A method for filling two-dimensional grain boundary gaps based on three-dimensional information

Abstract

Based on the three-dimensional information two-dimensional grain boundary filling method, the image is first collected and benchmarked using high-resolution two-dimensional imaging technology to ensure the accurate association of spatial information. After the grain boundaries of the first-layer metallographic image are hand-drawn and perfected, the image is pre-processed by grayscale, enhancement, filtering, etc. Specific functions and formulas are used to remove noise and binarize, and professional software is used to refine the grain boundaries. Grain corrosion is performed by setting a threshold to obtain the center marker point, and the eight-direction tracking is used to calculate the proportional parameters of adjacent grains. The grain edges of the first-layer metallographic image are detected and scaled, and a dedicated program is used to fill the second layer of grains until a complete metallographic image is output. This method also combines three-dimensional reconstruction technology to verify the filling results and optimize the algorithm. This technology effectively improves the structural integrity and performance of crystalline materials, improves the accuracy of grain boundary reconstruction, enhances the reliability of material performance prediction, reduces experimental and computational costs, and improves data processing efficiency. It is applicable to a variety of materials and opens up new avenues for material microscopic research and application.

Description

Technical Field

[0001] The present invention relates to the field of material science, and in particular to a method for filling two-dimensional grain boundary defects based on three-dimensional information. Background Art

[0002] In the field of materials science, obtaining metallographic images of metal specimens is a common laboratory testing method. Grain boundaries, the interfaces between adjacent grains, are the most common lattice defects in materials. Microscopic grain size, grain boundaries, and microstructure provide crucial insights into a material's mechanical properties, such as tensile strength, toughness, and plasticity. Consequently, they are widely used in various design and manufacturing fields.

[0003] The morphology of grains in metallographic images can be used to predict relevant material properties. Two-dimensional images of grains play a central role in the study of microstructure and structural properties, and are crucial for a deeper understanding of a material's microscopic characteristics. Grain boundaries are often obscured by their volume, making their extraction extremely challenging. In particular, extracting grain boundaries and connecting and reconstructing missing grain boundaries remain key challenges and key areas in metallographic image processing. Accurate grain boundary extraction and reconstruction are key techniques in metallographic analysis, and the accuracy of the results directly impacts the precision of the entire analysis process. The key to extracting and reconstructing grain boundaries lies in implementing efficient, precise, and appropriate image segmentation strategies. Metallographic analysis relies primarily on meticulous microscopic observation and accurate recording of key parameters such as grain size by professionals. However, given the inherent instability of manual analysis, including time consumption, variability in experience, and fluctuations in skill level, these factors can significantly increase the error range of analysis results. The accuracy and reliability of the analysis process significantly impact the accuracy of grain boundary extraction and reconstruction. Therefore, based on the deep integration of computer technology and image processing, this paper innovatively proposes a 3D thinking algorithm to solve the problem of filling missing grain boundaries in metallographic image processing. The microstructure of metal materials has a decisive influence on their mechanical properties, and the microstructure of metallographic images is an important basis for analyzing these properties. During the preparation of metal specimen images, defects such as blurred or missing grain boundaries inevitably occur, which pose significant challenges to grain boundary extraction. Given the inefficiency of manually processing complex metallographic images, automatic filling of missing grain boundaries is essential.

[0004] While advances have been made in grain boundary analysis using three-dimensional information (e.g., 3D electron microscopy images of crystals, X-ray tomography, and other techniques), accurately filling 2D grain boundary gaps based on this 3D information remains a technical challenge. Therefore, a method for filling 2D grain boundary gaps based on 3D information has been proposed, aiming to achieve efficient and accurate grain boundary repair. Summary of the Invention

[0005] In response to the above-mentioned deficiencies in the existing technology, in order to effectively improve the structural integrity and performance of crystal materials, the present invention provides a method for filling two-dimensional grain boundary gaps based on three-dimensional information. This method can fully utilize the concept of spatial position to fill the two-dimensional missing grain boundary map, thereby achieving an efficient, complete and accurate two-dimensional grain boundary map, and facilitating subsequent three-dimensional reconstruction.

