Method, device and mechanical apparatus for rapid assembly of optical lenses

Abstract

The application relates to the technical field of optical imaging, in particular to a method and device for rapidly assembling optical lenses and mechanical equipment, wherein the method comprises the following steps: acquiring a wave aberration image of a target optical system in a first target field of view; extracting a fingerprint image from the wave aberration image of the first target field of view, wherein the fingerprint image is an image obtained by superimposing a plurality of images at different intensities and rotation angles, and the fingerprint image provides information related to the misalignment amount in the wave aberration; determining the misalignment amount of the optical lens in the target optical system according to the fingerprint image; and assembling the optical lens of the target optical system based on the misalignment amount. Therefore, the problems of complex manual operation, limited system processing capacity, difficult acquisition of multi-field-of-view aberration information and application limitations of new technologies in the manufacturing and assembling process of high-performance optical systems are solved.

Description

Technical Field

[0001] This application relates to the field of optical imaging technology, and in particular to a method, apparatus and mechanical equipment for rapid assembly of optical lenses. Background Technology

[0002] With the rapid development of technology, the demand for high-performance optical systems has increased dramatically. Accurate and stable measurement results are crucial in fields such as astronomy, life sciences, and artificial intelligence. In the manufacturing process of high-performance optical systems, assembly and adjustment technology plays a key role. By precisely adjusting the position and angle of optical components, it effectively reduces errors generated during manufacturing, thereby making the system's imaging quality closer to the design standards.

[0003] However, with the ever-increasing demands for image quality, alignment techniques face growing challenges. Traditional alignment methods rely primarily on manual operation, which is not only time-consuming and labor-intensive but also makes it difficult to accurately predict and control the impact of different degrees of freedom on aberrations. Although computer-aided techniques such as sensitivity matrices provide some guidance for alignment, their effectiveness is often limited when dealing with large misalignments or complex systems, requiring iterative alignment processes to complete. Summary of the Invention

[0004] This application provides a method, apparatus, and mechanical equipment for rapid assembly of optical lenses to solve problems such as complex manual operation, limited system processing capacity, difficulty in obtaining multi-field aberration information, and limitations in the application of new technologies in the manufacturing and assembly process of high-performance optical systems.

[0005] The first aspect of this application provides a method for rapid assembly of optical lenses, comprising the following steps: acquiring a wavefront aberration image of a target optical system in a first target field of view; extracting a fingerprint image from the wavefront aberration image of the first target field of view, wherein the fingerprint image is an image obtained by superimposing multiple images with different intensities and rotation angles, providing information related to the misalignment in the wavefront aberration; determining the misalignment of the optical lenses in the target optical system based on the fingerprint image; and assembling the optical lenses of the target optical system based on the misalignment.

[0006] Optionally, the first target field of view includes a single target field of view or a multiple target field of view, and the wave aberration image is at least one of a single image, a complete image, and a cropped image.

[0007] Optionally, the expression of the fingerprint image is:

[0008]

[0009] Where, α j Z is the weight corresponding to the j-th Zernike coefficient. jZ is the value of the j-th Zenig coefficient of the mismatched system. j0 It is the value of the j-th Zernike coefficient of the nominal system, I j It is the normalized image of the j-th Zernike polynomial over the circular domain.

[0010] Optionally, determining the misalignment of the target optical system based on the fingerprint image and assembling the optical lenses of the target optical system based on the misalignment includes: inputting the fingerprint image into a primary misalignment model, the primary misalignment model outputting a primary misalignment of the target optical system, and performing a first misalignment assembly of the optical lenses of the target optical system based on the primary misalignment; and / or, inputting the fingerprint image of the optical lenses of the target optical system after the first misalignment assembly into a secondary misalignment model, the secondary misalignment model outputting a secondary misalignment of the target optical system, and performing a second misalignment assembly of the optical lenses of the target optical system based on the secondary misalignment.

[0011] Optionally, before inputting the fingerprint image into a primary misalignment model, the method includes: selecting an image of any second target field of view from the fingerprint image.

