Method and device for removing aspheric measurement backhaul error based on graph-to-graph neural network

Abstract

The invention discloses an aspheric measurement return error removal method and device based on a graph-to-graph neural network. The method comprises the steps of building a virtual interference system, making a graph-to-graph return error removal data set, building a graph-to-graph neural network model, training the graph-to-graph neural network model and solving an actual aspheric surface shape error. According to the method, the computer is used for simulating the optical system of the non-zero compensation method and generating the data set, training of the graph-to-graph neural network is completed, convenience and rapidness are achieved, backhaul errors can be eliminated without any priori knowledge and preprocessing for interference wavefront data of a group of actual optical systems, and good universality is achieved; a graph-to-graph neural network is used for eliminating return errors, high-frequency information in an interference wavefront graph and surface shape distribution of a detected aspheric surface is reserved, and the detection precision is higher; in the solving process, inherent system errors do not exist, the solving precision does not depend on the modeling precision of the system, the complex adjustment and calibration process is avoided, and rapid and high-precision aspheric surface shape error measurement is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of optical measurement, and in particular to a method for removing return errors in aspheric surface measurement based on an image-to-image neural network, and a device for removing return errors in aspheric surface measurement based on an image-to-image neural network, which are mainly used for fast and high-precision measurement of aspheric surface shape errors. Background Art

[0002] Aspheric mirrors are optical elements whose surface shape is not a common sphere, but a complex curve designed according to specific needs. Unlike spherical mirrors, the curvature of aspheric mirrors changes at different positions. This aspheric design enables it to have better optical performance, can effectively reduce optical aberrations such as spherical aberration and astigmatism, and improve imaging quality. It is widely used in complex and precise optical systems such as high-end camera lenses, astronomical telescopes, and medical imaging equipment. Therefore, strict requirements are placed on the precise measurement of aspheric surface errors.

[0003] Among various measurement technologies, interferometry has the advantages of high precision, non-contact, and short measurement time. Non-zero compensation in interferometry means that the compensator cannot completely compensate for the aberration caused by the ideal aspheric surface, resulting in residual wavefront aberration, which causes the measurement light to be unable to return along the original path. In this way, the compensator and the measured mirror do not need to correspond one to one, which can reduce the design and adjustment requirements of the compensator. Non-zero compensation is currently the main method to expand the dynamic range and achieve high-precision measurement, but the design of residual error introduces return error to the optical system, which is extremely unfavorable for precise measurement when the aspheric surface error is large.

[0004] In order to eliminate the influence of return error, people have proposed the half method and the inverse optimization method based on ray tracing to eliminate the return error in the non-zero compensation interference method for non-spherical surface measurement. When the non-spherical surface shape error is large, the half method itself has a large system error. The accuracy of the inverse optimization method in solving the non-spherical surface shape error mainly depends on the modeling accuracy of the system, including the inaccuracy of the optical component parameters and the inconsistency between the component position and the actual system, the randomness of temperature, humidity, and air flow velocity, as well as the selection of the detector resolution and the number of Zernike polynomial terms in the inverse optimization algorithm, which will affect the accuracy of the final measurement results. In addition, the inverse optimization method requires multiple iterative optimizations of the optimization variables, which greatly reduces the efficiency of solving the surface shape error and is not suitable for rapid measurement in industrial production.

[0005] The Chinese patent application with application number 202010105329X is a technical solution proposed by the inventor. It is a method for eliminating return errors in high-precision non-spherical surface error measurement. Computer simulation is used to generate several sets of surface error data of the measured surface and interference wavefront data of the optical system. These data are one-dimensional matrices. A neural network for eliminating return errors is constructed, and the neural network for eliminating return errors is trained. The trained neural network is used to solve the non-spherical surface error and solve the one-dimensional matrix. Although this fully connected neural network can directly obtain the measurement result by inputting the Zernike polynomial coefficients of the interference wavefront and outputting the Zernike coefficients of the measured surface shape, its accuracy is also limited by the limited Zernike polynomial coefficients, resulting in the loss of some high-frequency information and reducing the accuracy of the measurement result. Summary of the invention

[0006] In order to overcome the defects of the prior art, the technical problem to be solved by the present invention is to provide a method for removing return errors in aspheric surface measurement based on graph-to-graph neural network, which has good universality, can eliminate return errors, and combined with the ray tracing function of optical software, reduce the loss of high-frequency information, and realize fast and high-precision measurement of aspheric surface errors.