[0006] In order to solve the above technical problems, this application provides the following technical solutions:

[0007] A series of metallographic images of the sample are acquired using high-resolution two-dimensional imaging technology. All metallographic images are then benchmarked and registered based on specific markers using an image registration algorithm to accurately correlate spatial information. The image registration algorithm uses specific instructions from specialized image processing software (standardized functions or algorithms in image processing, such as MATLAB image processing functions) to achieve precise positioning.

[0008] The grain boundaries of the first metallographic layer are hand-drawn to complete them. Then all metallographic images are pre-processed. The pre-processing operations include grayscale processing using floating-point calculation method, image enhancement based on equalization technology formula, and image smoothing filtering combining Gaussian filtering and wavelet transform to ensure that the image quality meets the requirements of subsequent processing.

[0009] Use specific functions (MATLAB image processing functions) to eliminate noise in metallographic images, perform binarization operations according to predetermined formulas, and then use professional software tools to refine problems such as inconsistent grain boundary widths and burrs to obtain clear grain boundary images;

[0010] By setting different threshold disks, a multi-scale iterative etching operation is performed on each grain in the metallographic image while maintaining the geometric characteristics of the grain boundary. The grain center points are obtained at a specific iterative state before the grain completely disappears, and these center points are recorded as seed record points to provide a basis for subsequent analysis.

[0011] For each seed record mark point, a tracking strategy of eight directions at 45° intervals is implemented. Starting from the mark point, the tracking continues until the grain boundary is encountered. During this process, the length of each line segment is accurately recorded, and the ratio parameters of adjacent grains are then calculated to determine the relative relationship between the grains.

[0012] Perform edge detection and contour extraction on each grain in the first layer of the complete metallographic image, and perform scaling operations based on the proportional parameters calculated between two adjacent layers to provide an accurate template for subsequent filling.

[0013] A specially developed program is used to fill each grain of the second-layer metallographic diagram in turn. During the filling process, intelligent judgment and operation are performed based on the results of previous calculations and processing. The process is repeated until all grains in the layer are filled, and a complete metallographic diagram of the second layer is obtained.

[0014] The above filling process is executed cyclically, and the metallographic images of subsequent layers are processed in sequence until the complete metallographic images of all layers are output, thereby achieving comprehensive and automatic filling of grain boundaries and ensuring the accuracy and consistency of the filling results.

[0015] The present invention provides a method for filling two-dimensional grain boundary defects based on three-dimensional information, comprising:

[0016] Step 1: Read in a series of metallographic image stacks of the sample and perform benchmark registration on all metallographic images to correlate spatial information;

[0017] Step 2: Complete the first layer of complete grain boundary metallographic diagram by hand-drawing and pre-process all metallographic diagrams;

[0018] Step 3: Remove noise and perform binarization before performing refinement processing;

[0019] Step 4: Iteratively etch each grain in the metallographic image to find the grain center and obtain multiple seed recording mark points;

[0020] Step 5: Perform eight-direction tracking on each grain to calculate the proportional parameters of its adjacent layers;

[0021] Step 6: Perform edge detection and extraction on the first layer of the complete metallographic image and scale it accordingly;

[0022] Step 7: Fill each grain of the second layer metallographic diagram in turn to obtain a complete second layer metallographic diagram;

[0023] Step 8: Obtain complete metallographic images of all layers in sequence and output them.

[0024] Preferably, the step 1 includes:

[0025] Step 1.1: After heat treatment, the sample is mounted, ground, marked with diamond points, polished, etched, and photographed to obtain the first metallographic image. The polishing, etching, and photographing operations are repeated to obtain the second, third, and other series of metallographic images of the same area.

