Rapid positioning method for scanning electron microscope

Abstract

The invention discloses a rapid positioning method for a scanning electron microscope. The rapid positioning method comprises the following steps: acquiring a reference image and a target image of a sample; decomposing the reference image and the target image into a plurality of areas, and respectively extracting target points of each area in the reference image and the target image; matching the target point pairs in the same area of the reference image and the target image, calculating the deformation tensor of the area where the target point is located in the target image, and distinguishing the deformation type of each area in the target image; according to the local deformation type, affine or thin-plate spline transformation models are dynamically allocated to a plurality of areas of the target image; splicing the plurality of sub-models to obtain a complete three-dimensional model of the sample; based on the obtained three-dimensional model, mapping coordinates of a target point in real time; according to the method, the accurate sample three-dimensional model is obtained by using the transformation model according to the deformation tensors of different areas of the target image, so that high-precision reconstruction of the target point at the position of the warped edge curved surface is realized, and the positioning precision of the target point in the three-dimensional space is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of scanning electron microscopes, and particularly to a rapid positioning method for a scanning electron microscope. Background Art

[0002] A scanning electron microscope (SEM for short) is a microscope technology with high resolution and high magnification, and is widely used in the fields of materials science, biomedicine, electronic engineering, etc. The scanning electron microscope scans the surface of a sample and uses the signals generated by the interaction between the sample and the electron beam to obtain a microscopic image of the sample. The SEM can provide extremely detailed information on the surface of the sample, including morphology, composition, and structure, etc., and is an indispensable tool in modern scientific research and industrial inspection.

[0003] In scientific research, the microscopic morphology of samples has attracted more and more attention from researchers. The scanning electron microscope is a powerful tool for studying the microscopic morphology of objects. Sometimes, researchers need to perform in-situ observation on samples under the electron microscope, that is, mark some target points of interest during the first observation, and after processing under certain conditions (temperature, stress, chemical reaction), observe these target points again to determine the influence of external conditions on the microscopic morphology of the samples. However, after the samples (such as copper foils, stainless steel thin sheets) are processed under certain conditions (temperature, stress, chemical reaction), dynamic deformations (such as warping, swelling) may occur, so that the target points cannot be quickly found during the second observation.

[0004] Therefore, it is necessary to design a rapid positioning method for a scanning electron microscope to solve the problem that the target points cannot be quickly found during the second observation as described above. Summary of the Invention

[0005] The present invention overcomes the deficiencies of the prior art and provides a rapid positioning method for a scanning electron microscope.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a rapid positioning method for a scanning electron microscope, comprising the following steps:

[0007] Step S1, obtaining a reference image and a target image of the sample;

[0008] Step S2, decomposing the reference image and the target image into several regions through dynamic grid division, and respectively extracting the target points of each region in the reference image and the target image;

[0009] Step S3, matching the target point pairs in the same regions of the reference image and the target image, calculating the deformation tensor of the region where the target points are located in the target image, and distinguishing the deformation types of each region in the target image;

[0010] Step S4. Dynamically allocate affine or thin plate spline transformation models to several regions of the target image according to the local deformation type;

[0011] Step S5. Stitch together the several sub-models obtained in Step S4 to obtain a complete three-dimensional model of the sample;

[0012] Step S6. Based on the obtained three-dimensional model, map the coordinates of the target points in real time.

[0013] In a preferred embodiment of the present invention, in Step S1, the reference image is an image obtained by observing the sample for the first time at an inclination of ±5°, and the target image is an image obtained by observing the sample for the first time at an inclination of ±5°.

[0014] In a preferred embodiment of the present invention, in Step S2, decomposing the reference image and the target image into several regions includes: uniformly dividing the reference image and the target image obtained from the sample at the same inclination angle into uniform rectangular grids, identifying the structures in the images using the gradient magnitude threshold, and adjusting the initial grid boundaries.

[0015] In a preferred embodiment of the present invention, in Step S2, extracting the target points of each region includes: recording the region where the target point positions are located in the reference image, recording their coordinates, and determining the points corresponding to the target points in the reference image through descriptor matching within the grid region of the target image at the same angle.

