system 200 includes: The image acquisition unit 201 is used to acquire a target digital image and extract the color distribution, texture features, and pigment density features of the target digital image. The result acquisition unit 202 is used to generate initial printing parameters based on preset material optical properties, ink absorption properties, equipment printhead configuration, ink properties, and operating parameters through a machine learning prediction model; perform image transfer operation on the material surface according to the initial printing parameters to achieve seamless integrated transfer of the six sides of the stone and its internal holes; and acquire the actual image result after the transfer is completed. The initial transfer process parameters include: pattern output resolution of 200-400 DPI, printing ink volume of 50%-300%, vacuum heating furnace temperature of 100-300℃, and transfer time of 5-60 minutes. The model update unit 203 is used to compare the actual image result with the target digital image, calculate the color difference and structural similarity, and generate error data; automatically adjust the printing parameters according to the error data to generate corrected printing parameters; and input the error data and corrected printing parameters into the machine learning prediction model to update the parameters of the machine learning prediction model.

[0088] In some embodiments, acquiring the target digital image includes: receiving a digital image file in a preset image format; converting the received digital image file to a format to generate a target digital image in a uniform format; and performing noise removal and resolution unification processing on the target digital image.

[0089] In some embodiments, the extracted color distribution, texture features, and pigment density features of the target digital image include: extracting the red, green, and blue channel values ​​of each pixel in the target digital image; converting them to generate color distribution data for each pixel; extracting texture feature data of the target digital image; performing region segmentation on the target digital image; calculating the pigment deposition density of each segmented region; and forming pigment density feature data.

[0090] In some embodiments, generating initial printing parameters using a machine learning prediction model based on preset material optical properties, ink absorption properties, and device printhead configuration, ink properties, and operating parameters includes: retrieving material parameters for the corresponding material from a pre-stored material database, the material parameters including optical properties, ink absorption properties, and surface treatment parameters; retrieving device parameters for the corresponding device from a pre-stored device database, the device parameters including printhead configuration, ink properties, operating speed, operating temperature, and operating pressure; inputting the extracted image features and the retrieved material and device parameters into the machine learning prediction model; and generating an initial color map, initial printing speed, initial printing temperature, and initial printing pressure using the machine learning prediction model.

[0091] In some embodiments, performing the image transfer operation on the material surface according to the initial printing parameters includes: converting the color data of the target digital image according to the initial color map; controlling the operation of the printing equipment according to the initial printing speed, initial printing temperature, and initial printing pressure; and depositing ink layer by layer on the material surface to complete the image transfer.

[0092] In some embodiments, acquiring the actual image result after the transfer is completed includes: taking a picture of the material surface after the transfer is completed to obtain visible light data of the actual image; scanning the material surface after the transfer is completed to obtain spectral data of the actual image; and integrating the visible light data and spectral data into the actual image result.

[0093] In some embodiments, comparing the actual image result with the target digital image, calculating color difference and structural similarity, and generating error data includes: aligning the actual image result with the target digital image pixel by pixel; calculating the color difference value of the corresponding pixel; calculating the brightness similarity, contrast similarity, and structural similarity between the actual image result and the target digital image; generating difference distribution data based on the color difference value and structural similarity, and marking areas where the difference exceeds a preset threshold.

[0094] In some embodiments, the step of automatically adjusting printing parameters based on error data to generate corrected printing parameters includes: determining the printing parameter deviation corresponding to each difference region based on difference distribution data; adjusting the corresponding value of the color map table for each difference region; adjusting the printing speed, printing temperature, and printing pressure of the corresponding region; and integrating all adjusted parameters to generate corrected printing parameters.

[0095] In some embodiments, the step of inputting error data and corrected printing parameters into a machine learning prediction model to update the parameters of the machine learning prediction model includes: using error data and corrected printing parameters as training samples; inputting the training samples into the machine learning prediction model; adjusting the weights of the machine learning prediction model in conjunction with a pre-established physical model; and completing the parameter update of the machine learning prediction model.

[0096] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the machine learning-based closed-loop adaptive image transfer system and its modules described above can be found in the corresponding embodiments of the machine learning-based closed-loop adaptive image transfer method, and will not be repeated here.

[0097] The aforementioned machine learning-based closed-loop adaptive image transfer method can be implemented as a computer program, which can be used in various ways, such as... Figure 2 It runs on the device shown.

[0098] Please see Figure 3 , Figure 3 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application. The computer device includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and internal memory.

[0099] The storage medium may store operating devices and computer programs. The computer program includes program instructions that, when executed, cause the processor to perform any machine learning-based closed-loop adaptive image transfer method.