[0026] In step 1.2, ImageJ software was used to perform rigorous image processing on a series of metallographic images for specific regions. Specifically, using diamond-shaped markers as reference points, plugins, registration, and SIFT commands were used to ensure precise vertical registration of all images for consistent and accurate image analysis. By precisely cropping the desired region, a series of image stacks were generated.

[0027] Preferably, the step 2 includes:

[0028] Step 2.1 Manually improve the first metallographic image to ensure the integrity of its grain boundaries. The grain boundary filling work of all subsequent metallographic images can be completed automatically by computer;

[0029] Step 2.2: grayscale processing, image enhancement, and image smoothing filtering are performed on all metallographic images;

[0030] Step 2.2.1 Grayscale processing uses floating point calculation method;

[0031] Step 2.2.2 Image enhancement uses the equalization technique formula;

[0032] Step 2.2.3 Image smoothing filtering uses Gaussian filtering and wavelet transform.

[0033] Preferably, the step 3 includes:

[0034] Step 3.1 Use the imfill function to eliminate the noise in the metallographic image;

[0035] Step 3.2: Use the formula to perform binarization operation on the metallographic image;

[0036] Step 3.3: Use MATLAB to refine the undesirable phenomena such as inconsistent grain boundary width and burrs.

[0037] Preferably, the specific steps in step 4 are:

[0038] Step 4.1 sets different thresholds, disk, and performs multi-scale iterative etching on the given image while maintaining the geometric features of the grain boundary. During this process, each time an etching operation is completed, the grain boundary will shrink inward by one layer until the grain completely disappears, and the grain center point is obtained.

[0039] In step 4.2, the grain boundary curvature is described using the mathematical model Gaussian curvature or mean curvature, the degree of curvature is calculated through the local grain boundary geometric characteristics, and the shape and surface structure of the filling area are inferred.

[0040] Step 4.2.1 uses the local grain boundary tangent plane equation and numerical methods to calculate the curvature, dynamically adjust the shape of the filling area, and combine the crystal symmetry and space group information to ensure that the lattice structure of the filling area is consistent with the surrounding area.

[0041] Step 4.2.2: Use a clustering algorithm to cluster the crystal structure regions and analyze the similarities and differences between clusters. Combining the clustering results with physical properties, a data-driven similarity assessment framework is constructed. By adjusting the similarity threshold, the properties of similar grains can be accurately predicted or filled in.

[0042] In step 4.2.3, for the crystal structure, similarity metrics such as geometric distance, phase similarity, and energy minimization can be used to assign a similarity score to each grain, influencing the infill strategy. A dynamic adjustment mechanism adjusts the infill threshold and similarity criteria in real time based on the grain distribution characteristics. Incremental learning is used to optimize parameter configurations as data accumulates and the algorithm iterates.

[0043] Step 4.3 records the center point of each grain before it disappears, and obtains multiple seed recording mark points.

[0044] Preferably, the step 5 includes:

[0045] Step 5.1: Implement a tracking strategy in eight directions (45° apart) for each marker point;

[0046] In step 5.2, ensure that the line starts from the marked point and continues to track until the grain boundary is encountered. During this period, the length of each line segment is recorded and the scale parameter is calculated.

[0047] Preferably, the step 6 includes:

[0048] Step 6.1: Perform edge detection on each grain in the first layer of the complete metallographic image to extract the outline;

[0049] Step 6.2: Scale according to the ratio parameters of the two adjacent layers.

[0050] Preferably, the step 7 includes:

[0051] Step 7.1: Use the program to fill a grain;

[0052] Step 7.2 calculates and selects the optimal Boundary IoU metric to evaluate the matching degree between the object boundary in the segmentation result and the true boundary, and measures the boundary accuracy.

[0053] Step 7.3 loops to fill all the grains in one layer of metallographic image.

[0054] Preferably, after obtaining and outputting the complete metallographic images of all layers in sequence in step 8, a step of verifying the filling results is also included. The verification method uses three-dimensional reconstruction technology to ensure that the filled grain boundaries meet the overall structure and performance requirements of the crystal. If the verification results do not meet the requirements, the filling algorithm is dynamically optimized and adjusted based on the verification feedback.

