Misalignment error correction model construction method, misalignment error correction method and system
By constructing an offset error correction model and a residual neural network, the problem of offset error affecting the high-precision detection of large numerical aperture spherical optical elements was solved, and high-precision correction and surface shape detection were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SICHUAN UNIV
- Filing Date
- 2026-02-27
- Publication Date
- 2026-05-29
AI Technical Summary
Existing technologies are insufficient to effectively eliminate the effects of misalignment errors on large numerical aperture spherical optical elements in high-precision detection, resulting in inadequate detection accuracy.
By constructing an offset error correction model, a training dataset is generated using a Fizeau-type spherical interferometer system, and a residual neural network is used to train the neural network to output the offset error type, error amount, and error direction, thereby achieving high-precision correction.
It improves the accuracy of misalignment error correction, effectively eliminates misalignment errors in high-precision detection, and obtains high-precision surface shape information.
Smart Images
Figure CN122116041A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of optical element interferometry technology, and more specifically, to a method for constructing an offset error correction model, an offset error correction method, and a system. Background Technology
[0002] With the development of optical technology and technologies in various fields, the requirements for the surface shape of various optical components are becoming increasingly stringent. Spherical optical components are widely used in various scenarios due to their advantages such as simple fabrication and low cost. To obtain high-precision components that meet manufacturing requirements, high-precision inspection technology is essential. Commonly used surface shape inspection techniques can be divided into two types: traditional methods and interferometric inspection methods. Traditional methods include visual inspection, contour analysis, and knife-edge analysis. Compared to traditional methods, interferometric inspection methods are widely used in high-precision surface shape inspection due to their non-contact measurement, high accuracy, and high efficiency.
[0003] In actual interferometric testing, in addition to environmental influences and errors caused by problems with the device itself, the misalignment error of the spherical surface under test can also have a significant impact on the measurement results. The misalignment error of the spherical surface refers to the deviation between the actual position and the ideal position of the spherical surface under test during the measurement process. The misalignment error will cause the measured interferogram to contain not only the surface shape information of the spherical surface under test, but also the misalignment error information, which is not conducive to the judgment of the surface shape. Common misalignment errors include: (1) the vertex of the spherical surface under test has a small angle of tilt around the optical axis; (2) the spherical surface under test has a certain lateral offset relative to its ideal position; (3) the spherical surface under test has a certain axial offset relative to its ideal position, that is, the wavefront defocusing situation. The traditional misalignment error correction method is to eliminate the DC term, tilt term and defocusing term introduced by the misalignment error by fitting the measured surface shape information with Zernike. This method is simple and fast, and does not require knowledge of the radius of curvature (R) and numerical aperture (NA) of the spherical surface under test. However, this method is only suitable for situations where the NA of the spherical surface under test is small and the measurement accuracy requirement is not high.
[0004] Therefore, when the NA of the sphere under test increases and the accuracy requirement increases, the existing correction methods are unable to eliminate the influence of misalignment error, which in turn affects the judgment of the surface shape of the optical element and makes it difficult to complete high-precision detection. Summary of the Invention
[0005] The purpose of this invention is to provide a method for constructing an offset error correction model, an offset error correction method, and a system, which can improve the accuracy of offset error correction for optical elements with large NA spherical surfaces.
[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution:
[0007] In a first aspect, the present invention provides a method for constructing an offset error correction model, the method comprising:
[0008] Obtain sample fringe patterns free from misalignment errors;
[0009] By applying optical path differences of different types of misalignment errors to the sample fringe patterns, adjusting the error amount and direction of the sample fringe patterns, multiple sample interferograms are obtained. Each sample interferogram is then mapped one-to-one with its corresponding error amount, error direction, and misalignment error type to obtain the training dataset; where one sample interferogram corresponds to one type of misalignment error.
[0010] The neural network to be trained is trained using the training dataset, and the misalignment error correction model is obtained when the loss function converges.
[0011] In one implementation scheme, a Fizeau-type spherical interferometer system is used to perform numerical simulations of each type of misalignment error in terms of error magnitude and error direction.
[0012] In one implementation, the misalignment error types include defocus fringes in the z-direction, translational and rotational fringes in the x-direction, and translational and rotational fringes in the y-direction.
[0013] In one implementation, the neural network includes a residual neural network and three fully connected neural networks connected in parallel with the output of the residual neural network; wherein the three fully connected neural networks are used to process the misalignment error type classification, error quantity calculation and error direction prediction, respectively.