[0007] The technical solution of the present invention is: the method for removing the return error of aspheric surface measurement based on graph-to-graph neural network comprises the following steps:

[0008] (1) Build a virtual interferometer system in the optical tracking software based on the interferometer optical path used in the actual aspheric surface error detection work;

[0009] (2) Produce a data set for image-to-image return error elimination. Apply random surface error distribution to the DUT in the optical tracing software, calculate the corresponding interference wavefront map, save the two-dimensional information calculated by the simulation software, repeat the simulation a large number of times, and produce a data set for image-to-image return error elimination.

[0010] (3) Building a graph-to-graph neural network model, building a graph-to-graph neural network model, using the interference wavefront image as the network data and the measured aspheric surface shape distribution as the network output, so that the network input dimension is the same as the dimension of the interference wavefront image, and the network output dimension is the same as the dimension of the surface shape error distribution;

[0011] (4) training a graph-to-graph neural network model, training the graph-to-graph neural network on a graph-to-graph return error elimination dataset to obtain a graph-to-graph return error elimination network model;

[0012] (5) Solve the actual aspheric surface shape error, put the interference wavefront image obtained in the actual detection work into the trained image-to-image return error elimination network model, and obtain the surface shape distribution result of the actual aspheric surface after eliminating the return error.

[0013] The present invention establishes a method and device for removing return errors in non-zero interference aspheric surface measurement based on a picture-to-picture neural network, through the steps of building a virtual interference system, producing a picture-to-picture return error elimination data set, building a picture-to-picture neural network model, training the picture-to-picture neural network model, and solving the actual aspheric surface shape error. The present invention utilizes a computer to simulate an optical system of a non-zero compensation method and generates a data set, completing the training of a graph-to-graph neural network, which is convenient and fast. For a set of interference wavefront data of an actual optical system, the return error in the actual optical system can be eliminated without any prior knowledge and preprocessing, and has good universality. The present invention uses a graph-generated graph neural network to eliminate the return error, retains the high-frequency information in the interference wavefront image and the measured aspheric surface shape distribution, and has higher detection accuracy. In the process of solving the aspheric surface shape error, there is no inherent system error and the solution accuracy does not depend on the modeling accuracy of the system, including the inaccuracy of optical element parameters and the inconsistency of element position with the actual system, the randomness of temperature, humidity, and airflow velocity, etc. will not affect the final result, avoiding complex assembly and calibration processes, and realizing fast and high-precision aspheric surface shape error measurement.

[0014] A device for removing return errors in aspheric surface measurement based on a graph-to-graph neural network is also provided, comprising:

[0015] Build a virtual interference system module, which is configured to build a virtual interference system in the optical tracing software according to the interference optical path used in the actual aspheric surface error detection work; create a picture-to-picture return error elimination data set module, which is configured to apply random surface error distribution to the measured object in the optical tracing software, calculate the corresponding interference wavefront map, save the two-dimensional information calculated by the simulation software, repeat the simulation in large quantities, and create a picture-to-picture return error elimination data set;

[0016] Build a graph-to-graph neural network model module, which is configured to build a graph-to-graph neural network model, with the interference wavefront image as the network data and the measured aspheric surface shape distribution as the network output, so that the network input dimension is the same as the dimension of the interference wavefront image, and the network output dimension is the same as the dimension of the surface shape error distribution;

[0017] A training graph-to-graph neural network model module is configured to train a graph-to-graph neural network on a graph-to-graph return error elimination dataset to obtain a graph-to-graph return error elimination network model;

[0018] The module for solving the actual aspheric surface shape error is configured to put the interference wavefront image obtained in the actual detection work into the trained image-to-image return error elimination network model to obtain the surface shape distribution result of the actual measured aspheric surface after eliminating the return error. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of the method for removing return errors in aspheric surface measurement based on graph-to-graph neural network according to the present invention.

[0020] Figure 2 It is the interference wavefront diagram obtained by the measured aspheric surface in the interferometric measurement.

[0021] Figure 3 It is the true value of the surface shape distribution of the measured aspheric surface.

[0022] Figure 4 It is the calculation result of the measured aspheric surface shape error output by the image-to-image neural network.

[0023] Figure 5 It is the difference between the aspheric surface error result output by the graph-to-graph neural network and the true value. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0025] It should be noted that the term "comprises" and any variations in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products or equipment.