[0100] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0101] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When executed by a processor, the computer program enables the processor to perform any machine learning-based closed-loop adaptive image transfer method.

[0102] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the terminal to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0103] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0104] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: Acquire the target digital image; extract the color distribution, texture features, and pigment density features of the target digital image; Based on preset material optical properties, ink absorption properties, and equipment printhead configuration, ink properties, and operating parameters, initial printing parameters are generated through a machine learning prediction model. An image transfer operation is then performed on the material surface according to these initial printing parameters, achieving seamless integrated transfer of the stone's six sides and internal pores. The actual image results after the transfer are collected. The initial transfer process parameters include: pattern output resolution of 200-400 DPI, ink volume of 50%-300%, vacuum furnace heating temperature of 100-300℃, and transfer time of 5-60 minutes. The actual image result is compared with the target digital image to calculate color difference and structural similarity, generating error data; the printing parameters are automatically adjusted based on the error data to generate corrected printing parameters; the error data and corrected printing parameters are input into the machine learning prediction model to update the parameters of the machine learning prediction model.

[0105] In some embodiments, acquiring the target digital image includes: receiving a digital image file in a preset image format; converting the received digital image file to a format to generate a target digital image in a uniform format; and performing noise removal and resolution unification processing on the target digital image.

[0106] In some embodiments, the extracted color distribution, texture features, and pigment density features of the target digital image include: extracting the red, green, and blue channel values ​​of each pixel in the target digital image; converting them to generate color distribution data for each pixel; extracting texture feature data of the target digital image; performing region segmentation on the target digital image; calculating the pigment deposition density of each segmented region; and forming pigment density feature data.

[0107] In some embodiments, generating initial printing parameters using a machine learning prediction model based on preset material optical properties, ink absorption properties, and device printhead configuration, ink properties, and operating parameters includes: retrieving material parameters for the corresponding material from a pre-stored material database, the material parameters including optical properties, ink absorption properties, and surface treatment parameters; retrieving device parameters for the corresponding device from a pre-stored device database, the device parameters including printhead configuration, ink properties, operating speed, operating temperature, and operating pressure; inputting the extracted image features and the retrieved material and device parameters into the machine learning prediction model; and generating an initial color map, initial printing speed, initial printing temperature, and initial printing pressure using the machine learning prediction model.

[0108] In some embodiments, performing the image transfer operation on the material surface according to the initial printing parameters includes: converting the color data of the target digital image according to the initial color map; controlling the operation of the printing equipment according to the initial printing speed, initial printing temperature, and initial printing pressure; and depositing ink layer by layer on the material surface to complete the image transfer.

[0109] In some embodiments, acquiring the actual image result after the transfer is completed includes: taking a picture of the material surface after the transfer is completed to obtain visible light data of the actual image; scanning the material surface after the transfer is completed to obtain spectral data of the actual image; and integrating the visible light data and spectral data into the actual image result.

[0110] In some embodiments, comparing the actual image result with the target digital image, calculating color difference and structural similarity, and generating error data includes: aligning the actual image result with the target digital image pixel by pixel; calculating the color difference value of the corresponding pixel; calculating the brightness similarity, contrast similarity, and structural similarity between the actual image result and the target digital image; generating difference distribution data based on the color difference value and structural similarity, and marking areas where the difference exceeds a preset threshold.

[0111] In some embodiments, the step of automatically adjusting printing parameters based on error data to generate corrected printing parameters includes: determining the printing parameter deviation corresponding to each difference region based on difference distribution data; adjusting the corresponding value of the color map table for each difference region; adjusting the printing speed, printing temperature, and printing pressure of the corresponding region; and integrating all adjusted parameters to generate corrected printing parameters.

[0112] In some embodiments, the step of inputting error data and corrected printing parameters into a machine learning prediction model to update the parameters of the machine learning prediction model includes: using error data and corrected printing parameters as training samples; inputting the training samples into the machine learning prediction model; adjusting the weights of the machine learning prediction model in conjunction with a pre-established physical model; and completing the parameter update of the machine learning prediction model.

[0113] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the steps of the machine learning-based closed-loop adaptive image transfer method provided in any embodiment of this application.

[0114] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0115] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A closed-loop adaptive image transfer method based on machine learning, characterized in that, include: Acquire the target digital image; extract the color distribution, texture features, and pigment density features of the target digital image; Based on the preset material optical properties, ink absorption properties, as well as the equipment printhead configuration, ink properties, and operating parameters, the initial printing parameters are generated through a machine learning prediction model. Based on the initial printing parameters, an image transfer operation is performed on the material surface to achieve seamless integrated transfer of the six sides of the stone and the internal holes. Collect the actual image result after the transfer is completed; The initial transfer process parameters include: pattern output resolution of 200-400 DPI, printing ink volume of 50%-300%, vacuum heating oven temperature of 100-300℃, and transfer time of 5-60 minutes. The actual image result is compared with the target digital image to calculate color difference and structural similarity, generating error data; the printing parameters are automatically adjusted based on the error data to generate corrected printing parameters; the error data and corrected printing parameters are input into the machine learning prediction model to update the parameters of the machine learning prediction model.