[0014] In one implementation, the loss function is a weighted sum of the misalignment error type classification loss, the error quantity calculation loss, and the error direction prediction loss.
[0015] In one implementation, the misalignment error type classification loss and error direction prediction loss are cross-entropy loss functions, and the error quantity calculation loss is the mean square error (MSE) loss.
[0016] A second aspect of the present invention provides an offset error correction method for correcting offset errors occurring in the surface shape detection process of spherical optical elements, the method comprising:
[0017] Obtain the interference pattern to be corrected, which contains misalignment error, generated during the surface shape detection process of the spherical optical element;
[0018] Based on the interferogram to be corrected and the pre-constructed misalignment error correction model, the misalignment error type, error amount, and error direction of the interferogram to be corrected are determined; wherein, the misalignment error correction model is constructed using a misalignment error correction model construction method provided in the first aspect of the present invention;
[0019] Misalignment error correction is performed according to the misalignment error type, error amount, and error direction.
[0020] A second aspect of the present invention provides a system for constructing an offset error correction model, the system comprising:
[0021] The sample acquisition unit is used to acquire sample fringe patterns that do not have offset errors;
[0022] The training set determination unit is used to apply optical path differences of different types of misalignment errors to the sample fringe patterns, adjust the error amount and error direction of the sample fringe patterns, obtain multiple sample interferograms, and map each sample interferogram to its corresponding error amount, error direction and misalignment error type to obtain the training dataset; wherein, one sample interferogram corresponds to one type of misalignment error.
[0023] The model training unit uses the training dataset to train the neural network to be trained, and obtains the misalignment error correction model when the loss function converges.
[0024] In one implementation, the misalignment error types include defocus fringes in the z-direction, translational and rotational fringes in the x-direction, and translational and rotational fringes in the y-direction.
[0025] A fourth aspect of the present invention provides an misalignment error correction system for correcting misalignment errors occurring in the surface shape detection process of spherical optical elements, the system comprising:
[0026] The interferogram acquisition unit is used to acquire the interferogram to be corrected, which contains misalignment errors generated by the spherical optical element during the surface shape detection process.
[0027] An error determination unit is used to determine the type, magnitude, and direction of the misalignment error of the interferogram to be corrected based on the interferogram to be corrected and a pre-built misalignment error correction model; wherein the misalignment error correction model is constructed using a misalignment error correction model construction method provided in the first aspect of the present invention;
[0028] The correction unit is used to correct the misalignment error according to the misalignment error type, error amount, and error direction.
[0029] Compared with the prior art, the present invention has the following beneficial effects:
[0030] The technical solution provided by this invention first applies optical path differences of different types of misalignment errors to sample fringe patterns, then adjusts the error amount and direction of the sample fringe patterns to obtain multiple sample interferograms. These sample interferograms are then mapped one-to-one with their corresponding error amounts, directions, and misalignment error types to obtain a training dataset. Finally, the training dataset is used to train the neural network to be trained. When the loss function converges, an misalignment error correction model capable of outputting the misalignment error type, amount, and direction is obtained. The trained neural network can output the misalignment error type, amount, and direction, with a very small error between the error amount and the true error amount, thus improving the accuracy of misalignment error correction. Attached Figure Description
[0031] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:
[0032] Figure 1 This is a flowchart illustrating a method for constructing an offset error correction model according to an embodiment of the present invention.
[0033] Figure 2 A schematic diagram illustrating the adjustment error provided in an embodiment of the present invention;
[0034] Figure 3 A flowchart illustrating an offset error correction method provided in an embodiment of the present invention;
[0035] Figure 4 This is a principle block diagram of an offset error correction model construction system provided in an embodiment of the present invention;
[0036] Figure 5 This is a schematic diagram of an offset error correction system provided in an embodiment of the present invention. Detailed Implementation
[0037] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.
[0038] It should be noted that the terms "comprising" or "may include" used in the various embodiments of this application indicate the presence of the claimed function, operation, or element, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in the various embodiments of this application, the terms "comprising," "having," and their cognates are intended only to indicate a specific feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or adding one or more combinations of the foregoing.
[0039] It should be understood that terms such as "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0040] When the NA of the sphere under test increases and the accuracy requirement increases, the existing correction methods are unable to eliminate the influence of misalignment error, which in turn affects the judgment of the surface shape of the optical element and makes it difficult to complete high-precision detection.
