A method and system for generating photolithography elements for holographic laser printing

Abstract

The present invention provides a method and system for generating photolithography primitives for holographic laser printing, the method comprising the following steps: obtaining a plane line draft design drawing of a target product; using an edge recognition algorithm to identify the line edge features in the plane line draft design drawing after preprocessing, and generating and marking the line pattern features of the plane line draft design drawing according to the line edge features; extracting an image data set based on the line pattern features; combining a convolutional neural network and a support vector machine to construct a visual effect generation model; completing model training of the visual effect generation model through the image data set; inputting the plane line draft design drawing into the visual effect generation model that has completed model training, generating a photolithography visual effect filling scheme through the visual effect generation model; arranging a dot matrix grating in the plane line draft design drawing according to the photolithography visual effect filling scheme to form a target photolithography primitive corresponding to the target product. The present invention has the effect of quickly and stably generating photolithography primitives.

Description

Technical Field

[0001] The present invention belongs to the technical field of holographic laser printing, and in particular relates to a method and system for generating photolithographic image elements for holographic laser printing. Background Art

[0002] Holographic laser printing technology is widely used in anti-counterfeiting, packaging, art decoration and other fields. Its main function is to generate patterns with specific visual effects on the surface of materials through photolithography technology. These patterns not only have beautiful visual effects, but also can prevent counterfeiting to a certain extent. However, with the increasing market demand for personalized and complex patterns, traditional photolithography technology faces unprecedented challenges. Specifically, holographic laser printing involves complex pattern design and precise visual effect filling, which requires a high degree of detail processing and precise parameter control. Traditional methods rely on manual design, which is not only time-consuming, but also prone to human errors, resulting in unstable quality of the image elements. The limitations of the technology make it difficult for traditional methods to complete high-quality photolithography image element generation in a short time, thereby limiting the production efficiency and market competitiveness of the product. Summary of the invention

[0003] The present invention provides a method and system for generating photolithography elements for holographic laser printing, so as to solve the problems of low efficiency and unstable quality of photolithography element generation.

[0004] In a first aspect, the present invention provides a method for generating photolithographic primitives for holographic laser printing, the method comprising the following steps:

[0005] Obtain the graphic line drawing of the target product;

[0006] Preprocessing the plane line draft design drawing;

[0007] Using an edge recognition algorithm to identify line edge features in the preprocessed plane line draft design drawing, and generating and marking line pattern features of the plane line draft design drawing based on the line edge features;

[0008] Extracting an image data set from a preset pattern database based on the line pattern features, the image data set includes a plurality of sample image data, each of the sample image data includes a plurality of closed line areas, and each of the closed line areas of the sample image data is pre-filled and annotated with different types of lithography visual effects;

[0009] Combine convolutional neural networks and support vector machines to build a visual effect generation model;

[0010] Complete model training of the visual effect generation model through the image data set;

[0011] Inputting the plane line draft design drawing into the visual effect generation model that has completed model training, and generating a lithography visual effect filling scheme for all closed line areas in the plane line draft design drawing through the visual effect generation model;

[0012] According to the photolithography visual effect filling scheme, dot matrix gratings are arranged in the plane line draft design drawing to form target photolithography primitives corresponding to the target product.

[0013] Optionally, the step of combining a convolutional neural network and a support vector machine to construct a visual effect generation model comprises the following steps:

[0014] Building a visual effect filling sub-model based on a convolutional neural network, wherein the visual effect filling sub-model is used to generate visual effect filling parameters of an input image;

[0015] Building a lithography hotspot detection sub-model based on a support vector machine, wherein the lithography hotspot detection sub-model is used to detect the number of lithography hotspots in an input image;

[0016] The visual effect filling sub-model, the preset effect filler and the lithography hotspot detection sub-model are sequentially spliced ​​end to end to form a visual effect generation model. The effect filler is used to simulate the filling of the corresponding lithography visual effect on the closed line area in the input image according to the visual effect filling parameters.

[0017] Optionally, the visual effect filling sub-model includes a global feature extraction module, a local feature extraction module, a feature fusion module and a filling parameter generation module, the lithography hotspot detection sub-model includes a frequency domain feature extraction module, a support vector machine parameter configuration module and a support vector machine classification module, and there is a feedback adjustment mechanism between the lithography hotspot detection sub-model and the effect filler.

[0018] The step of completing the model training of the visual effect generation model through the image data set comprises the following steps:

[0019] Arrange all the sample image data in the image data set into a first image training set;

[0020] Identify closed line regions of all the sample image data in the image data set based on the line edge features;

[0021] Deleting the photolithography visual effects in the closed line areas of all the sample image data and retaining the annotation labels of the photolithography visual effects to obtain a plurality of sample line image data;

[0022] Integrate all the sample line image data into a second image training set;

[0023] Segmenting the closed line areas of all the sample line image data in the second image training set into sample image blocks by using an image segmentation algorithm, and integrating all the sample image blocks into an image block training set;

[0024] completing independent model training of the visual effect filling sub-model by combining the second image training set and the image block training set;

[0025] The independent model training of the lithography hotspot detection sub-model is completed using the first image training set.

[0026] Optionally, the step of combining the second image training set and the image block training set to complete the independent model training of the visual effect filling sub-model comprises the following steps:

[0027] Initializing module parameters of the global feature extraction module, the local feature extraction module, and the filling parameter generation module, and weight parameters of the feature fusion module;

[0028] Inputting the second image training set into the global feature extraction module, and using the global feature extraction module to extract global line pattern features of the sample line image data in the second image training set, wherein the global line pattern features include the number of closed line areas in the sample line image data, the region similarity, and the region spacing;

[0029] Inputting the image block training set into the local feature extraction module, and using the local feature extraction module to extract local line pattern features of the sample image blocks in the image block training set, wherein the local line pattern features include regional complexity, the number of regional line border endpoints, regional area and regional shape of the closed line area in the sample image block;

[0030] Inputting the global line pattern feature and the local line pattern feature into the feature fusion module, and fusing the global line pattern feature and the local line pattern feature into a multi-scale line pattern feature through the feature fusion module based on the weight parameter;

[0031] Inputting the multi-scale line pattern features into the filling parameter generation module, the filling parameter generation module randomly generates initial visual effect filling parameters according to the multi-scale line pattern features, the initial visual effect filling parameters including initial types of lithography visual effects corresponding to all the closed line areas in the sample line image data;

[0032] A cross entropy loss function is constructed in combination with the annotation label and the initial type of the lithography visual effect, and the module parameters and the weight parameters are optimized in a manner of minimizing the cross entropy loss function until the module parameters and the weight parameters reach optimal module parameters and optimal weight parameters, respectively.