[0016] In a preferred embodiment of the present invention, in Step S3, it includes the following sub-steps:

[0017] Step S31. Select all the target points in the reference image and the directly observed target points in the target image, match all the target points in the reference image with the directly observed target points in the target image, denote them as target point pairs, and determine the same regions of the reference image and the target image;

[0018] Step S32. Based on the matched target point pairs, calculate the displacement field u=(u x , u y , u z ) within the regions corresponding to the target point pairs, and predict the displacements of the unknown target points by weighting the distances between the known target points and the unknown target points in the target image, where u x =x2 - x1, u y =y2 - y1, u z =z2 - z1, (x1, y1, z1) and (y2, y2, z2) are the coordinates of the same target points in the reference image and the target image respectively. Based on the displacement field of the matched point pairs, obtain the displacement gradient tensor I is the identity matrix, is the gradient of the displacement vector;

[0019] Step S33: Calculate the displacement gradient tensor F. Perform polar decomposition on the deformation tensor F to obtain the rotation matrix R and the right stretch tensor U. Calculate the eigenvalues of U, which are the stretch ratios in the corresponding principal directions. Define the principal strains as: ε1 = λ1 - 1, ε2 = λ2 - 1, and take the maximum absolute principal strain ε max = max(|ε1|, |ε2|) as the deformation measure;

[0020] Step S34: Determine the deformation types of several regions according to the deformation measure. If ε max is less than the threshold, determine that this region has small deformation. If ε max is greater than or equal to the threshold, determine that this region has large deformation.

[0021] In a preferred embodiment of the present invention, in the step S4, for the regions with small deformation, use the affine transformation model, specifically where, is the linear transformation matrix, is the translation vector, is the three-dimensional coordinate of a certain target point in the reference image, is the three-dimensional coordinate of the corresponding point in the target image.

[0022] In a preferred embodiment of the present invention, in the step S4, for the regions with large deformation, use the thin plate spline transformation model, specifically where φ is the radial basis function (describing non-linear deformation), (x i , y i , z i ) are the control points (i.e., the matched target points), A is the affine transformation matrix, and w i is the weight.

[0023] In a preferred embodiment of the present invention, in step S5, it includes the following sub-steps:

[0024] Step S51: Obtain the sub-models of each region;

[0025] Step S52: Take the coordinate system when the reference image is not tilted as the global reference, with its z-axis perpendicular to the sample stage, and convert the parameters of each sub-model to the global coordinate system;

[0026] Step S53: For adjacent regions, extract the common control points at the boundaries and force their global coordinates to be consistent;

[0027] Step S54: Based on the dual-view images tilted by ±5°, calculate the disparity through stereo matching to restore the depth, merge the global coordinates of all matched point pairs, and form a sparse point cloud;

[0028] Step S55: Generate a continuous surface from the sparse point cloud, and then construct a three-dimensional model.

[0029] In a preferred embodiment of the present invention, in the step S6, based on the coordinates of the target points in the second observation image, through the transformation relationship from the target image to the three-dimensional model, the coordinates of the target points are mapped into the three-dimensional model in real time. By comparing the positions of the target points in the first and second observations, combined with the deformation tensor and the deformation type, the three-dimensional coordinates of the target points are dynamically adjusted.

[0030] The present invention solves the defects existing in the background art and has the following beneficial effects:

[0031] (1) The present invention provides a fast positioning method for a scanning electron microscope. By dynamically dividing an image into several regions, selecting an appropriate transformation model according to the deformation degree of each region, and combining the sub-models obtained from the transformation models, a seamless transition from local to global is achieved, a complete three-dimensional model is generated, the differences in local deformations are effectively captured, the deficiency of a global single model in describing complex deformations is avoided, and thus the local warping or bending characteristics can be accurately determined. At the same time, a high-precision three-dimensional basis is provided for real-time mapping of the coordinates of target points, thereby quickly positioning the target points, significantly improving the efficiency and reliability of in-situ observation of the scanning electron microscope.

[0032] (2) The present invention provides a fast positioning method for a scanning electron microscope. By classifying each region according to the deformation tensors of different regions and selecting the corresponding affine transformation model and thin plate spline transformation model according to the type of the region, while ensuring high precision, the calculation efficiency is significantly improved, the use of complex calculation models for each region is avoided, the overall calculation burden is reduced, and at the same time, the actual deformation conditions of each region can be accurately reflected. And during splicing, the deformation transition between different regions is natural, the splicing error is reduced, and the smoothness and consistency of the overall three-dimensional model are ensured, so as to achieve high-precision reconstruction of the target points at the position of the warped edge surface, and further improve the positioning accuracy of the target points in the three-dimensional space.