[0055] Compared with the prior art, the method of filling two-dimensional grain boundary defects based on three-dimensional information of the present invention has at least the following beneficial effects:

[0056] 1. Improved accuracy of grain boundary reconstruction: Traditional two-dimensional grain boundary analysis may have problems with missing or incomplete information. However, this invention can more accurately restore the integrity of grain boundaries by combining three-dimensional information, avoiding the limitations of two-dimensional data.

[0057] 2. Improved the reliability of material property prediction: Grain boundaries play a vital role in material properties. By filling the missing grain boundaries with this method, the structural characteristics of the material can be more realistically reflected, thus providing a more accurate basis for the prediction and optimization of material properties.

[0058] 3. Reduced experimental and computational costs: By effectively filling in 3D information, the number of actual grain boundary experiments required is reduced, reducing experimental costs. At the same time, the improved data integrity during the calculation process optimizes simulation and computational efficiency.

[0059] 4. Improved data processing efficiency: This method can automatically process large amounts of crystal structure data, reduce the time for manual intervention and data processing, and improve scientific research efficiency.

[0060] 5. The present invention has wide applicability: The filling method of the present invention is not only applicable to metal materials, but can also be extended to the grain boundary reconstruction of different materials such as ceramics and semiconductors, and has strong applicability and versatility.

[0061] The following further illustrates a method for filling two-dimensional grain boundary defects based on three-dimensional information according to the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 : A flow chart of a method for filling two-dimensional grain boundary gaps based on three-dimensional information according to the present invention;

[0063] Figure 2 : A series of pictures showing a method for filling two-dimensional grain boundary gaps based on three-dimensional information according to the present invention;

[0064] Figure 3 : This is the image after histogram equalization of the method for filling two-dimensional grain boundary gaps based on three-dimensional information of the present invention;

[0065] Figure 4 : The present invention provides a method for filling two-dimensional grain boundary gaps based on three-dimensional information to remove noise from the image;

[0066] Figure 5 : The present invention provides a binarized image of a method for filling a two-dimensional grain boundary defect based on three-dimensional information;

[0067] Figure 6 : Schematic diagram of an iterative etching method for filling two-dimensional grain boundary defects based on three-dimensional information in the present invention;

[0068] Figure 7 : The present invention provides a method for filling two-dimensional grain boundary gaps based on three-dimensional information and a grain center marking point image;

[0069] Figure 8 : The present invention provides a binary complete grain boundary image based on a method for filling two-dimensional grain boundary gaps based on three-dimensional information. DETAILED DESCRIPTION

[0070] This invention provides a method for filling missing two-dimensional grain boundaries based on three-dimensional information. By combining three-dimensional information acquisition, missing region analysis and modeling, grain boundary filling algorithms, filling result verification, and optimization and adjustment, it achieves efficient and accurate grain boundary repair. This method not only improves the accuracy of grain boundary reconstruction and enhances the reliability of material property prediction, but also reduces experimental and computational costs, improves data processing efficiency, and has broad applicability. The invention primarily relates to the following aspects:

[0071] 1. 3D information acquisition

[0072] High-resolution two-dimensional imaging techniques, such as optical microscopy and scanning electron microscopy, are used to obtain three-dimensional structural information of crystalline materials. This information includes the morphology and location of grain boundaries, as well as specific information about missing parts, providing basic information support for subsequent grain boundary filling.

[0073] 2. Missing Area Analysis and Modeling

[0074] Based on the collected 3D information, computer image processing techniques, such as image segmentation and feature extraction, are used to analyze the missing grain boundary areas. Simultaneously, 3D modeling techniques, such as geometric modeling and physical modeling, are used to model the missing areas and accurately identify their specific location and shape.