[0033] Optionally, the step of completing the independent model training of the lithography hotspot detection sub-model by using the first image training set comprises the following steps:

[0034] Initializing the SVM parameters of the support vector machine classification module;

[0035] Inputting the first image training set into the frequency domain feature extraction module, and using the frequency domain feature extraction module and a discrete cosine transform method to extract sample frequency features of the sample image data in the first image training set;

[0036] The sample frequency features are input into the support vector machine parameter configuration module, and the sample frequency features and the SVM parameters are fused and encoded into a first chromosome code through the support vector machine parameter configuration module. The first chromosome code is optimally solved based on a genetic algorithm to obtain an optimal sample frequency feature subset and an optimal SVM parameter.

[0037] Optionally, the step of inputting the plane line draft design drawing into the visual effect generation model that has completed model training, and generating a lithography visual effect filling scheme for all closed line areas in the plane line draft design drawing through the visual effect generation model comprises the following steps:

[0038] Inputting the plane line draft design drawing into the effect filler and the visual effect filling sub-model that has completed model training, and using the visual effect filling sub-model to generate initial visual effect filling parameters of the plane line draft design drawing;

[0039] By means of the effect filler and according to the initial visual effect filling parameters of the plane line draft design drawing, all closed line areas in the plane line draft design drawing are simulated and filled with corresponding lithography visual effects to obtain a lithography primitive simulation image corresponding to the target product;

[0040] Inputting the photolithography primitive simulation image into the photolithography hotspot detection sub-model that has completed model training, and outputting the number of photolithography hotspots in the photolithography primitive simulation image through the photolithography hotspot detection sub-model;

[0041] If the number of the lithography hot spots is less than a preset number threshold, a lithography visual effect filling scheme for all closed line areas in the plane line draft design drawing is generated according to the initial visual effect filling parameters.

[0042] Optionally, the method further comprises the following steps:

[0043] If the number of the lithography hot spots is greater than or equal to the number threshold, encoding the initial visual effect filling parameter as a second chromosome code;

[0044] Adopting a genetic algorithm and using the lithography hotspot detection sub-model to iteratively update the second chromosome code until the number of the lithography hotspots is less than the number threshold, and taking the second chromosome code of the current iteration number as the target chromosome code;

[0045] Decoding the target chromosome into optimal visual effect filling parameters;

[0046] A lithography visual effect filling scheme is generated for all closed line areas in the plane line draft design drawing according to the optimal visual effect filling parameters.

[0047] Optionally, the step of using an edge recognition algorithm to identify line edge features in the preprocessed plane line draft design drawing, and generating and marking line pattern features of the plane line draft design drawing according to the line edge features comprises the following steps:

[0048] Using an edge recognition algorithm to identify line edge features in the pre-processed plane line drawing design;

[0049] Performing skeleton processing on the plane line draft design drawing based on the line edge features to obtain a line skeleton structure of the plane line draft design drawing;

[0050] Analyze and obtain line topological features in the line skeleton structure;

[0051] Using a feature extraction algorithm to extract line geometric features from the line edge features;

[0052] The line pattern features of the plane line draft design drawing are generated and marked by combining the line topological features and the line geometric features.

[0053] In a second aspect, the present invention further provides a photolithography element generation system for holographic laser printing, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the computer program.

[0054] The beneficial effects of the present invention are:

[0055] By introducing advanced image processing and machine learning technologies, the present invention can automatically identify and process complex line pattern features, significantly reducing manual intervention and human errors. In addition, the present invention improves the accuracy and consistency of photolithography primitive generation through optimized model training and parameter adjustment, thereby ensuring that the visual effect of each closed area meets the expected results. This automated process not only improves design efficiency, but also quickly responds to market demand for complex and personalized patterns, significantly improving production efficiency and product quality, making holographic laser printing more competitive in multiple application fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic flow chart of a method for generating photolithography elements for holographic laser printing in one embodiment of the present application.

[0057] Figure 2 A schematic diagram of the model structure of a visual effect generation model in one embodiment of the present application. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0059] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0060] Figure 1 FIG. 1 is a flow chart of a method for generating photolithographic primitives for holographic laser printing in one embodiment. It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1At least part of the steps in the above method may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps. Figure 1 As shown, the present invention discloses a method for generating photolithographic primitives for holographic laser printing, which specifically includes the following steps:.

[0061] S101. Obtain a planar line drawing of a target product.

[0062] Among them, it is necessary to ensure that the design is in a high-quality vector graphic format, such as SVG or AI files, so as to ensure the clarity and accuracy of the lines. The design can be scanned and vectorized by hand-drawing, or directly drawn using computer-aided design (CAD) software. In this step, it is necessary to consider the final effect of the target product and ensure that the layout, density and complexity of the lines meet the expected visual effects and technical requirements. For example, if the target product is a high-end packaging box, the design needs to contain fine lines and complex patterns to show the high-end and uniqueness of the product.

[0063] S102. Preprocess the graphic line drawing.

[0064] Among them, the preprocessing of the plane line draft design drawing is to optimize the image quality and simplify the subsequent processing steps. First, the design drawing needs to be denoised to remove possible noise and noise to ensure the clarity of the lines. This can be achieved using image processing techniques such as median filtering or Gaussian filtering. Next, the design drawing needs to be binarized to convert the image into black and white mode for subsequent edge recognition and feature extraction. Binarization can be performed by setting a threshold, setting pixels above the threshold to white and pixels below the threshold to black. Finally, morphological processing, such as dilation and erosion operations, is required to enhance the continuity and integrity of the lines. These processing steps can be implemented using open source image processing libraries such as OpenCV. The preprocessed design drawing will be easier to identify edges and extract features, ensuring the accuracy and efficiency of subsequent steps.

[0065] S103. Use an edge recognition algorithm to identify line edge features in the pre-processed plane line draft design drawing, and generate and annotate line pattern features of the plane line draft design drawing based on the line edge features.

[0066] Among them, the edge recognition algorithm can adopt the Canny edge detection algorithm, which identifies significant edges by calculating the gradient of the image. First, the pre-processed design drawing needs to be Gaussian blurred to reduce the interference of noise on edge detection. Next, the significant edges are identified by calculating the gradient amplitude and direction of the image. Then, the non-maximum suppression technology is used to refine the edge lines and remove the insignificant edges. Finally, the final edge image is obtained through double threshold processing and edge connection. Based on the identified edge features, skeletonization can be performed to simplify the lines into a single-pixel wide skeleton structure. By analyzing the topological features of the skeleton structure, such as endpoints, intersections, and branch points, the geometric features of the lines, such as length, curvature, and angle, can be extracted. Combining these topological features and geometric features, the line pattern features of the plane line draft design drawing can be generated and annotated, providing basic data for subsequent visual effect generation.

[0067] S104. Extracting an image data set from a preset pattern database based on line pattern features.

[0068] Among them, a large number of different types of sample image data are pre-stored in the pattern database, and each sample image data contains several closed line areas, which are pre-filled and annotated with different types of lithography visual effects, such as suspension effect, relief effect and microarray effect. First, based on the line pattern characteristics, such as the number of areas, area similarity and area spacing, sample image data similar to the design drawing is screened out from the database. Next, the screened sample image data needs to be further analyzed and matched to ensure that the shape, size and layout of its closed line area are consistent with the closed line area in the design drawing.