[0033] (3) The present invention provides a fast positioning method for a scanning electron microscope. By constructing a three-dimensional model of the second observed sample, not only can the three-dimensional spatial changes after the deformation of the sample be captured, but also the positioning of the target points can be dynamically adjusted through the three-dimensional model. Even if the target points have a large spatial offset, the three-dimensional model can still accurately find the new positions of the target points through geometric matching, thereby accurately positioning the positions of the target points, ensuring the traceability of the target points, and at the same time improving the repeatability and reliability of the experiment, and reducing the influence caused by operation errors or environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings;

[0035] Figure 1 It is a flowchart of a fast positioning method for a scanning electron microscope according to the present invention. Detailed implementation manners

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0037] Many specific details are set forth in the following description in order to fully understand the present invention, but the present invention can also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited by the specific embodiments disclosed below.

[0038] As Figure 1 shown, a fast positioning method for a scanning electron microscope includes the following steps:

[0039] Step S1, obtaining a reference image and a target image of the sample; the reference image is an image obtained by observing the sample for the first time at an inclination of ±5°, and the target image is an image obtained by observing the sample for the first time at an inclination of ±5°.

[0040] Among them, the reference image is the initial state of the sample without being affected by external conditions, and the target image is the processed state. By comparing the two, the degree of deformation can be quantified. By adjusting the tilt angle of the sample stage to +5°, the reference image is obtained, and the position of the target point to be observed is marked. After applying external conditions (such as heating to 300°C) to the sample, the deformation caused by thermal expansion is captured to obtain the target image.

[0041] Step S2, decomposing the reference image and the target image into several regions through dynamic grid division, and respectively extracting the target points of each region in the reference image and the target image;

[0042] In the step S2, decomposing the reference image and the target image into several regions includes: uniformly dividing the reference image and the target image obtained from the sample at the same inclination angle into uniform rectangular grids, identifying the structures in the image using the gradient amplitude threshold, and adjusting the initial grid boundary;

[0043] The extraction of target points for each region includes: recording the region where the target point positions in the reference image are located, recording their coordinates, and within the grid region of the target image at the same angle, determining the points corresponding to the target points in the reference image through descriptor matching.

[0044] Specifically, the reference image and the target image are divided into uniform rectangular grids, and the grid size is set according to the image resolution and the target scale. For example, for an image of 1024×1024 pixels, the initial grid size can be set to 128×128 pixels (a total of 64 grids).

[0045] Step S3: Match the target point pairs in the same regions of the reference image and the target image, calculate the deformation tensor of the region where the target points in the target image are located, and distinguish the deformation types of each region in the target image;

[0046] Step S4: Dynamically allocate affine or thin plate spline transformation models to several regions of the target image according to the local deformation types;

[0047] Step S5: Stitch the several sub-models obtained in Step S4 to obtain the complete three-dimensional model of the sample;

[0048] Step S6: Based on the obtained three-dimensional model, map the coordinates of the target points in real time.

[0049] In a preferred embodiment of the present invention, in the said Step S3, it includes the following sub-steps:

[0050] Step S31: Select all the target points in the reference image and the target points directly observed in the target image, match all the target points in the reference image with the target points directly observed in the target image, record them as target point pairs, and judge the same regions of the reference image and the target image;

[0051] For all the target points in the reference image and the target points directly observed in the target image, use the descriptor matching algorithm for pairing to find the target points at the corresponding same physical positions in the reference image and the target image, record them as target point pairs. Once the target point pair matching is completed, the region to which each target point pair belongs can be judged according to the grid division information. For example, if the target point in the reference image is in grid region A, the matching point in the target image should also be in the same region (i.e., grid region A).

[0052] Step S32: Based on the matched target point pairs, calculate the displacement field u=(u x , u y , u z ) in the region corresponding to the target point pairs, and predict the displacement of the unknown target points by weighting the distances between the known target points and the unknown target points in the target image, where u x= x2 - x1, u y = y2 - y1, u z = z2 - z1, (x1, y1, z1) and (y2, y2, z2) are the coordinates of the same target point in the reference image and the target image respectively. Based on the displacement field of the matching point pairs, the displacement gradient tensor is obtained I is the identity matrix is the gradient of the displacement vector