[0075] 3. Grain boundary filling algorithm

[0076] A grain boundary filling algorithm based on 3D information was designed to automatically or semi-automatically generate missing grain boundary information. The algorithm's core objective is to infer a reasonable filling solution based on the morphological characteristics of the surrounding grain boundaries, such as curvature and orientation, as well as crystal structure rules, such as crystal symmetry and growth. This ensures that the filled grain boundary is consistent in morphology and structure with the surrounding grain boundaries.

[0077] 4. Filling result verification

[0078] 3D reconstruction techniques, such as volume rendering and surface reconstruction, are used to verify the infill results. By comparing the infilled grain boundaries with the overall crystal structure, we ensure that the infilled grain boundaries meet the overall structural and performance requirements of the crystal. If the verification results do not meet the requirements, the infill algorithm is dynamically optimized and adjusted based on the verification feedback until the infill quality reaches the expected level.

[0079] 5. Optimization and Adjustment

[0080] During the infill process, the algorithm is dynamically optimized based on the infill results. This optimization includes parameter adjustment and optimization of the algorithm structure. Through continuous optimization and adjustment, the accuracy and efficiency of the infill algorithm are improved, ensuring the stability and reliability of the infill quality.

[0081] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0082] like Figure 1 As shown, combined with Figure 2-Figure 8 The schematic diagram of the present invention shown is a method for filling two-dimensional grain boundary defects based on three-dimensional information, comprising:

[0083] Step 1: Read in a series of metallographic image stacks of the sample (such as Figure 2 As shown, this operation can generate a series of metallographic image stacks to provide basic data for subsequent processing):

[0084] A stack of 132 sample metallographic images with a size of 105 × 880 × 390 μm was read in, and all metallographic images were benchmarked and registered to correlate spatial information.

[0085] Step 1.1: After heat treatment, the sample is mounted, ground, marked with diamond points, polished, etched, and photographed to obtain the first metallographic image. The polishing, etching, and photographing operations are repeated to obtain the second, third, and other series of metallographic images of the same area.

[0086] In step 1.2, during the processing of specific areas, ImageJ software is used to perform rigorous image processing operations on a series of grain metallographic images. Specifically, using diamond markers as references, plugins, registration, and SIFT commands are used to ensure that all images are accurately aligned in the upper and lower positions to achieve consistency and accuracy in image analysis. By accurately cropping the required area, a series of image stacks can be generated. (See Figure 2 )

[0087] Step 2: Hand-draw the complete first layer of metallographic diagram:

[0088] All metallographic images were preprocessed.

[0089] Step 2.1 Manually improve the first metallographic image to ensure the integrity of its grain boundaries. The grain boundary filling work of all subsequent metallographic images can be completed automatically by computer;

[0090] Step 2.2: grayscale processing, image enhancement, and image smoothing filtering are performed on all metallographic images;

[0091] Step 2.2.1 Grayscale processing uses floating point calculation method;

[0092] Step 2.2.2 Image enhancement uses the equalization technique formula;

[0093] Step 2.2.3 Image smoothing filtering uses Gaussian filtering and wavelet transform.

[0094] Step 3: Remove noise and perform binarization before thinning ( Figure 4 Shows the image effect after removing noise. Figure 5 A binary image is presented, and after these processing, the image is more conducive to subsequent analysis).

[0095] Step 3.1 Use the imfill function to eliminate the noise in the metallographic image;

[0096] Step 3.2: Use the formula to perform binarization operation on the metallographic image;

[0097] Step 3.3: Use MATLAB to refine the undesirable phenomena such as inconsistent grain boundary width and burrs.

[0098] Step 4: Iteratively etch each grain in the metallographic image to find the grain center and obtain multiple seed record marking points ( Figure 6 The iterative erosion process is shown. Figure 7 The final grain center mark point image is obtained. These results lay the foundation for further analysis of grain relationships).

[0099] Step 4.1 sets different thresholds, disk, and performs multi-scale iterative etching on the given image while maintaining the geometric features of the grain boundary. During this process, each time an etching operation is completed, the grain boundary will shrink inward by one layer until the grain completely disappears, and the grain center point is obtained.