[0069] Specifically, shape descriptors can be used to quantitatively describe the shape features of each closed area. The similarity between the shape descriptors of each closed area in the design drawing and the closed area in the sample image data can be calculated using Euclidean distance or cosine similarity. A similarity threshold is set, and only matching pairs with similarity higher than the threshold are retained. Then the area and perimeter of each closed area in the design drawing and the sample image are calculated. The area and perimeter are normalized to eliminate the influence of image size differences. Compare the normalized area and perimeter, set an allowable error range (such as ±10%), and only retain matching pairs within the error range. For layout matching, the center of gravity of the image can be used as a reference point to calculate the relative position and direction of each closed area relative to the center of gravity. Construct a spatial relationship graph to represent the topological relationship between closed areas. Compare the spatial relationship graphs of the design drawing and the sample image, and use a graph matching algorithm (such as a graph isomorphism algorithm) to evaluate the layout similarity. Finally, weights are assigned to shape matching, size matching, and layout matching, and a comprehensive matching score is calculated. The sample image data is sorted according to the comprehensive matching score, and several samples with the highest scores are selected. In this way, an image dataset containing multiple sample image data can be extracted to provide training data for the subsequent visual effect generation model. The quality and diversity of the image dataset will directly affect the performance of the visual effect generation model, so it is necessary to ensure the richness of the pattern database and the representativeness of the sample image data.

[0070] S105. Combining convolutional neural networks and support vector machines to build a visual effect generation model.

[0071] Among them, the convolutional neural network (CNN) is a deep learning model that is good at processing image data, while the support vector machine (SVM) is a classic machine learning algorithm that is good at processing small samples and high-dimensional data. First, a visual effect filling sub-model is constructed based on the convolutional neural network, which is used to generate visual effect filling parameters for the input image. The visual effect filling sub-model includes a global feature extraction module, a local feature extraction module, a feature fusion module, and a filling parameter generation module. The global feature extraction module is used to extract the global line pattern features of the image, the local feature extraction module is used to extract the local line pattern features of the image, the feature fusion module is used to fuse the global features and local features into multi-scale features, and the filling parameter generation module is used to generate visual effect filling parameters based on the multi-scale features.

[0072] Next, a lithography hotspot detection sub-model is constructed based on the support vector machine. This sub-model is used to detect the number of lithography hotspots in the input image. The lithography hotspot detection sub-model includes a frequency domain feature extraction module, a support vector machine parameter configuration module, and a support vector machine classification module. The frequency domain feature extraction module is used to extract the frequency domain features of the image, the support vector machine parameter configuration module is used to configure the parameters of the support vector machine, and the support vector machine classification module is used to detect the number of lithography hotspots based on the frequency domain features and the support vector machine parameters. Finally, the visual effect filling sub-model, the preset effect filler, and the lithography hotspot detection sub-model are spliced ​​end to end to form a complete visual effect generation model. The effect filler is used to simulate the filling of the corresponding lithography visual effect on the closed line area in the input image according to the visual effect filling parameters.

[0073] S106. Complete model training of the visual effect generation model through the image dataset.

[0074] S107. Input the plane line draft design drawing into the visual effect generation model that has completed model training, and generate a lithography visual effect filling solution for all closed line areas in the plane line draft design drawing through the visual effect generation model.

[0075] S108. Arrange the dot lattice grating in the plane line draft design drawing according to the lithography visual effect filling scheme to form the target lithography element corresponding to the target product.

[0076] Among them, according to the generated lithography visual effect filling scheme, the lithography visual effect type and parameters of each closed line area are determined. Next, according to the lithography visual effect type and parameters, the layout method of the dot matrix grating is determined, including the density, arrangement method and angle of the dot matrix. Then, using the lithography equipment, the dot matrix grating is accurately laid out in the closed line area of ​​the plane line draft design to form a lithography primitive. The lithography equipment can use a laser etcher or an electron beam etcher, which has high precision and high resolution and can achieve fine lithography effects. In this way, lithography primitives with special visual effects, such as suspension effects, relief effects, and microarray effects, can be formed on the surface of the target product.

[0077] In one embodiment, referring to Figure 2 , combining convolutional neural network and support vector machine to build a visual effect generation model includes the following steps:

[0078] A visual effect filling sub-model is constructed based on a convolutional neural network, and the visual effect filling sub-model is used to generate visual effect filling parameters of an input image;

[0079] A lithography hotspot detection sub-model is constructed based on a support vector machine. The lithography hotspot detection sub-model is used to detect the number of lithography hotspots in an input image.

[0080] The visual effect filling sub-model, the preset effect filler and the lithography hotspot detection sub-model are sequentially spliced ​​end to end to form a visual effect generation model. The effect filler is used to simulate the filling of the corresponding lithography visual effect for the closed line area in the input image according to the visual effect filling parameters.

[0081] In this embodiment, if Figure 2 As shown, the visual effect filling sub-model includes a global feature extraction module, a local feature extraction module, a feature fusion module and a filling parameter generation module, the lithography hotspot detection sub-model includes a frequency domain feature extraction module, a support vector machine parameter configuration module and a support vector machine classification module, and there is a feedback adjustment mechanism between the lithography hotspot detection sub-model and the effect filler.

[0082] In this embodiment, the visual effect filling sub-model first performs preliminary processing on the input image through the global feature extraction module. The global feature extraction module uses multiple convolutional layers and pooling layers to gradually extract high-level features of the image. These convolutional layers can capture visual features such as edges, textures, and shapes in the image, while the pooling layers reduce the size of the feature map through downsampling operations while retaining important information. This process is similar to when humans observe an image, they first notice the overall outline and main structure. In this way, the global feature extraction module is able to generate a set of feature maps that contain overall information about the image.

[0083] The local feature extraction module further refines the global features. This module also consists of multiple convolutional layers and pooling layers, but its design focuses more on capturing detailed features in the image. The local feature extraction module can identify small texture changes, local shape features, and other small but critical visual information in the image. In this way, the local feature extraction module complements the output of the global feature extraction module, so that the final feature representation contains both the overall information of the image and rich local details.

[0084] The function of the feature fusion module is to effectively fuse global features and local features. This module usually uses a multi-layer perceptron (MLP) or other forms of neural network structure to splice, weight and combine global features and local features to generate a comprehensive feature representation. This process is similar to combining the overall impression of an image with detailed observations to form a more comprehensive and accurate understanding. The output of the feature fusion module will be passed as input to the filling parameter generation module.