[0053] For the target points in the same region of the reference image and the target image recorded in step S2, the matching of the target points in the reference image and the target image is achieved through the descriptor matching algorithm. Since the sample observed for the second time has been processed, there will be a situation of warped edges and surfaces. If the target point is on this warped edge surface, it is difficult to observe it in the target image. The target points not observed in the target image are marked as missing points compared with the reference image, and the deformation tensor calculation under sparse data is used. Specifically, based on the known matching points in the region, the displacement of the unknown target point is predicted. Specifically, the radial basis function interpolation method is used to weight the distance between the known target point and the missing point to predict the displacement of the missing point: where ||x m - x i || represents the distance between the missing point and the known target point, w i represents the weight corresponding to the known target point, and the weight is calculated by methods such as the least squares method (confidence weight assignment). φ is the radial basis function. In this way, the vacancy of the displacement field is filled by RBF interpolation to ensure that there are enough data points in each region

[0054] Step S33: Calculate the deformation gradient through the spatial change rate of the displacement field, and decompose the deformation gradient into two parts: pure rotation and pure stretching. The rotation part describes the overall rotation of the sample (such as due to the offset of the sample stage), and the stretching part reflects the true deformation of the material itself (such as thermal expansion or plastic flow). Specifically, calculate the displacement gradient tensor F, and the polar decomposition result of the deformation tensor F is the rotation matrix R and the right stretch tensor U. Calculate the eigenvalues of U, which are the stretch ratios in the corresponding principal directions. Define the principal strains as: ε1 = λ1 - 1, ε2 = λ2 - 1, and take the maximum absolute principal strain ε max = max(|ε1|, |ε2|) as the deformation measure

[0055] By calculating the principal strain, the true deformation degree of the material can be accurately measured. A larger principal strain means that the material has undergone significant deformation, while a smaller principal strain may mean that the deformation of the sample is small or almost no deformation. In this way, the type in each region is determined, providing a basis for using different transformation models according to different types in the follow-up

[0056] Step S34: Determine the deformation types of several regions according to the deformation degree. If ε max is less than the threshold value, then determine that this region is a small deformation. If ε max is greater than or equal to the threshold value, then determine that this region is a large deformation, where the threshold value is preset according to the material type.

[0057] In the present invention, by first recording the target points in the same regions of the reference image and the target image, considering that since the sample for the second observation has been processed, there will be a situation of warped edges and curved surfaces. If the target point is on this warped edge and curved surface, it is difficult to observe it in the target image. By comparing the reference image, mark the target points not observed in the target image as missing points, and use the radial basis function (RBF) interpolation to predict the displacement of the missing points;

[0058] Accurately predict the displacement of the missing points through the displacement information of the known points, process the missing points or the target points that cannot be matched in the target image due to warping, ensure that each region has enough data points to describe the deformation within the region, and avoid inaccurate estimation due to data sparsity.

[0059] For the target points in the same regions of the recorded reference image and the target image, pair them with each other, and then calculate the displacement field within each region. Then, obtain the displacement gradient tensor through the displacement field. The same applies to the missing points. Calculate the deformation gradient through the spatial change rate of the displacement field, decompose the deformation gradient into two parts: pure rotation and pure stretching. The rotation part describes the overall rotation of the sample (such as due to the offset of the sample stage), and the stretching part reflects the true deformation of the material itself (such as thermal expansion or plastic flow). Define the principal strain. By calculating the principal strain, the true deformation degree of the material can be accurately measured. A larger principal strain means that the material has undergone significant deformation, while a smaller principal strain may mean that the deformation of the sample is smaller or almost no deformation. Based on this, determine the type within each region, thus providing a basis for using different transformation models according to different types in the subsequent process.

[0060] In the present invention, in the step S4, for the regions with small deformation, use the affine transformation model, specifically where, is the linear transformation matrix, is the translation vector, is the three-dimensional coordinate of a certain target point in the reference image, is the three-dimensional coordinate of the corresponding point in the target image.

[0061] In the step S4, for the regions with large deformation, use the thin plate spline transformation model, specifically where, φ is the radial basis function (describing non-linear deformation), (x i , y i , z i) is the control point (i.e., the matched target point), A is the affine transformation matrix, and w i is the weight.