[0026] like Figure 1 As shown, the method for removing the return error of aspheric surface measurement based on graph-to-graph neural network includes the following steps:

[0027] (1) Build a virtual interferometer system in the optical tracking software based on the interferometer optical path used in the actual aspheric surface error detection work;

[0028] (2) Produce a data set for image-to-image return error elimination. Apply random surface error distribution to the DUT in the optical tracing software, calculate the corresponding interference wavefront map, save the two-dimensional information calculated by the simulation software, repeat the simulation a large number of times, and produce a data set for image-to-image return error elimination.

[0029] (3) Building a graph-to-graph neural network model, building a graph-to-graph neural network model, using the interference wavefront image as the network data and the measured aspheric surface shape distribution as the network output, so that the network input dimension is the same as the dimension of the interference wavefront image, and the network output dimension is the same as the dimension of the surface shape error distribution;

[0030] (4) training a graph-to-graph neural network model, training the graph-to-graph neural network on a graph-to-graph return error elimination dataset to obtain a graph-to-graph return error elimination network model;

[0031] (5) Solve the actual aspheric surface shape error, put the interference wavefront image obtained in the actual detection work into the trained image-to-image return error elimination network model, and obtain the surface shape distribution result of the actual aspheric surface after eliminating the return error.

[0032] The present invention establishes a method and device for removing return errors in non-zero interference aspheric surface measurement based on a picture-to-picture neural network, through the steps of building a virtual interference system, producing a picture-to-picture return error elimination data set, building a picture-to-picture neural network model, training the picture-to-picture neural network model, and solving the actual aspheric surface shape error. The present invention utilizes a computer to simulate an optical system of a non-zero compensation method and generates a data set, completing the training of a graph-to-graph neural network, which is convenient and fast. For a set of interference wavefront data of an actual optical system, the return error in the actual optical system can be eliminated without any prior knowledge and preprocessing, and has good universality. The present invention uses a graph-generated graph neural network to eliminate the return error, retains the high-frequency information in the interference wavefront image and the measured aspheric surface shape distribution, and has higher detection accuracy. In the process of solving the aspheric surface shape error, there is no inherent system error and the solution accuracy does not depend on the modeling accuracy of the system, including the inaccuracy of optical element parameters and the inconsistency of element position with the actual system, the randomness of temperature, humidity, and airflow velocity, etc. will not affect the final result, avoiding complex assembly and calibration processes, and realizing fast and high-precision aspheric surface shape error measurement.

[0033] Preferably, in step (1), the virtual interferometer constructed by the optical tracing software is consistent with the actual detection optical path.

[0034] Preferably, in the step (2), random surface errors are added to the aspheric surface to be measured in the virtual interferometer in the form of Zernike surfaces, and the corresponding interference wavefront image is obtained after ray tracing, and the resolution of the interference wavefront image and the surface error of the aspheric surface to be measured is above 224×224.

[0035] Preferably, in step (3), the graph-to-graph neural network model is a U-Net image generation network, ensuring that the network has strong complexity and performance.

[0036] Preferably, in step (4), the training of the graph-to-graph neural network pays close attention to the changes in loss function and accuracy to prevent under-training or over-fitting of the network model.

[0037] Preferably, in step (5), the interference wavefront actually detected is scaled to the same size as the data set so as to be placed in the network to correctly solve the surface error.

[0038] Preferably, in step (1), Zemax software is used to build a Fizeau interferometer that is consistent with the actual optical path.

[0039] Preferably, in the step (2), 78 Zernike coefficients are randomly generated, and the surface shape PV generated by the coefficients is between 2λ and 0.05λ; the randomly generated Zernike polynomial coefficients are substituted into the aspheric surface to be measured in the form of a Zernike sag surface in the Zemax software, and the surface shape generated by this group of coefficients is used as the true value of the surface shape error for standby; the optical path tracing is performed to record the interference wavefront image; the true value of the aspheric surface shape error and the interference wavefront image are used as a group of data with a resolution of 224×224, and 6052 groups of data are obtained by repeating the experiment, which together constitute a return error elimination data set; in the step (3), a swin-unet network model is built, and the input and output dimensions are both 224x224.

[0040] Preferably, in step (4), the swin-unet network model is trained in 6052 sets of data, the BatchSize is set to 32, a total of 300 rounds of training are performed, the optimizer is Adam, and the learning rate is 1e-4.