[0100] In step 4.2, the grain boundary curvature is described using the mathematical model Gaussian curvature or mean curvature, the degree of curvature is calculated through the local grain boundary geometric characteristics, and the shape and surface structure of the filling area are inferred.

[0101] Step 4.2.1 uses the local grain boundary tangent plane equation and numerical methods to calculate the curvature, dynamically adjust the shape of the filling area, and combine the crystal symmetry and space group information to ensure that the lattice structure of the filling area is consistent with the surrounding area.

[0102] Step 4.2.2: Use a clustering algorithm to cluster the crystal structure regions and analyze the similarities and differences between clusters. Combining the clustering results with physical properties, a data-driven similarity assessment framework is constructed. By adjusting the similarity threshold, the properties of similar grains can be accurately predicted or filled in.

[0103] In step 4.2.3, for the crystal structure, similarity metrics such as geometric distance, phase similarity, and energy minimization can be used to assign a similarity score to each grain, influencing the infill strategy. A dynamic adjustment mechanism adjusts the infill threshold and similarity criteria in real time based on the grain distribution characteristics. Incremental learning is used to optimize parameter configurations as data accumulates and the algorithm iterates.

[0104] Step 4.3 records the center point of each grain before it disappears, and obtains multiple seed recording mark points.

[0105] Step 5: Perform eight-direction tracking on each grain to calculate the adjacent grain ratio parameters.

[0106] Step 5.1: Implement a tracking strategy in eight directions (45° apart) for each marker point;

[0107] In step 5.2, ensure that the line starts from the marked point and continues to track until the grain boundary is encountered. During this period, the length of each line segment is recorded and the scale parameter is calculated.

[0108] Step 6: Perform edge detection and extraction on the first layer of the complete metallographic image and scale it accordingly;

[0109] Step 6.1: Perform edge detection on each grain in the first layer of the complete metallographic image to extract the outline;

[0110] Step 6.2: Scale according to the ratio parameters of the two adjacent layers.

[0111] Step 7: Fill each grain of the second layer metallographic diagram in turn to obtain a complete second layer metallographic diagram;

[0112] Step 7.1: Use the program to fill a grain;

[0113] Step 7.2 calculates and selects the optimal Boundary IoU metric to evaluate the matching degree between the object boundary in the segmentation result and the true boundary, and measures the boundary accuracy.

[0114] Step 7.3 loops to fill all the grains in one layer of metallographic image.

[0115] Step 8: Obtain and output complete metallographic images of all layers in sequence. Figure 8 The final output is a binary complete grain boundary image, which reflects the achievement of grain boundary filling achieved by the present invention.

[0116] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A method for filling two-dimensional grain boundary defects based on three-dimensional information, characterized in that: The following steps are involved: A series of metallographic images of the sample are acquired using high-resolution two-dimensional imaging technology. All metallographic images are then benchmarked and registered based on specific landmarks using an image registration algorithm to accurately correlate spatial information. The image registration algorithm utilizes specialized image processing software to achieve precise positioning. The grain boundaries of the first metallographic layer are hand-drawn to complete them. Then all metallographic images are pre-processed. The pre-processing operations include grayscale processing using floating-point calculation method, image enhancement based on equalization technology formula, and image smoothing filtering combining Gaussian filtering and wavelet transform to ensure that the image quality meets the requirements of subsequent processing. Use MATLAB image processing functions to eliminate noise in metallographic images: perform binarization operations according to a predetermined formula, and then use software tools to refine existing defects such as inconsistent grain boundary widths and burrs to obtain clear grain boundary images; By setting different threshold disks, a multi-scale iterative etching operation is performed on each grain in the metallographic image while maintaining the geometric characteristics of the grain boundary. The grain center points are obtained at a specific iteration state before each grain completely disappears, and these center points are recorded as seed record points to provide a basis for subsequent analysis, which includes:

1. Set different thresholds disk and perform multi-scale iterative etching on the given image while maintaining the geometric features of the grain boundary. During this process, each time an etching operation is completed, the grain boundary will shrink one layer inward until the grain completely disappears in an iterative state, and the grain center point is obtained.