[0041] Therefore, in the technical solution provided by this invention, firstly, optical path differences of different types of misalignment errors are applied to the sample fringe patterns. Then, the error amount and error direction of the sample fringe patterns are adjusted to obtain multiple sample interferograms. The sample interferograms are then mapped one-to-one with their corresponding error amounts, error directions, and misalignment error types to obtain a training dataset. Finally, the training dataset is used to train the neural network to be trained. When the loss function converges, an misalignment error correction model with the ability to output the misalignment error type, error amount, and error direction is obtained. The trained neural network can output the misalignment error type, error amount, and error direction, and the error amount is very close to the true error amount, thus improving the accuracy of misalignment error correction.
[0042] like Figure 1 As shown, Figure 1 This is a flowchart illustrating a method for constructing an offset error correction model according to an embodiment of the present invention. The method includes:
[0043] S101, Obtain a sample fringe pattern without misalignment error.
[0044] Specifically, the generation of sample stripe patterns without misalignment error can be achieved using existing technologies, and this embodiment will not provide further explanation.
[0045] S102, apply optical path difference of different types of misalignment error to the sample fringe pattern, adjust the error amount and error direction of the sample fringe pattern to obtain multiple sample interferograms, and match each sample interferogram with the corresponding error amount, error direction and misalignment error type to obtain the training dataset; wherein, one sample interferogram corresponds to one type of misalignment error.
[0046] In this embodiment, when misalignment error occurs, the center of curvature of the sphere to be measured is selected. Taking the origin as the reference point, and neglecting backlash error, the optical path difference at any point P(x,y,z) on the surface to be measured is: OPD=2*( - ), in, For the ideal optical path, The optical path length is given by the presence of error, and O is the center of the reference spherical wavefront.
[0047] When the sphere under test is in the ideal position, its center should coincide with O. A certain range of random error amount and direction are applied to each type of misalignment error. The corresponding sample interferograms for different types of misalignment errors are obtained according to the optical path difference formula. This achieves a one-to-one correspondence between each sample interferogram and its corresponding error amount, error type, and error direction, thus obtaining a training dataset of image-error amounts.
[0048] This invention employs a Fizeau-type spherical interferometer system to achieve numerical simulation of the error magnitude and direction for each type of misalignment error. Specifically, the Fizeau-type spherical interferometer system is an extremely important and commonly used interferometer in optical manufacturing and testing, specifically designed to measure the surface accuracy (aberrations, aperture number N, local error ΔN) and radius of curvature of spherical optical elements (such as lenses and spherical mirrors). Its core idea originates from the Fizeau interference principle: utilizing an air wedge formed between a standard reference sphere and the measured sphere to generate interference fringes of equal thickness for high-precision measurement. The Fizeau interference is a common-path interference. Its basic structure is as follows: a collimated beam of light passes through a beam splitter and is incident perpendicularly onto a reference plane. Part of the light is reflected from the lower surface of the reference plane (as the reference beam), and the other part passes through the reference plane, reaches the measured surface, and is reflected back (as the test beam). When the two beams meet again, interference occurs. The interference occurs between two physical surfaces (the reference plane and the measured surface), and the light paths almost completely overlap, therefore it is insensitive to vibration and air turbulence, and its structure is simple and stable.
[0049] The Fizeau-type spherical interferometry system mainly consists of a light source, an imaging system, a beam splitter, a collimating lens, a reference sphere, and the sphere under test. To ensure the clarity and stability of the interference fringes, a helium-neon laser with a wavelength of 632.8 nm is selected as the light source. This wavelength laser possesses excellent monochromaticity and coherence, providing a reliable optical basis for high-precision surface shape detection. The light emitted from the light source is collimated by the collimating lens, converting the emitted beam into a parallel beam with a flat wavefront and consistent propagation direction. This beam is then incident on the beam splitter, which splits it into two coherent beams. One beam is reflected by the reference sphere to become the reference beam, and the other beam is reflected by the sphere under test to become the test beam. The two beams interfere, which is captured and recorded by the imaging system. Since the surface shape information of the sphere is also recorded by the interference fringes, both the sphere under test and the reference sphere are chosen as ideal spheres without surface shape errors to better correct for errors. Under ideal measurement conditions, the actual position of the sphere under test coincides with its ideal position. At this time, the intensity distribution I of the interference field can be expressed as: ,in, , These are the light intensities of the two coherent beams, respectively. This represents the wavefront phase difference between two coherent beams. When the tested spherical surface is in an ideal position, the change in the distribution of interference fringes directly indicates the surface shape of the tested spherical surface.