[0085] The filling parameter generation module is responsible for generating visual effect filling parameters based on the fused features. This module usually uses a fully connected layer or other forms of neural network structure to map the input features to a specific filling parameter space. The filling parameters include specific values ​​of various visual effects such as color, texture, light and shadow. These parameters will be used in the subsequent effect filler to simulate and fill the visual effects of the input image. In this way, the visual effect filling sub-model can generate visual effect filling parameters that match the input image, providing the necessary basic information for subsequent image processing steps.

[0086] The first step of the lithography hotspot detection sub-model is to perform frequency domain analysis on the input image through the frequency domain feature extraction module, so as to extract spectral features according to its spectral characteristics. The support vector machine parameter configuration module sets the parameters of the support vector machine (SVM) according to the extracted frequency domain features. SVM is a classic machine learning algorithm that is good at handling classification problems in high-dimensional space. The support vector machine parameter configuration module selects appropriate kernel functions and hyperparameters by analyzing frequency domain features, so that SVM can effectively classify in feature space. The selection of kernel functions and the adjustment of parameters are the key to SVM performance. The parameter configuration module usually uses methods such as cross-validation to ensure that the selected parameters can perform well on both training data and test data. The support vector machine classification module is the core part of the lithography hotspot detection sub-model. This module uses the configured SVM to classify the input image and detect the lithography hotspots in it. Lithography hotspots refer to areas where defects or problems may occur during the lithography process, usually manifested as high-frequency areas or specific texture patterns in the image. The support vector machine classification module classifies each pixel or area in the image by analyzing frequency domain features and marks possible lithography hotspots. This process is similar to automatically identifying those areas of detail that require special attention in a complex image, thereby providing guidance for the lithography process.

[0087] The visual effect filling sub-model, the preset effect filler and the lithography hotspot detection sub-model are sequentially spliced ​​end to end to form a complete visual effect generation model. The effect filler simulates filling of the closed line area in the input image according to the visual effect filling parameters. The effect filler usually uses image processing algorithms or generative adversarial networks (GAN) and other technologies to convert the filling parameters into specific visual effects. In this way, the effect filler can generate realistic visual effects in the input image, making the image look more complete and beautiful. There is a feedback adjustment mechanism between the lithography hotspot detection sub-model and the effect filler. The function of this mechanism is to adjust the filling strategy of the effect filler according to the results of the lithography hotspot detection, so as to optimize the final visual effect. Specifically, when the lithography hotspot detection sub-model detects that there are lithography hotspots in certain areas, the feedback adjustment mechanism will notify the effect filler to use different filling parameters or adjust the filling strategy in these areas to reduce the number of lithography hotspots. In this way, the feedback adjustment mechanism can dynamically optimize the visual effect generation process to ensure that the final image is both beautiful and meets the requirements of the lithography process.

[0088] In one embodiment, completing the model training of the visual effect generation model using the image dataset includes the following steps:

[0089] Arrange all sample image data in the image data set into a first image training set;

[0090] Based on the line edge features, the closed line areas of all sample image data in the image data set are identified;

[0091] Deleting the photolithography visual effects in the closed line areas of all sample image data and retaining the annotation labels of the photolithography visual effects to obtain a plurality of sample line image data;

[0092] Integrate all sample line image data into a second image training set;

[0093] Segmenting the closed line areas of all sample line image data in the second image training set into sample image blocks by using an image segmentation algorithm, and integrating all sample image blocks into an image block training set;

[0094] The independent model training of the visual effect filling sub-model is completed by combining the second image training set and the image block training set;

[0095] The independent model training of the lithography hotspot detection sub-model is completed using the first image training set.

[0096] In this embodiment, first, all sample image data in the image data set are organized into a first image training set. Next, the closed line areas of all sample image data in the image data set are identified based on line edge features. The goal of this step is to find closed line areas in the image, which are usually important structures in the image, such as contours or boundaries involving patterns. In specific implementation, edge detection algorithms, such as the Canny edge detection algorithm, can be used to first extract edge features in the image. Then, morphological operations, such as expansion and corrosion, are used to further process the edge image to remove noise and fill the gaps between edges. Next, a contour detection algorithm, such as the findContours function in OpenCV, is used to identify all contours in the image. Finally, closed line areas are screened out by analyzing the shape and closure of the contour. These closed line areas will serve as the focus of subsequent processing, providing a basis for visual effect filling and lithography hotspot detection of the image.

[0097] After identifying the closed line areas, delete the lithography visual effects in the closed line areas of all sample image data, and retain the annotated labels of the lithography visual effects to obtain multiple sample line image data. The goal of this step is to remove the lithography visual effects in the image and only retain the line structure and the corresponding label information. In specific implementation, you can first use the image masking technology to cover or fill these areas with background colors or transparent colors based on the closed line areas identified in the previous step. In this way, the area that originally contained the lithography visual effect will be deleted, and only the line structure will be retained. Then, match these processed image data with the original lithography visual effect labels, and retain the label information and attach it to the processed image data. In this way, multiple sample line image data can be generated, which not only contains the line structure of the image, but also retains the label information of the lithography visual effect, providing a clear target for subsequent model training.

[0098] Integrate all sample line image data into a second image training set. The goal of this step is to aggregate the processed sample line image data to form a new training set. In specific implementation, all sample line image data can be classified and labeled according to certain rules, such as grouping according to the line type of the image, the size and shape of the closed area, and adding corresponding labels to each image. These labels may include information such as the line features of the image and the attributes of the closed area. After completing these processes, all sample line image data and labels are integrated into a unified database or folder to form a second image training set.

[0099] The closed line areas of all sample line image data in the second image training set are segmented into sample image blocks by an image segmentation algorithm, and all sample image blocks are integrated into an image block training set. The goal of this step is to segment the closed line areas in the large image into smaller image blocks for the convenience of model training and processing. In specific implementation, an image segmentation algorithm, such as a region-based segmentation algorithm or an edge-based segmentation algorithm, can be used to first locate the boundaries of the closed line area. Then, based on these boundaries, the closed line area is cut into several small image blocks. Each image block will contain one or more closed line areas and corresponding label information. Next, these image blocks are classified and labeled according to certain rules, and integrated into a unified database or folder to form an image block training set. In this way, a large amount of small-size image block data can be generated to provide more fine-grained data support for model training.

[0100] The independent model training of the visual effect filling sub-model is completed by combining the second image training set and the image block training set. The goal of this step is to use the data of the two training sets to train a model that can generate visual effect filling parameters. In the specific implementation, the second image training set and the image block training set can be first divided into a training set and a validation set according to a certain ratio. Then, a convolutional neural network (CNN) model is designed, which includes a global feature extraction module, a local feature extraction module, a feature fusion module, and a filling parameter generation module. Next, the data of the training set is input into the model, and the parameters of the model are gradually adjusted using the back propagation algorithm and the gradient descent optimization algorithm so that it can accurately generate visual effect filling parameters. During the training process, the data of the validation set can be used to evaluate and adjust the model to ensure the generalization ability and robustness of the model. In this way, an efficient visual effect filling sub-model can be trained to provide reliable parameter generation for visual effect filling of images.