[0041] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes the steps of the above-mentioned embodiment method, and the storage medium can be: ROM / RAM, disk, CD, memory card, etc. Therefore, corresponding to the method of the present invention, the present invention also includes a device for removing return errors of aspheric surface measurement based on a graph-to-graph neural network, which is usually represented in the form of functional modules corresponding to the steps of the method. The device includes:

[0042] Build a virtual interference system module, which is configured to build a virtual interference system in the optical tracing software according to the interference optical path used in the actual aspheric surface error detection work; create a picture-to-picture return error elimination data set module, which is configured to apply random surface error distribution to the measured object in the optical tracing software, calculate the corresponding interference wavefront map, save the two-dimensional information calculated by the simulation software, repeat the simulation in large quantities, and create a picture-to-picture return error elimination data set;

[0043] Build a graph-to-graph neural network model module, which is configured to build a graph-to-graph neural network model, with the interference wavefront image as the network data and the measured aspheric surface shape distribution as the network output, so that the network input dimension is the same as the dimension of the interference wavefront image, and the network output dimension is the same as the dimension of the surface shape error distribution;

[0044] A training graph-to-graph neural network model module is configured to train a graph-to-graph neural network on a graph-to-graph return error elimination dataset to obtain a graph-to-graph return error elimination network model;

[0045] The module for solving the actual aspheric surface shape error is configured to put the interference wavefront image obtained in the actual detection work into the trained image-to-image return error elimination network model to obtain the surface shape distribution result of the actual measured aspheric surface after eliminating the return error.

[0046] A specific implementation example of the present invention is described in detail below. A method for removing the return error of non-zero interference aspheric surface measurement based on a graph-to-graph neural network is implemented in the following manner:

[0047] The process of removing the return error of non-zero interference aspheric surface measurement based on graph-to-graph neural network is established as follows: Figure 1 As shown, the specific implementation steps are:

[0048] Step 1: Build a virtual interference system

[0049] In the example, Zemax software is used to build a Fizeau interferometer that is consistent with the actual optical path.

[0050] Step 2: Create a return error elimination data set

[0051] In this example, 78 Zernike coefficients are randomly generated, and the surface shape PV generated by the coefficients is guaranteed to be between 2λ and 0.05λ. In Zemax, the randomly generated Zernike polynomial coefficients are brought into the measured aspheric surface in the form of a Zernike sag surface, and the surface shape generated by this group of coefficients is used as the true value of the surface shape error. Figure 3 Run the optical path tracing and record the interference wavefront diagram, as shown in Figure 2 As shown in the figure, the true value of the aspheric surface error and the interference wavefront diagram are taken as a set of data with a resolution of 224*224. Repeating the experiment obtains 6052 sets of data, which together constitute the return error elimination data set.

[0052] Step 3: Build a graph-to-graph neural network model

[0053] In this example, a swin-unet network model is built, and the input and output dimensions are both 224*224.

[0054] Step 4: Train the graph-to-graph neural network model

[0055] In this example, the swin-unet network model is trained on 6052 sets of data, the batch size is set to 32, a total of 300 rounds of training, the optimizer is Adam, and the learning rate is 1e-4.

[0056] Step 5: Solving the actual aspheric surface error

[0057] In this example, the data in the validation set is put into the network model, and the surface error distribution is as follows: Figure 4 As shown, the error between it and the true value is Figure 5 As shown in Figure 2, the return error elimination of the aspheric surface error distribution is completed.

[0058] The beneficial effects of the present invention are as follows:

[0059] 1. The present invention uses a graph-generated neural network to eliminate the return error of aspheric surface error detection. Without Zernike fitting, the aspheric surface error distribution and the interference wavefront map are directly matched, effectively retaining the high-frequency information therein.

[0060] 2. The image-generated image network used in the present invention can eliminate the return error in a set of interference wavefront data of an actual optical system after training, without any prior knowledge and preprocessing, and realize the rapid measurement of aspheric surface error.

[0061] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the protection scope of the technical solution of the present invention.

Claims

1. A method for removing return errors in aspheric surface measurement based on a graph-to-graph neural network, characterized in that: It includes the following steps: (1) Build a virtual interferometer system in the optical tracking software based on the interferometer optical path used in the actual aspheric surface error detection work; (2) Produce a data set for image-to-image return error elimination. Apply random surface error distribution to the DUT in the optical tracing software, calculate the corresponding interference wavefront map, save the two-dimensional information calculated by the simulation software, repeat the simulation a large number of times, and produce a data set for image-to-image return error elimination. (3) Building a graph-to-graph neural network model, building a graph-to-graph neural network model, using the interference wavefront image as the network data and the measured aspheric surface shape distribution as the network output, so that the network input dimension is the same as the dimension of the interference wavefront image, and the network output dimension is the same as the dimension of the surface shape error distribution; (4) training a graph-to-graph neural network model, training the graph-to-graph neural network on a graph-to-graph return error elimination dataset to obtain a graph-to-graph return error elimination network model; (5) Solve the actual aspheric surface shape error, put the interference wavefront image obtained in the actual detection work into the trained image-to-image return error elimination network model, and obtain the surface shape distribution result of the actual aspheric surface after eliminating the return error.