2. The method according to claim 1, characterized in that, The acquisition of the target digital image includes: Receives digital image files in a preset image format; The received digital image files are converted to generate target digital images in a uniform format. The target digital image is subjected to noise removal and resolution unification processing.

3. The method according to claim 1, characterized in that, The extracted color distribution, texture features, and pigment density features of the target digital image include: Extract the red, green, and blue channel values ​​of each pixel in the target digital image; convert and generate color distribution data for each pixel; Extract texture feature data from the target digital image; The target digital image is segmented into regions, and the pigment deposition density of each segmented region is calculated to form pigment density feature data.

4. The method according to claim 1, characterized in that, The process of generating initial printing parameters based on preset material optical properties, ink absorption properties, printhead configuration, ink properties, and operating parameters using a machine learning prediction model includes: Retrieve the material parameters of the corresponding material from a pre-stored material database. The material parameters include optical properties, ink absorption properties, and surface treatment parameters. The device parameters of the corresponding device are retrieved from the pre-stored device database. These device parameters include printhead configuration, ink characteristics, operating speed, operating temperature, and operating pressure. The extracted image features, along with the retrieved material and equipment parameters, are input into the machine learning prediction model. The initial color map, initial printing speed, initial printing temperature, and initial printing pressure are generated using a machine learning prediction model.

5. The method according to claim 1, characterized in that, The step of performing an image transfer operation on the material surface according to the initial printing parameters includes: Convert the color data of the target digital image according to the initial color map table; The printing equipment is controlled based on the initial printing speed, initial printing temperature, and initial printing pressure. Ink is deposited layer by layer on the surface of the material to complete the image transfer.

6. The method according to claim 1, characterized in that, The actual image results after the acquisition and transfer are completed include: The surface of the transferred material is photographed to obtain visible light data of the actual image; The surface of the transferred material is scanned to obtain the spectral data of the actual image; Integrate visible light data and spectral data into actual image results.

7. The method according to claim 1, characterized in that, The step of comparing the actual image result with the target digital image, calculating color difference and structural similarity, and generating error data includes: Align the actual image result with the target digital image pixel by pixel; Calculate the color difference value of the corresponding pixel; Calculate the brightness similarity, contrast similarity, and structural similarity between the actual image result and the target digital image; Generate difference distribution data based on color difference values ​​and structural similarity, and mark areas where the difference exceeds a preset threshold.

8. The method according to claim 1, characterized in that, The automatic adjustment of printing parameters based on error data to generate corrected printing parameters includes: Based on the difference distribution data, determine the printing parameter deviation corresponding to each difference region; For each area of ​​difference, adjust the corresponding values ​​in the color map table; Adjust the printing speed, printing temperature, and printing pressure for the corresponding area; Integrate all adjusted parameters to generate corrected printing parameters.

9. The method according to claim 1, characterized in that, The step of inputting error data and corrected printing parameters into the machine learning prediction model to update the parameters of the machine learning prediction model includes: Use the error data and the corrected printing parameters as training samples; Input the training samples into the machine learning prediction model; The weights of the machine learning prediction model are adjusted based on the pre-established physical model. Complete the parameter update of the machine learning prediction model.

10. A closed-loop adaptive image transfer system based on machine learning, for implementing the method as described in any one of claims 1-9, characterized in that, include: The image acquisition unit is used to acquire the target digital image and extract the color distribution, texture features, and pigment density features of the target digital image. The result acquisition unit is used to generate initial printing parameters based on preset material optical properties, ink absorption properties, as well as equipment printhead configuration, ink properties, and operating parameters, through a machine learning prediction model. Based on the initial printing parameters, an image transfer operation is performed on the material surface to achieve seamless integrated transfer of the six sides of the stone and the internal holes. Collect the actual image result after the transfer is completed; The initial transfer process parameters include: pattern output resolution of 200-400 DPI, printing ink volume of 50%-300%, vacuum heating oven temperature of 100-300℃, and transfer time of 5-60 minutes. The model update unit is used to compare the actual image results with the target digital image, calculate the color difference and structural similarity, and generate error data; automatically adjust the printing parameters based on the error data to generate corrected printing parameters; and input the error data and corrected printing parameters into the machine learning prediction model to update the parameters of the machine learning prediction model.

Images