[0101] Finally, the independent model training of the lithography hotspot detection sub-model is completed using the first image training set. The goal of this step is to use the data of the first image training set to train a model that can detect lithography hotspots. In specific implementation, the first image training set can be first divided into a training set and a validation set according to a certain ratio. Then, a support vector machine (SVM) model is designed, which includes a frequency domain feature extraction module, a support vector machine parameter configuration module, and a support vector machine classification module. Next, the data of the training set is input into the model, and the frequency domain features of the image are extracted using the frequency domain feature extraction module, and the parameters of the SVM are set according to these features. Then, the SVM is used for classification to detect the lithography hotspots in the image. During the training process, the data of the validation set can be used to evaluate and adjust the model to ensure the accuracy and robustness of the model. In this way, an efficient lithography hotspot detection sub-model can be trained to provide reliable hotspot detection for the lithography process of the image.

[0102] In one implementation manner, completing the independent model training of the visual effect filling sub-model in combination with the second image training set and the image block training set includes the following steps:

[0103] Initialize the module parameters of the global feature extraction module, the local feature extraction module, and the filling parameter generation module, as well as the weight parameters of the feature fusion module;

[0104] Inputting the second image training set into the global feature extraction module, and using the global feature extraction module to extract global line pattern features of the sample line image data in the second image training set, the global line pattern features including the number of closed line areas in the sample line image data, the region similarity and the region spacing;

[0105] The image block training set is input into the local feature extraction module, and the local line pattern features of the sample image blocks in the image block training set are extracted by the local feature extraction module, wherein the local line pattern features include the regional complexity of the closed line area in the sample image block, the number of regional line border endpoints, the regional area and the regional shape;

[0106] The global line pattern features and the local line pattern features are input into a feature fusion module, and the global line pattern features and the local line pattern features are fused into a multi-scale line pattern feature through the feature fusion module based on a weight parameter;

[0107] Inputting the multi-scale line pattern features into a filling parameter generation module, the filling parameter generation module randomly generates initial visual effect filling parameters according to the multi-scale line pattern features, the initial visual effect filling parameters including initial types of lithography visual effects corresponding to all closed line areas in the sample line image data;

[0108] A cross-entropy loss function is constructed by combining the annotation labels and the initial type of lithography visual effects, and the module parameters and weight parameters are optimized in a way of minimizing the cross-entropy loss function until the module parameters and weight parameters reach the optimal module parameters and optimal weight parameters respectively.

[0109] In this embodiment, first, initialize the module parameters of the global feature extraction module, the local feature extraction module, the filling parameter generation module, and the weight parameters of the feature fusion module. The global feature extraction module, the local feature extraction module and the filling parameter generation module are each responsible for the extraction and processing of features at different levels. The initialization of the global feature extraction module can adopt a random initialization method, such as using a uniform distribution or a normal distribution to generate initial parameter values. The initialization of the local feature extraction module can adopt the parameters of the pre-trained model, such as using the convolutional neural network parameters pre-trained on ImageNet. The initialization of the filling parameter generation module can adopt zero initialization or small value random initialization to ensure that the generated initial filling parameters will not have too much impact on the training process. The weight parameter initialization of the feature fusion module can also adopt a random initialization method to ensure that different features have equal weights in the early stage.

[0110] The global feature extraction module usually adopts a convolutional neural network (CNN) architecture to extract high-level features of the image through multi-layer convolution and pooling operations. In specific implementation, each sample line image data in the second image training set can be input into the global feature extraction module, and the global line pattern features can be extracted through multi-layer convolution and pooling operations. These features include the number of closed line areas in the sample line image data, the region similarity, and the region spacing. The number of regions can be obtained by counting the number of closed line areas in the image, the region similarity can be obtained by calculating the similarity measure between closed line areas, such as using cosine similarity or Euclidean distance, and the region spacing can be obtained by calculating the distance between closed line areas, such as using the minimum distance or average distance.

[0111] The local feature extraction module can also adopt the convolutional neural network (CNN) architecture, but its focus is on extracting detailed features in the image block. In specific implementation, each sample image block in the image block training set can be input into the local feature extraction module, and after multiple layers of convolution and pooling operations, the local line pattern features are extracted. These features include the regional complexity of the closed line area in the sample image block, the number of regional line border endpoints, the regional area and the regional shape. The regional complexity can be obtained by calculating the edge density or the number of line intersections of the closed line area, the number of regional line border endpoints can be obtained by counting the number of endpoints in the closed line area, the regional area can be obtained by calculating the number of pixels in the closed line area, and the regional shape can be described by shape descriptors, such as Hu moments or Fourier descriptors.

[0112] The feature fusion module is designed to integrate features at different levels to form a unified feature representation. In specific implementation, the global line pattern features and local line pattern features can be fused by weighted summation or splicing. The weighted summation method is to weighted average the global features and local features according to the weight parameter, for example, multiply the global features by the weight parameter α, and multiply the local features by the weight parameter β, and then add the two. The splicing method is to splice the global features and local features in the feature dimension to form a higher-dimensional feature vector. In this way, the information of global features and local features can be integrated to form multi-scale line pattern features, providing comprehensive information support for the subsequent filling parameter generation.

[0113] The filling parameter generation module is designed to generate initial visual effect filling parameters based on the input features, which will be used to guide the visual effect filling of the image. In specific implementation, a fully connected neural network can be designed to input multi-scale line pattern features into the network, and then generate initial visual effect filling parameters after multiple layers of fully connected layers and activation functions. These parameters include the initial types of lithography visual effects corresponding to all closed line areas in the sample line image data, such as filling color, texture type, etc. The generated initial parameters can be randomly generated, such as randomly sampling parameter values ​​within a certain range, to ensure that the initial parameters have a certain diversity.

[0114] The cross entropy loss function is a commonly used classification loss function that measures the difference between the predicted result and the true label. In specific implementation, the generated initial visual effect filling parameters can be compared with the annotated labels to calculate the cross entropy loss. Then, the back propagation algorithm and the gradient descent optimization algorithm are used to gradually adjust the module parameters of the global feature extraction module, the local feature extraction module, the filling parameter generation module, and the weight parameters of the feature fusion module to minimize the cross entropy loss. In each iteration, the parameters of each module are updated according to the gradient information of the loss function until the loss function converges or the preset stop condition is reached.

[0115] In one embodiment, completing the independent model training of the lithography hotspot detection sub-model using the first image training set includes the following steps:

[0116] Initialize the SVM parameters of the support vector machine classification module;

[0117] Inputting the first image training set into the frequency domain feature extraction module, using the frequency domain feature extraction module and extracting the sample frequency features of the sample image data in the first image training set through a discrete cosine transform method;

[0118] The sample frequency features are input into the support vector machine parameter configuration module, and the sample frequency features and SVM parameters are fused and encoded into the first chromosome code through the support vector machine parameter configuration module. The optimal solution of the first chromosome code is solved based on the genetic algorithm to obtain the optimal sample frequency feature subset and the optimal SVM parameters.