[0050] However, in actual measurements, due to various unavoidable factors, the actual position of the sphere under test will deviate from the ideal position, resulting in misalignment error. In this case, the light intensity distribution I will deviate from the ideal model, and its expression will become:
[0051] ,in, The phase difference in the optical path introduced by the misalignment error. It is the wavelength of the light source, and OPD is the introduced optical path difference.
[0052] Given the diverse forms of misalignment errors, the need for large datasets to train neural networks, and the environmental influences on actual experimental data acquisition, this interferometric system randomly generates sample interference fringe patterns for various misalignment error types. It should be noted that, for example... Figure 2 As shown, the types of misalignment errors include defocus fringes in the z-direction, translational and rotational fringes in the x-direction, and translational and rotational fringes in the y-direction.
[0053] Compared to a sample fringe pattern without error, the intensity expression for a sample interferogram with misalignment error includes an additional optical path difference introduced by the error. The expression for this optical path difference differs depending on the type of misalignment error. For any misalignment error, the expression for the optical path difference depends on the magnitude of the error. By changing the values related to the error in the optical path difference formula, sample interferograms with different error amounts can be obtained. Furthermore, by changing the expression for the optical path difference and then changing the values related to the error, sample interferograms with other error types can be obtained.
[0054] The training dataset simulated 16,000 sample interferograms, with 3,200 sample interferograms for each error type. 12,000 samples were evenly selected as the training set, and 3,000 samples were used as the validation set to evaluate the network's training performance. After training, the remaining 1,000 samples were used to test the trained model and evaluate its generalization ability.
[0055] S103 uses the training dataset to train the neural network to be trained, and obtains the misalignment error correction model when the loss function converges.
[0056] Specifically, a neural network consists of an input layer, a convolutional layer, a pooling layer, and an output layer. The input layer generates a simulated interferogram, and the output layer is divided into three parts: an output of the type of misalignment error, an output of the error magnitude, and an output of the error direction.
[0057] Before training the neural network using the training data, data augmentation is performed on the simulation dataset to generate diverse samples, improving the network's performance and robustness. Different types of misalignment interference fringes have different characteristics and varying sensitivities to different data augmentation techniques. Therefore, different data augmentation techniques are applied to each type of fringe. For defocused fringes, random translation, random rotation, random horizontal flipping, random vertical flipping, and random noise are used for data augmentation. For translational and rotational fringes, random translation and random noise are used for data augmentation.
[0058] The augmented training dataset is input into the neural network for training, where the network learns the features of the interferogram for each type of error and its corresponding error magnitude. The convolutional neural network model used in this invention includes a feature extraction encoder and three task-specific fully connected decoders, simultaneously performing aberration type classification, target value regression, and sign prediction. The encoder uses a pre-trained ResNet-50 backbone network (i.e., a residual neural network, where all convolutional kernels are 3×3 or 1×1, with the initial convolutional layer using 7×7 kernels. Max pooling is used, with a pooling window size of 2×2). Compared to other neural networks, the ResNet-50 network solves the gradient vanishing problem in deep networks through residual connections, enabling the training of very deep networks without performance loss, making it suitable for multi-task learning. The ResNet-50 network mainly consists of four downsampling stages used to extract low- to high-level features of the interference fringes. A bridging layer is designed between the encoder and decoder, directly passing the encoder's feature vectors to the three task-specific fully connected decoders. The decoder consists of three independent fully connected neural networks, handling error offset type classification, target value regression (error magnitude), and sign prediction (error direction) respectively. The decoder output directly generates the classification, regression, and sign prediction results without upsampling. The neural network employs a multi-task loss function, combining classification loss, regression loss, and sign loss. The overall loss is a weighted sum of the three losses, where the offset error type classification loss and error direction prediction loss are cross-entropy loss functions, and the error magnitude calculation loss is the mean squared error (MSE) loss. The optimizer used for training employs AdamW, combining mixed-precision training and gradient accumulation to improve training stability.