[0062] In step S4, the deformation tensors of different regions classify each region, and corresponding affine transformation models and thin plate spline transformation models are selected according to the type of the region. The dynamic model selection method based on the deformation type can accurately adapt to the deformation characteristics of different regions. For small deformation regions, the affine transformation model is used to describe linear changes such as translation, rotation, and scaling, which can quickly calculate and reduce the computational burden. For large deformation regions, the thin plate spline transformation model can accurately capture complex non-linear deformations (such as warping and surface deformation), avoiding fitting errors caused by using overly simplified linear models. Therefore, selecting the corresponding affine transformation model and thin plate spline transformation model according to the type of the region not only ensures the efficient processing of small deformation regions but also guarantees the high-precision modeling of large deformation regions, improving the accuracy of the overall 3D reconstruction;

[0063] At the same time, the dynamic selection of the transformation model improves the computational efficiency. By using different transformation models for different regions, it is possible to avoid using computationally intensive non-linear models (such as the thin plate spline model) for all regions, thereby reducing unnecessary consumption of computational resources while ensuring high precision, making the overall modeling process more efficient, saving not only time but also maintaining a high computing speed when dealing with large-scale data;

[0064] Moreover, selecting a suitable transformation model based on the deformation type can effectively optimize the model stitching process. Modeling different regions with different deformation types separately can ensure a natural transition between regions during stitching, avoiding problems such as seams or discontinuous deformations caused by mismatched transformation models, ensuring the overall consistency and smooth transition of the 3D model, and thus providing a more reliable and stable result. When dealing with complex deformation regions, it can ensure the smoothness and accuracy of the final model in all regions.

[0065] Therefore, for step S4, by classifying each region according to the deformation tensors of different regions and selecting the corresponding affine transformation model and thin plate spline transformation model, it significantly improves the computational efficiency while ensuring high precision, avoids using complex computational models for each region, reduces the overall computational burden, can accurately reflect the actual deformation of each region, and can make the deformation transition between different regions more natural during stitching, reducing stitching errors, ensuring the smoothness and consistency of the overall 3D model, thereby achieving high-precision reconstruction of the target points at the warped surface position, and further improving the positioning accuracy of the target points in 3D space.

[0066] In the present invention, in step S5, the following sub-steps are included:

[0067] Step S51: Obtain the sub-models of each region;

[0068] Step S52: Taking the coordinate system when the reference image is not tilted as the global reference, with its z-axis perpendicular to the sample stage, convert the parameters of each sub-model to the global coordinate system;

[0069] Step S53: For adjacent regions, extract the common control points at the boundary and force their global coordinates to be consistent;

[0070] Step S54: Based on the dual-view images tilted at ±5°, calculate the disparity through stereo matching to restore the depth, merge the global coordinates of all matching point pairs to form a sparse point cloud;

[0071] Step S55: Generate a continuous surface from the sparse point cloud and then construct a three-dimensional model.

[0072] In step S5, by independently modeling each region, according to the small deformation or large deformation of each region, select an appropriate affine transformation model or thin plate spline transformation model, and then generate several sub-models. For the several sub-models, use common control points to ensure accurate docking between different regions, avoid misalignment or distortion during splicing. Through depth recovery technology, convert the sparse point cloud in the target image into a continuous three-dimensional surface, ensure the smooth transition of the model, and then ensure high precision while avoiding the accumulation of errors. By splicing different transformation models into a unified three-dimensional model, it can truly and smoothly reflect the microscopic morphology and deformation of the sample, ensure the precise capture of details, and avoid the situation where it is difficult to accurately and quickly capture the target point on the warped surface.

[0073] In the present invention, in the said step S6, based on the coordinates of the target points in the second observation image, through the transformation relationship from the target image to the three-dimensional model, map the coordinates of the target points to the three-dimensional model in real time. By comparing the positions of the target points in the first and second observations, combined with the deformation tensor and the deformation type, dynamically adjust the three-dimensional coordinates of the target points.

[0074] By constructing the three-dimensional model of the second observed sample, the present invention can not only capture the three-dimensional spatial changes after the sample deformation, but also dynamically adjust the positioning of the target points through the three-dimensional model. Even if the target points have a large spatial offset, the three-dimensional model can still accurately find the new positions of the target points through geometric matching, thereby accurately positioning the positions of the target points, ensuring the retrospective nature of the target points, improving the repeatability and reliability of the experiment at the same time, and reducing the influence caused by operation errors or environmental changes.

[0075] Based on the inspiration of the ideal embodiments of the present invention, through the above description, relevant personnel can completely make various changes and modifications without departing from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and the technical scope must be determined according to the scope of the claims.

Claims

1. A rapid positioning method for a scanning electron microscope, characterized in that: The following steps are involved: Step S1, obtaining a reference image and a target image of a sample; Step S2, decomposing the reference image and the target image into a plurality of regions by dynamic grid division, and extracting target points in each region of the reference image and the target image respectively; Step S3, matching the target point pairs in the same area of ​​the reference image and the target image, calculating the deformation tensor of the area where the target point is located in the target image, and distinguishing the deformation type of each area in the target image; Step S4, dynamically assigning affine or thin plate spline transformation models to several regions of the target image according to the local deformation type; Step S5, splicing the several sub-models obtained in step S4 to obtain a complete three-dimensional model of the sample; Step S6: Based on the obtained three-dimensional model, the coordinates of the target point are mapped in real time.