[0119] In this embodiment, first, the SVM parameters of the support vector machine (SVM) classification module are initialized. Support vector machine is a supervised learning model for classification and regression analysis. Its core idea is to find an optimal hyperplane to separate data points of different categories. Initializing SVM parameters is to ensure that the model has a reasonable starting point in the early stage of training. SVM parameters mainly include kernel function type, penalty parameter C, kernel parameter (such as γ of Gaussian kernel), etc. The kernel function type determines the mapping method of data in high-dimensional space. Common kernel functions include linear kernel, polynomial kernel, Gaussian kernel, etc. The penalty parameter C controls the tolerance of the model to misclassification. The larger the C value, the lower the tolerance of the model to misclassification, and the more it tends to find a more accurate classification boundary. The kernel parameter γ determines the width of the Gaussian kernel. The larger the γ value, the higher the complexity of the model. Initializing these parameters can be determined by empirical values ​​or by cross-validation. For example, the initial kernel function type can be set to Gaussian kernel, the C value can be 1, and the γ value can be set to 0.01. In this way, a good starting point can be provided for model training, ensuring that the SVM classification module can effectively learn and adjust parameters during the training process.

[0120] Next, the first image training set is input into the frequency domain feature extraction module, and the sample frequency features of the sample image data in the first image training set are extracted using the frequency domain feature extraction module and the discrete cosine transform (DCT) method. The frequency domain feature extraction module is designed to convert image data from the time domain to the frequency domain in order to extract the frequency features of the image. Discrete cosine transform is a commonly used frequency domain transformation method. By performing a cosine transform on the image data, the pixel values ​​of the image can be represented as a set of frequency components. In specific implementation, each sample image data in the first image training set can be input into the frequency domain feature extraction module, and preprocessing, such as grayscale and normalization, is first performed, and then the image data is converted to the frequency domain through discrete cosine transform. The formula for discrete cosine transform is:

[0121] ;

[0122] in, Represents the pixel value of the image in the time domain, represents the frequency components of the image in the frequency domain, Indicates the size of the image.

[0123] The design purpose of the support vector machine parameter configuration module is to fuse the sample frequency features and SVM parameters so as to optimize them through the genetic algorithm. In specific implementation, the sample frequency features and SVM parameters can be represented as a chromosome code, for example, the sample frequency features are represented as a feature vector, the SVM parameters are represented as a parameter vector, and then the two are spliced ​​into a chromosome code. The genetic algorithm is an optimization algorithm based on natural selection and genetic mechanism, which can gradually approach the optimal solution through selection, crossover and mutation operations. First, a set of chromosome codes is initialized as the initial population, and then the chromosomes with higher fitness are selected for reproduction through selection operations. The fitness can be evaluated by cross-validation, such as calculating the accuracy of SVM classification. Next, the selected chromosomes are crossed through the crossover operation to generate new chromosomes, such as using a single-point crossover or a multi-point crossover. Finally, the newly generated chromosomes are mutated through the mutation operation, such as randomly changing the values ​​of certain genes in the chromosome. Through multiple iterations, chromosomes with higher fitness are gradually screened out, and finally the optimal sample frequency feature subset and the optimal SVM parameters are obtained.

[0124] Through the above steps, the sample frequency features of the first image training set can be extracted and the SVM parameters can be optimized, and finally the optimal sample frequency feature subset and the optimal SVM parameters can be obtained, providing efficient feature representation and classification model for subsequent image classification.

[0125] In one embodiment, the plane line draft design drawing is input into the visual effect generation model that has completed model training, and the lithography visual effect filling scheme for all closed line areas in the plane line draft design drawing is generated by the visual effect generation model, including the following steps:

[0126] Inputting the plane line draft design drawing into the effect filler and the visual effect filling sub-model that has completed model training, and using the visual effect filling sub-model to generate initial visual effect filling parameters of the plane line draft design drawing;

[0127] Through the effect filler and according to the initial visual effect filling parameters of the plane line draft design, all closed line areas in the plane line draft design are simulated and filled with corresponding lithography visual effects to obtain a lithography primitive simulation image corresponding to the target product;

[0128] Inputting the photolithography primitive simulation image into the photolithography hotspot detection sub-model that has completed model training, and outputting the number of photolithography hotspots in the photolithography primitive simulation image through the photolithography hotspot detection sub-model;

[0129] If the number of lithography hot spots is less than a preset threshold, a lithography visual effect filling solution is generated for all closed line areas in the plane line draft design drawing according to the initial visual effect filling parameters.

[0130] In this embodiment, if the number of lithography hot spots is greater than or equal to the number threshold, the initial visual effect filling parameter is encoded as a second chromosome code;

[0131] Adopting a genetic algorithm and utilizing a photolithography hotspot detection sub-model to iteratively update the second chromosome code until the number of photolithography hotspots is less than a threshold value, and taking the second chromosome code of the current iteration number as the target chromosome code;

[0132] Decode the target chromosome encoding into optimal visual effect filling parameters;

[0133] Generate a lithography visual effect filling solution for all closed line areas in the plane line draft design based on the optimal visual effect filling parameters.

[0134] In this embodiment, the plane line draft design drawing is first input into the effect filler and the visual effect filling sub-model that has completed model training. The visual effect filling sub-model uses a convolutional neural network (CNN) to extract the global and local features of the plane line draft design drawing, and then fuses these features through a feature fusion module to finally generate initial visual effect filling parameters. Next, through the effect filler and according to the initial visual effect filling parameters of the plane line draft design drawing, the corresponding lithography visual effects are simulated and filled into all closed line areas in the plane line draft design drawing. The role of the effect filler is to fill the corresponding lithography visual effects into each closed line area of ​​the plane line draft design drawing according to the parameters generated by the visual effect filling sub-model. For example, if the filling parameters of a certain area indicate that the area needs to be filled with a specific laser texture, the effect filler will simulate the corresponding laser effect in the area according to this parameter. In this way, a simulated image containing all lithography effects can be obtained, that is, a simulated image of the lithography primitives corresponding to the target product.

[0135] Then, the simulated image of the lithography primitive is input into the lithography hotspot detection sub-model that has completed model training. The lithography hotspot detection sub-model detects the number of lithography hotspots in the simulated image of the lithography primitive through the frequency domain feature extraction module and the support vector machine (SVM) classification module. Lithography hotspots are areas where defects are prone to occur in the lithography process. By detecting these hotspots, the quality of the lithography primitive can be evaluated. The frequency domain feature extraction module first uses discrete cosine transform (DCT) to extract the frequency domain features of the image. These features reflect the distribution of different frequency components in the image. Then, the support vector machine classification module determines the number of lithography hotspots in the image based on these frequency domain features. If the number of lithography hotspots detected is less than the preset threshold, it means that the current visual effect filling parameters are good enough, and the final lithography visual effect filling solution can be directly generated.