2. Use the mathematical model Gaussian curvature or mean curvature to describe the grain boundary curvature, calculate the degree of curvature through the local grain boundary geometric characteristics, and infer the shape of the filling area and the surface structure; 2.1 The curvature is calculated by using the local grain boundary tangent plane equation and numerical methods, and the shape of the filling area is dynamically adjusted. In combination with crystal symmetry and space group information, the lattice structure of the filling area is ensured to be consistent with the surrounding area. 2.2 Use clustering algorithms to cluster crystal structure regions and analyze the similarities and differences between clusters; combine clustering results with physical properties to build a data-driven similarity assessment framework; and adjust the similarity threshold to accurately predict or fill in the characteristics of similar grains; 2.3 Based on the crystal structure, a similarity metric is used to assign a similarity score to each grain, which affects the filling strategy. A dynamic adjustment mechanism is used to adjust the filling threshold and similarity standard in real time based on the grain distribution characteristics. Incremental learning is used to optimize parameter configuration as data accumulates and the algorithm iterates.

3. Record the center point of each grain before it disappears to obtain multiple seed recording mark points; For each seed record mark point, a tracking strategy of eight directions at 45° intervals is implemented. Starting from the mark point, the tracking continues until the grain boundary is encountered. During this process, the length of each line segment is accurately recorded, and the ratio parameters of adjacent grains are then calculated to determine the relative relationship between the grains. Perform edge detection and contour extraction on each grain in the first layer of the complete metallographic image, and perform scaling operations based on the proportional parameters calculated between two adjacent layers to provide an accurate template for subsequent filling. The program is used to fill each grain of the second layer metallographic diagram in turn. During the filling process, intelligent judgment and operation are performed based on the results of the previous calculation and processing. The program is cycled until all the grains of the layer are filled and a complete metallographic diagram of the second layer is obtained. The above filling process is executed cyclically, and the metallographic images of subsequent layers are processed in sequence until the complete metallographic images of all layers are output, thereby achieving comprehensive and automatic filling of grain boundaries and ensuring the accuracy and consistency of the filling results.

2. A method for filling two-dimensional grain boundary defects based on three-dimensional information, characterized in that: include: Step 1: Read in a series of metallographic image stacks of the sample and perform benchmark registration on all metallographic images to correlate spatial information; Step 2: Complete the first layer of complete grain boundary metallographic diagram by hand-drawing and pre-process all metallographic diagrams; Step 3: Remove noise and perform binarization before performing refinement processing; Step 4: Iteratively etch each grain in the metallographic image to find the grain center and obtain multiple seed record marking points. Step 4.1 sets different thresholds, disk, and performs multi-scale iterative etching on the given image while maintaining the geometric features of the grain boundary. During this process, each time an etching operation is completed, the grain boundary will shrink inward by one layer until the grain completely disappears, and the grain center point is obtained. Step 4.2 uses a mathematical model, Gaussian curvature or mean curvature, to describe the grain boundary curvature, calculates the degree of curvature based on the local grain boundary geometric characteristics, and infers the shape and surface structure of the filled area; Step 4.2.1: Use the local grain boundary tangent plane equation and numerical methods to calculate the curvature, dynamically adjust the shape of the filled area, and combine crystal symmetry and space group information to ensure that the lattice structure of the filled area is consistent with the surrounding area; Step 4.2.2: Use a clustering algorithm to cluster the crystal structure regions and analyze the similarities and differences between clusters. Combine the clustering results with physical properties to build a data-driven similarity assessment framework. Adjust the similarity threshold to accurately predict or fill in the properties of similar grains. Step 4.2.3: Based on the crystal structure, a similarity metric is used to assign a similarity score to each grain, which affects the filling strategy. A dynamic adjustment mechanism is used to adjust the filling threshold and similarity criteria in real time based on the grain distribution characteristics. Incremental learning is used to optimize the parameter configuration as data accumulates and the algorithm iterates. Step 4.3: Record the center point of each grain before it disappears to obtain multiple seed recording mark points; Step 5: Perform eight-direction tracking on each grain to calculate the proportional parameters of its adjacent layers; Step 6: Perform edge detection on each grain in the first layer of the complete metallographic image to extract the outline and then scale it accordingly; Step 7: Fill each grain of the second layer metallographic diagram in turn to obtain a complete second layer metallographic diagram; Step 8: Obtain complete metallographic images of all layers in sequence and output them.