2. A rapid positioning method for a scanning electron microscope according to claim 1, characterized in that: In step S1, the reference image is an image obtained by observing the sample for the first time at an inclination of ±5°, and the target image is an image obtained by observing the sample for the first time at an inclination of ±5°.

3. The rapid positioning method for a scanning electron microscope according to claim 1, characterized in that: In step S2, decomposing the reference image and the target image into several regions includes: uniformly dividing the reference image and the target image obtained by the same tilt angle sample into uniform rectangular grids, using the gradient amplitude threshold to identify the structure in the image, and adjusting the initial grid boundary.

4. The rapid positioning method for a scanning electron microscope according to claim 1, characterized in that: In step S2, extracting the target point in each area includes: recording the area of ​​the target point position in the reference image, recording its coordinates, and determining the point corresponding to the target point of the reference image by descriptor matching in the grid area of ​​the target image at the same angle.

5. The rapid positioning method for a scanning electron microscope according to claim 1, characterized in that: In the step S3, the following sub-steps are included: Step S31, selecting all target points in the reference image and target points directly observed in the target image, matching all target points in the reference image with target points directly observed in the target image, recording them as target point pairs, and determining the same area of ​​the reference image and the target image; Step S32: Based on the matched target point pairs, calculate the displacement field u in the area corresponding to the target point pairs = (u x ,u y ,u z ), and predict the displacement of the unknown target point by weighting the distance between the known target point and the unknown target point in the target image, where u x =x2-x1,u y =y2-y1,u z =z2-z1, (x1, y1, z1) and (y2, y2, z2) are the coordinates of the same target point in the reference image and the target image respectively. Based on the displacement field of the matching point pair, the displacement gradient tensor is obtained. I is the identity matrix, is the gradient of the displacement vector; Step S33, calculate the displacement gradient tensor F, decompose the deformation tensor F into the rotation matrix R and the right stretch tensor U, calculate the eigenvalue of U, corresponding to the stretch ratio in the main direction, define the principal strain as: ε1 = λ1-1, ε2 = λ2-1, take the maximum absolute principal strain ε max =max(|ε1|,|ε2|) as the deformation measure; Step S34: Determine the deformation type of several regions according to the deformation measure. If ε max If ε is less than the threshold, the region is judged to be a small deformation. max If it is greater than or equal to the threshold, the area is judged to be a large deformation.

6. A rapid positioning method for a scanning electron microscope according to claim 5, characterized in that: In step S4, the region with small deformation is transformed using an affine variation model, specifically: in, is the linear transformation matrix, is the translation vector, is the three-dimensional coordinate of a target point in the reference image, are the three-dimensional coordinates of the corresponding points in the target image.

7. The rapid positioning method for a scanning electron microscope according to claim 5, characterized in that: In step S4, the region with large deformation is transformed into a model using a thin plate spline, specifically: Where φ is the radial basis function (describing nonlinear deformation), (x i ,y i , z i ) is the control point (i.e. the matching target point), A is the affine transformation matrix, w i is the weight.

8. The rapid positioning method for a scanning electron microscope according to claim 1, characterized in that: In step S5, the following sub-steps are included: Step S51, obtaining a sub-model of each region; Step S52, taking the coordinate system of the reference image when it is not tilted as the global reference, with its z-axis perpendicular to the sample stage, converting the parameters of each sub-model to the global coordinate system; Step S53: For adjacent regions, extract common control points at the boundaries and force their global coordinates to be consistent; Step S54: Based on the dual-view images tilted at ±5°, calculate the disparity through stereo matching, restore the depth, and merge the global coordinates of all matching point pairs to form a sparse point cloud; Step S55: Generate a continuous surface from the sparse point cloud, and then construct a three-dimensional model.

9. The rapid positioning method for a scanning electron microscope according to claim 1, characterized in that: In step S6, based on the coordinates of the target point in the second observation image, the coordinates of the target point are mapped to the three-dimensional model in real time through the transformation relationship from the target image to the three-dimensional model. By comparing the positions of the target point in the first and second observations, combined with the deformation tensor and deformation type, the three-dimensional coordinates of the target point are dynamically adjusted.

Images