[0059] like Figure 3 As shown, this embodiment of the invention also provides an offset error correction method for correcting offset errors that occur in spherical optical elements during surface shape detection. The method includes:
[0060] S301, Obtain the interference pattern to be corrected, which contains misalignment error, generated by the spherical optical element during the surface shape detection process;
[0061] S302, based on the interferogram to be corrected and the pre-constructed misalignment error correction model, determine the misalignment error type, error amount, and error direction of the interferogram to be corrected; wherein, the misalignment error correction model is based on the above... Figure 1 The method described above for constructing an offset error correction model is used to construct the model.
[0062] S303, perform offset error correction according to the offset error type, error amount and error direction.
[0063] A spherical surface with NA=0.9 and R=0.09m was selected for testing. Based on the aforementioned simulation of interference fringes, 1000 simulated interferograms with misalignment errors were generated. For each type of misalignment error, 200 samples were used for testing and correction. The misalignment error correction model constructed above was used to correct these samples. The results show that after correction using this model, the average RMS of the 200 interferograms with defocus errors is 0.0490 Å; the average RMS of the 200 interferograms with x and y direction offsets after correction are 0.1161 Å and 0.2103 Å, respectively; and the average RMS of the 200 interferograms with x and y direction rotation errors after correction are 0.0803 Å and 0.1182 Å, respectively. To more thoroughly eliminate the influence of errors, the corrected interferogram was further fine-tuned using Zernike fitting. After fine-tuning, the average RMS was only 0.000948, 0.008166, 0.013620, 0.001628, and 0.002504, which meets the requirements of high-precision detection.
[0064] Accordingly, this invention, through an offset error correction model capable of outputting offset error type, error quantity, and error direction, can perform high-precision correction on the system, obtaining high-precision surface shape information. This invention provides a high-precision correction method for offset errors in interferometric detection of large numerical aperture spherical surfaces.
[0065] like Figure 4 As shown, this embodiment of the invention also provides an offset error correction model construction system, the system comprising:
[0066] The sample acquisition unit 410 is used to acquire sample fringe patterns that do not have misalignment errors;
[0067] The training set determination unit 420 is used to apply optical path differences of different types of misalignment errors to the sample fringe patterns, adjust the error amount and error direction of the sample fringe patterns, obtain multiple sample interferograms, and map each sample interferogram to its corresponding error amount, error direction and misalignment error type to obtain the training dataset; wherein, one sample interferogram corresponds to one type of misalignment error.
[0068] The model training unit 430 uses the training dataset to train the neural network to be trained, and obtains the misalignment error correction model when the loss function converges.
[0069] This application provides an offset error correction model construction system, which is similar to the above-mentioned... Figure 1The method for constructing an offset error correction model shown is a technical solution based on the same inventive concept. Through the detailed description of the method for constructing an offset error correction model provided in the above embodiments, those skilled in the art can clearly understand the implementation process of an offset error correction model construction system in this embodiment. Therefore, for the sake of brevity, it will not be described again here.
[0070] Accordingly, in the technical solution provided by this invention, firstly, optical path differences of different types of misalignment errors are applied to the sample fringe patterns. Then, the error amount and error direction of the sample fringe patterns are adjusted to obtain sample interferograms with multiple misalignment error types. The sample interferograms are then mapped one-to-one with their corresponding error amounts, error directions, and misalignment error types to obtain a training dataset. Finally, the training dataset is used to train the neural network to be trained. When the loss function converges, an misalignment error correction model capable of outputting the misalignment error type, error amount, and error direction is obtained. The trained neural network can output the misalignment error type, error amount, and error direction, and the error amount is very close to the true error amount, thus improving the accuracy of misalignment error correction.
[0071] like Figure 5 As shown, this embodiment of the invention also provides an misalignment error correction system for correcting misalignment errors that occur in spherical optical elements during surface shape detection. The system includes:
[0072] Interference pattern acquisition unit 510 is used to acquire the interference pattern to be corrected that has misalignment error generated by the spherical optical element during the surface shape detection process;
[0073] The error determination unit 520 is used to determine the type, magnitude, and direction of the misalignment error of the interferogram to be corrected based on the interferogram to be corrected and a pre-built misalignment error correction model; wherein the misalignment error correction model is constructed using a misalignment error correction model construction method provided in the first aspect of the present invention;
[0074] The correction unit 530 is used to correct the misalignment error according to the misalignment error type, error amount and error direction.