3. The method for filling two-dimensional grain boundary defects based on three-dimensional information according to claim 2, characterized in that: The step 1 includes: Step 1.1: After heat treatment, the sample is subjected to operations including mounting, grinding, diamond marking, polishing, etching, and photographing to obtain the first metallographic image. The polishing, etching, and photographing operations are repeated to obtain a series of metallographic images of the same area. In step 1.2, during the region-specific processing, ImageJ software was used to perform rigorous image processing on a series of metallographic images. Using diamond-shaped markers as references, plugins, registration, and SIFT commands were used to ensure precise vertical registration of all images for consistent and accurate image analysis. A series of image stacks were generated by precisely cropping the desired region.

4. The method for filling two-dimensional grain boundary defects based on three-dimensional information according to claim 2, characterized in that: The step 2 includes: Step 2.1: Manually improve the first metallographic image to ensure the integrity of its grain boundaries. The grain boundary filling work of all subsequent metallographic images can be completed automatically by computer. Step 2.2: grayscale processing, image enhancement, and image smoothing filtering are performed on all metallographic images; Step 2.2.1 Grayscale processing uses floating point calculation method; Step 2.2.2 Image enhancement uses the equalization technique formula; Step 2.2.3 Image smoothing filtering uses Gaussian filtering and wavelet transform.

5. The method for filling two-dimensional grain boundary defects based on three-dimensional information according to claim 2, characterized in that: The step 3 includes: Step 3.1 Use the imfill function to eliminate the noise in the metallographic image; Step 3.2: Use the formula to perform binarization operation on the metallographic image; Step 3.3 uses MATLAB to refine the existing undesirable phenomena including inconsistent grain boundary width and burrs.

6. The method for filling two-dimensional grain boundary defects based on three-dimensional information according to claim 2, characterized in that: The step 5 includes: Step 5.1: Implement a tracking strategy for each marker point in eight directions at 45° intervals. In step 5.2, ensure that the line starts from the marked point and continues to track until the grain boundary is encountered. During this period, the length of each line segment is recorded and the scale parameter is calculated.

7. The method for filling two-dimensional grain boundary defects based on three-dimensional information according to claim 2, characterized in that: The step 6 includes: Step 6.1: Perform edge detection on each grain in the first layer of the complete metallographic image to extract the outline; Step 6.2: Scale according to the ratio parameters of the two adjacent layers.

8. The method for filling two-dimensional grain boundary defects based on three-dimensional information according to claim 2, characterized in that: The step 7 includes: Step 7.1: Use the program to fill a grain; Step 7.2 calculates and selects the optimal Boundary IoU metric to evaluate the matching degree between the object boundary in the segmentation result and the true boundary, and measures the boundary accuracy; Step 7.3 loops to fill all the grains in one layer of metallographic image.

9. The method for filling two-dimensional grain boundary defects based on three-dimensional information according to claim 2, characterized in that: After obtaining and outputting the complete metallographic images of all layers in sequence in step 8, the step of verifying the filling results is also included. The verification method uses three-dimensional reconstruction technology to ensure that the filled grain boundaries meet the overall structure and performance requirements of the crystal. If the verification results do not meet the requirements, the filling algorithm is dynamically optimized and adjusted based on the verification feedback.

Images