2. The method for removing return errors in aspheric surface measurement based on graph-to-graph neural network according to claim 1 is characterized in that: In the step (1), the virtual interferometer constructed by the optical tracing software is consistent with the actual detection optical path.

3. The method for removing return errors in aspheric surface measurement based on graph-to-graph neural network according to claim 2 is characterized in that: In the step (2), random surface errors are added to the aspheric surface to be measured in the virtual interferometer in the form of Zernike surfaces, and the corresponding interference wavefront image is obtained after ray tracing. The resolution of the interference wavefront image and the aspheric surface error to be measured is above 224×224.

4. The method for removing return errors in aspheric surface measurement based on graph-to-graph neural network according to claim 2, characterized in that: In the step (3), the graph-to-graph neural network model is a U-Net image generation network.

5. The method for removing return errors in aspheric surface measurement based on graph-to-graph neural network according to claim 3 or 4, characterized in that: In step (4), the training of the graph-to-graph neural network is to pay attention to the changes in the loss function and accuracy at all times to prevent the network model from being under-trained or over-fitting.

6. The method for removing return errors in aspheric surface measurement based on graph-to-graph neural network according to claim 5, characterized in that: In the step (5), the interference wavefront actually detected is scaled to the same size as the data set so as to be put into the network to correctly solve the surface error.

7. The method for removing return errors in aspheric surface measurement based on graph-to-graph neural network according to claim 1, characterized in that: In the step (1), Zemax software is used to build a Fizeau interferometer that is consistent with the actual optical path.

8. The method for removing return errors in aspheric surface measurement based on graph-to-graph neural network according to claim 7 is characterized in that: In the step (2), 78 Zernike coefficients are randomly generated, and the surface shape PV generated by the coefficients is between 2λ and 0.05λ; the randomly generated Zernike polynomial coefficients are introduced into the aspheric surface to be measured in the form of a Zernike sag surface in the Zemax software, and the surface shape generated by this group of coefficients is used as the true value of the surface shape error for standby; the optical path tracing is performed to record the interference wavefront image; the true value of the aspheric surface shape error and the interference wavefront image are used as a group of data with a resolution of 224×224, and 6052 groups of data are obtained by repeating the experiment, which together constitute a return error elimination data set; in the step (3), a swin-unet network model is built, and the input and output dimensions are both 224×224.

9. The method for removing return errors in aspheric surface measurement based on graph-to-graph neural network according to claim 8, characterized in that: In the step (4), the swin-unet network model is trained on 6052 sets of data, the batch size is set to 32, a total of 300 rounds of training are performed, the optimizer is Adam, and the learning rate is 1e-4.

10. A device for removing return errors in aspheric surface measurement based on a graph-to-graph neural network, characterized in that: It includes: Build a virtual interference system module, which is configured to build a virtual interference system in the optical tracing software according to the interference optical path used in the actual aspheric surface error detection work; create a picture-to-picture return error elimination data set module, which is configured to apply random surface error distribution to the measured object in the optical tracing software, calculate the corresponding interference wavefront map, save the two-dimensional information calculated by the simulation software, repeat the simulation in large quantities, and create a picture-to-picture return error elimination data set; Build a graph-to-graph neural network model module, which is configured to build a graph-to-graph neural network model, with the interference wavefront image as the network data and the measured aspheric surface shape distribution as the network output, so that the network input dimension is the same as the dimension of the interference wavefront image, and the network output dimension is the same as the dimension of the surface shape error distribution; A training graph-to-graph neural network model module is configured to train a graph-to-graph neural network on a graph-to-graph return error elimination dataset to obtain a graph-to-graph return error elimination network model; The module for solving the actual aspheric surface shape error is configured to put the interference wavefront image obtained in the actual detection work into the trained image-to-image return error elimination network model to obtain the surface shape distribution result of the actual measured aspheric surface after eliminating the return error.

Images