[0136] If the number of photolithography hotspots is greater than or equal to the quantity threshold, the initial visual effect filling parameters are encoded as the second chromosome code. Genetic algorithm is an optimization algorithm that simulates natural selection and genetic mechanisms. By encoding parameters, these parameters can be optimized using genetic algorithms. Specifically, the initial visual effect filling parameters are encoded as chromosomes to form an initial population. Each chromosome represents a possible solution. Through the selection, crossover and mutation operations of the genetic algorithm, new chromosomes can be continuously generated, thereby gradually approaching the optimal solution. The genetic algorithm is used and the photolithography hotspot detection submodel is used to iteratively update the second chromosome code until the number of photolithography hotspots is less than the quantity threshold. The core of the genetic algorithm is to iteratively update the chromosome through selection, crossover and mutation operations. The selection operation selects chromosomes with higher fitness for retention and crossover according to the feedback of the photolithography hotspot detection submodel. The crossover operation generates new chromosomes by exchanging some genes of the chromosome. The mutation operation increases the diversity of the population by randomly changing some genes of the chromosome. Through multiple iterations, the fitness of the chromosome is gradually improved, and finally a chromosome code with a number of photolithography hotspots less than the quantity threshold is obtained.

[0137] The second chromosome encoding of the current iteration number is used as the target chromosome encoding. After multiple iterations, the chromosome with the highest fitness is selected as the target chromosome encoding. This encoding represents an optimized set of visual effect filling parameters, which can be used to generate the final lithography visual effect filling scheme. Decode the target chromosome encoding into the optimal visual effect filling parameters. The decoding process is the process of converting the chromosome encoding back to the visual effect filling parameters. Through decoding, an optimal set of visual effect filling parameters can be obtained, which can generate high-quality lithography visual effects in the plane line draft design drawing. According to the optimal visual effect filling parameters, a lithography visual effect filling scheme for all closed line areas in the plane line draft design drawing is generated. Finally, according to the optimal visual effect filling parameters, all closed line areas in the plane line draft design drawing are filled with corresponding lithography visual effects. In this way, a high-quality lithography visual effect filling scheme can be generated to ensure that the final product has good visual effects and fewer defects in the actual lithography process.

[0138] In one embodiment, using an edge recognition algorithm to identify line edge features in a preprocessed plane line draft design drawing, and generating and marking line pattern features of the plane line draft design drawing according to the line edge features includes the following steps:

[0139] Use edge recognition algorithm to identify the line edge features in the pre-processed plane line draft design;

[0140] The plane line draft design drawing is skeletonized based on the line edge features to obtain the line skeleton structure of the plane line draft design drawing;

[0141] Analyze and obtain the line topological features in the line skeleton structure;

[0142] A feature extraction algorithm is used to extract line geometric features from line edge features;

[0143] The line pattern features of the plane line draft design are generated and annotated by combining the line topological features and line geometric features.

[0144] In this embodiment, the edge recognition algorithm is used to identify the edge features of the lines in the plane line draft design after preprocessing. The edge recognition algorithm is an important step in image processing. By detecting the change of pixel intensity in the image, the edge features in the image can be identified. Commonly used edge recognition algorithms include Sobel operator, Canny edge detection algorithm, etc. Taking the Canny edge detection algorithm as an example, the preprocessed plane line draft design is first subjected to Gaussian blur processing to reduce the interference of noise on edge detection. Then, the gradient intensity and direction of each pixel in the image are calculated, and only the pixel points whose gradient intensity reaches the local maximum are retained through the non-maximum suppression step. Finally, double threshold processing is applied to connect the edges to form a complete edge map. Through these steps, the line edge features in the plane line draft design can be clearly identified, laying the foundation for subsequent processing.

[0145] The plane line draft design drawing is skeletonized based on the line edge features to obtain the line skeleton structure of the plane line draft design drawing. Skeletonization is a technology that converts lines of uneven width in an image into single-pixel width. Commonly used algorithms include the Zhang-Suen thinning algorithm. Taking the Zhang-Suen thinning algorithm as an example, each pixel in the image is first traversed, and the pixel is judged whether it is a boundary point based on the pixel value of the eight neighborhoods. Then, the boundary points that do not affect the connectivity of the lines are deleted according to certain rules, and the above steps are repeated until the image no longer changes. Through skeletonization, the lines in the plane line draft design drawing can be simplified into a skeleton structure with a single pixel width. This structure retains the topological information of the lines and removes redundant redundant information, making subsequent topological analysis and feature extraction more efficient and accurate.

[0146] The line topological features in the line skeleton structure are analyzed. Line topological features refer to the connection relationship and topological structure of the line, including endpoints, intersections, bifurcation points, etc. These features of the line can be extracted by performing topological analysis on the line skeleton structure. Specifically, each pixel in the skeleton image can be traversed and the type of the pixel can be determined based on the eight-neighborhood information of the pixel. If a pixel has only one neighbor, the pixel is an endpoint; if there are two neighbors, the pixel is part of the line segment; if there are three or more neighbors, the pixel is an intersection or bifurcation point. In this way, all topological features in the line skeleton structure can be obtained, providing basic data for subsequent geometric feature extraction and pattern feature generation.

[0147] The feature extraction algorithm is used to extract the line geometric features from the line edge features. The line geometric features include the length, curvature, direction and other information of the line. Taking Hough transform as an example, the skeleton image is first converted into parameter space, and the straight line segments in the image are detected by accumulating votes in the parameter space. For the curve segments, they can be described by methods such as polynomial fitting or spline interpolation. Through these methods, the geometric features of the line can be extracted, such as the length and direction of the straight line segment, the curvature of the curve segment, etc. These geometric features can not only reflect the shape and structure of the line, but also provide important information for the subsequent pattern feature generation.

[0148] The line pattern features of the plane line draft design are generated and annotated by combining the line topology features and line geometry features. Line pattern features refer to high-level features generated by integrating the topology and geometry information of the lines, such as patterns of specific shapes, repeated structures, etc. By combining the previously extracted topology features and geometric features, richer and more detailed pattern features can be generated. For example, based on the endpoints and intersection information of the lines, specific geometric shapes such as rectangles and triangles can be detected; based on the direction and curvature information of the lines, symmetrical structures or periodically repeated parts in the pattern can be identified. Finally, these line pattern features are annotated to provide a complete information framework for the plane line draft design.

[0149] The present invention also discloses a photolithography element generation system for holographic laser printing, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method described in any one of the above embodiments is implemented.

[0150] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0151] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD) or a flash memory card (FC) equipped on the computer device, and the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output, and this application does not impose any restrictions on this.

[0152] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as above, which are not provided in detail for the sake of simplicity.