[0075] An offset error correction system in this application embodiment is similar to the one described above. Figure 3 The offset error correction method shown is a technical solution based on the same inventive concept. Through the detailed description of the offset error correction method provided in the above embodiments, those skilled in the art can clearly understand the implementation process of an offset error correction system in this embodiment. Therefore, for the sake of brevity, it will not be described again here.
[0076] Accordingly, this invention, through an offset error correction model capable of outputting offset error type, error quantity, and error direction, can perform high-precision correction on the system, obtaining high-precision surface shape information. This invention provides a high-precision correction method for offset errors in interferometric detection of large numerical aperture spherical surfaces.
[0077] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for constructing an offset error correction model, characterized in that the method... include: Obtain sample fringe patterns free from misalignment errors; By applying optical path differences of different types of misalignment errors to the sample fringe patterns, adjusting the error amount and direction of the sample fringe patterns, multiple sample interferograms are obtained. Each sample interferogram is then mapped one-to-one with its corresponding error amount, error direction, and misalignment error type to obtain the training dataset; where one sample interferogram corresponds to one type of misalignment error. The neural network to be trained is trained using the training dataset, and the misalignment error correction model is obtained when the loss function converges.
2. The method for constructing an offset error correction model according to claim 1, characterized in that, Numerical simulations of the error magnitude and direction for each type of misalignment error were performed using a Fizeau-type spherical interferometer system.
3. The method for constructing an offset error correction model according to claim 1, characterized in that, The types of misalignment errors include defocus fringes in the z-direction, translational and rotational fringes in the x-direction, and translational and rotational fringes in the y-direction.
4. The method for constructing an offset error correction model according to claim 1, characterized in that, The neural network includes a residual neural network and three fully connected neural networks connected in parallel with the output of the residual neural network; wherein, the three fully connected neural networks are used to process the misalignment error type classification, error quantity calculation and error direction prediction, respectively.
5. The method for constructing an offset error correction model according to claim 4, characterized in that, The loss function is a weighted sum of the misalignment error type classification loss, the error quantity calculation loss, and the error direction prediction loss.
6. The method for constructing an offset error correction model according to claim 5, characterized in that, The offset error type classification loss and error direction prediction loss are cross-entropy loss functions, and the error quantity calculation loss is the mean square error (MSE) loss.
7. A method for correcting offset error, characterized in that, Methods for correcting misalignment errors that occur during surface shape inspection of spherical optical elements include: Obtain the interference pattern to be corrected, which contains misalignment error, generated during the surface shape detection process of the spherical optical element; Based on the interferogram to be corrected and the pre-constructed misalignment error correction model, the misalignment error type, error amount, and error direction of the interferogram to be corrected are determined; wherein, the misalignment error correction model is constructed using a misalignment error correction model construction method according to any one of claims 1 to 6; Misalignment error correction is performed according to the misalignment error type, error amount, and error direction.
8. A system for constructing an offset error correction model, characterized in that, The system includes: The sample acquisition unit is used to acquire sample fringe patterns that do not have offset errors; The training set determination unit is used to apply optical path differences of different types of misalignment errors to the sample fringe patterns, adjust the error amount and error direction of the sample fringe patterns, obtain multiple sample interferograms, and map each sample interferogram to its corresponding error amount, error direction and misalignment error type to obtain the training dataset; wherein, one sample interferogram corresponds to one type of misalignment error. The model training unit uses the training dataset to train the neural network to be trained, and obtains the misalignment error correction model when the loss function converges.
9. The offset error correction model construction system according to claim 8, characterized in that, The types of misalignment errors include defocus fringes in the z-direction, translational and rotational fringes in the x-direction, and translational and rotational fringes in the y-direction.
10. An offset error correction system, characterized in that, The system is used to correct misalignment errors that occur during surface shape detection of spherical optical elements. It includes: The interferogram acquisition unit is used to acquire the interferogram to be corrected, which contains misalignment errors generated by the spherical optical element during the surface shape detection process. An error determination unit is used to determine the type, magnitude, and direction of the misalignment error of the interferogram to be corrected based on the interferogram to be corrected and a pre-built misalignment error correction model; wherein the misalignment error correction model is constructed using a misalignment error correction model construction method according to any one of claims 1 to 6; The correction unit is used to correct the misalignment error according to the misalignment error type, error amount, and error direction.