[0153] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A method for generating photolithographic primitives for holographic laser printing, characterized in that: The steps include: Obtain the graphic line drawing of the target product; Preprocessing the plane line draft design drawing; Using an edge recognition algorithm to identify line edge features in the preprocessed plane line draft design drawing, and generating and marking line pattern features of the plane line draft design drawing based on the line edge features; Extracting an image data set from a preset pattern database based on the line pattern features, the image data set includes a plurality of sample image data, each of the sample image data includes a plurality of closed line areas, and each of the closed line areas of the sample image data is pre-filled and annotated with different types of lithography visual effects; Building a visual effect filling sub-model based on a convolutional neural network, wherein the visual effect filling sub-model is used to generate visual effect filling parameters of an input image; Building a lithography hotspot detection sub-model based on a support vector machine, wherein the lithography hotspot detection sub-model is used to detect the number of lithography hotspots in an input image; The visual effect filling sub-model, the preset effect filler and the lithography hotspot detection sub-model are sequentially spliced ​​end to end to form a visual effect generation model, the effect filler is used to simulate the filling of the corresponding lithography visual effect on the closed line area in the input image according to the visual effect filling parameters, the visual effect filling sub-model includes a global feature extraction module, a local feature extraction module, a feature fusion module and a filling parameter generation module, the lithography hotspot detection sub-model includes a frequency domain feature extraction module, a support vector machine parameter configuration module and a support vector machine classification module, and a feedback adjustment mechanism is provided between the lithography hotspot detection sub-model and the effect filler; Complete model training of the visual effect generation model through the image data set; Inputting the plane line draft design drawing into the effect filler and the visual effect filling sub-model that has completed model training, and using the visual effect filling sub-model to generate initial visual effect filling parameters of the plane line draft design drawing; By means of the effect filler and according to the initial visual effect filling parameters of the plane line draft design drawing, all closed line areas in the plane line draft design drawing are simulated and filled with corresponding lithography visual effects to obtain a lithography primitive simulation image corresponding to the target product; Inputting the photolithography primitive simulation image into the photolithography hotspot detection sub-model that has completed model training, and outputting the number of photolithography hotspots in the photolithography primitive simulation image through the photolithography hotspot detection sub-model; If the number of the lithography hot spots is less than a preset number threshold, generating a lithography visual effect filling scheme for all closed line areas in the plane line draft design drawing according to the initial visual effect filling parameters; According to the photolithography visual effect filling scheme, dot matrix gratings are arranged in the plane line draft design drawing to form target photolithography primitives corresponding to the target product.

2. The method for generating photolithographic primitives for holographic laser printing according to claim 1, characterized in that: The step of completing the model training of the visual effect generation model through the image data set comprises the following steps: Arrange all the sample image data in the image data set into a first image training set; Identify closed line regions of all the sample image data in the image data set based on the line edge features; Deleting the photolithography visual effects in the closed line areas of all the sample image data and retaining the annotation labels of the photolithography visual effects to obtain a plurality of sample line image data; Integrate all the sample line image data into a second image training set; Segmenting the closed line areas of all the sample line image data in the second image training set into sample image blocks by using an image segmentation algorithm, and integrating all the sample image blocks into an image block training set; completing independent model training of the visual effect filling sub-model by combining the second image training set and the image block training set; The independent model training of the lithography hotspot detection sub-model is completed using the first image training set.

3. The method for generating photolithography primitives for holographic laser printing according to claim 2, characterized in that: The step of combining the second image training set and the image block training set to complete the independent model training of the visual effect filling sub-model comprises the following steps: Initializing module parameters of the global feature extraction module, the local feature extraction module, and the filling parameter generation module, and weight parameters of the feature fusion module; Inputting the second image training set into the global feature extraction module, and using the global feature extraction module to extract global line pattern features of the sample line image data in the second image training set, wherein the global line pattern features include the number of closed line areas in the sample line image data, the region similarity, and the region spacing; Inputting the image block training set into the local feature extraction module, and using the local feature extraction module to extract local line pattern features of the sample image blocks in the image block training set, wherein the local line pattern features include regional complexity, the number of regional line border endpoints, regional area and regional shape of the closed line area in the sample image block; Inputting the global line pattern feature and the local line pattern feature into the feature fusion module, and fusing the global line pattern feature and the local line pattern feature into a multi-scale line pattern feature through the feature fusion module based on the weight parameter; Inputting the multi-scale line pattern features into the filling parameter generation module, the filling parameter generation module randomly generates initial visual effect filling parameters according to the multi-scale line pattern features, the initial visual effect filling parameters including initial types of lithography visual effects corresponding to all the closed line areas in the sample line image data; A cross entropy loss function is constructed in combination with the annotation label and the initial type of the lithography visual effect, and the module parameters and the weight parameters are optimized in a manner of minimizing the cross entropy loss function until the module parameters and the weight parameters reach optimal module parameters and optimal weight parameters, respectively.

4. The method for generating photolithographic primitives for holographic laser printing according to claim 2, characterized in that: The independent model training of the lithography hotspot detection sub-model using the first image training set comprises the following steps: Initializing the SVM parameters of the support vector machine classification module; Inputting the first image training set into the frequency domain feature extraction module, and using the frequency domain feature extraction module and a discrete cosine transform method to extract sample frequency features of the sample image data in the first image training set; The sample frequency features are input into the support vector machine parameter configuration module, and the sample frequency features and the SVM parameters are fused and encoded into a first chromosome code through the support vector machine parameter configuration module. The first chromosome code is optimally solved based on a genetic algorithm to obtain an optimal sample frequency feature subset and an optimal SVM parameter.

5. The method for generating photolithographic primitives for holographic laser printing according to claim 1, characterized in that: The method further comprises the steps of: If the number of the lithography hot spots is greater than or equal to the number threshold, encoding the initial visual effect filling parameter as a second chromosome code; Adopting a genetic algorithm and using the lithography hotspot detection sub-model to iteratively update the second chromosome code until the number of the lithography hotspots is less than the number threshold, and taking the second chromosome code of the current iteration number as the target chromosome code; Decoding the target chromosome into optimal visual effect filling parameters; A lithography visual effect filling scheme is generated for all closed line areas in the plane line draft design drawing according to the optimal visual effect filling parameters.

6. The method for generating photolithographic primitives for holographic laser printing according to claim 1, characterized in that: The step of using an edge recognition algorithm to identify line edge features in the preprocessed plane line draft design drawing, and generating and marking line pattern features of the plane line draft design drawing according to the line edge features comprises the following steps: Using an edge recognition algorithm to identify line edge features in the pre-processed plane line drawing design; Performing skeleton processing on the plane line draft design drawing based on the line edge features to obtain a line skeleton structure of the plane line draft design drawing; Analyze and obtain line topological features in the line skeleton structure; Using a feature extraction algorithm to extract line geometric features from the line edge features; The line pattern features of the plane line draft design drawing are generated and marked by combining the line topological features and the line geometric features.

7. A photolithography element generation system for holographic laser printing, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

Images