[0012] Optionally, before inputting the fingerprint image into the primary misalignment model, the method further includes: generating multiple misalignment systems based on the designed nominal system; calculating the wavefront aberration of each misalignment system under a preset field of view, and generating a fingerprint image of each misalignment system based on the wavefront aberration of each misalignment system under the preset field of view; generating a training dataset for the first misalignment based on the fingerprint image of each misalignment system and the misalignment amount of the optical lens; and training the neural network using the training dataset for the first misalignment until the training stopping condition is met to obtain the primary misalignment model.

[0013] Optionally, before inputting the fingerprint image of the optical lens of the target optical system after a first misalignment assembly into the secondary misalignment model, the method further includes: inputting the training dataset of the first misalignment assembly into the primary misalignment model, wherein the primary misalignment model outputs the primary misalignment values ​​of multiple misalignment systems, performing a first misalignment assembly on the multiple misalignment systems based on the primary misalignment values; generating a training dataset for the second misalignment assembly based on the fingerprint image and misalignment values ​​of the multiple misalignment systems after the first misalignment assembly, and training the primary misalignment model using the training dataset until the training stopping condition is met to obtain the secondary misalignment model.

[0014] Optionally, the training stopping condition includes at least one of the following: the training duration reaches the target duration and the training loss value is less than the target value. The training loss value is calculated using a network loss function, which is:

[0015]

[0016] Where m is the number of degrees of freedom in the assembly and adjustment, and x i x is the actual misalignment in the i-th degree of freedom. i ′ is the network prediction misalignment of the i-th degree of freedom.

[0017] A second aspect of this application provides a rapid assembly apparatus for optical lenses, comprising: an acquisition module for acquiring a wavefront aberration image of a target optical system in a first target field of view; an extraction module for extracting a fingerprint image from the wavefront aberration image of the first target field of view, wherein the fingerprint image is an image obtained by superimposing multiple images at different intensities and rotation angles, providing information related to the misalignment amount in the wavefront aberration; and a processing module for determining the misalignment amount of the optical lenses in the target optical system based on the fingerprint image, and assembling the optical lenses of the target optical system based on the misalignment amount.

[0018] A third aspect of this application provides a mechanical device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the rapid assembly method for optical lenses as described in any of the preceding claims.

[0019] Therefore, this application has at least the following beneficial effects:

[0020] This application's embodiments utilize the one-to-one correspondence between fingerprint images and the misalignment state of an optical system. Compared to traditional wavefront aberration Zernike coefficients, fingerprint images more richly reveal the misalignment information of the optical system. By relying solely on fingerprint images from a single field of view, the misalignment of the optical system can be accurately calculated, greatly simplifying the calculation process. When constructing the training dataset, a large number of misaligned systems are generated based on the nominal system using the Monte Carlo method, and their wavefront aberrations under specific fields of view are extracted to generate fingerprint images. These fingerprint images, along with the corresponding component misalignments, constitute the training dataset, which is then efficiently trained using a ResNet network. The misalignment calculated by the network is accurately compensated into the misaligned system, resulting in a significant improvement in the imaging quality of the assembled system, achieving a high degree of consistency with the nominal system and approaching the diffraction limit. This solves the technical problems encountered in the manufacturing and assembly of high-performance optical systems, such as complex manual operations, limited system processing capabilities, difficulties in acquiring multi-field aberration information, and limitations in the application of new technologies.

[0021] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0022] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0023] Figure 1 This is a flowchart of a rapid assembly method for optical lenses provided according to an embodiment of this application;

[0024] Figure 2 This is a schematic diagram of an RC system with two lenses according to an embodiment of this application;

[0025] Figure 3 This is a schematic diagram of fingerprint images of a set of misaligned systems in the (0°, 0°) field of view of an RC system provided according to an embodiment of this application;

[0026] Figure 4 This is a schematic diagram of fingerprint images of a set of misaligned systems in the field of view (0.67°, 0.67°) of the RC system provided according to an embodiment of this application;

[0027] Figure 5 This is a schematic diagram of the normalized image over the circular domain corresponding to the Zernike polynomials Z5-Z9 provided in the embodiments of this application;

[0028] Figure 6 This is a schematic diagram of an image on a circular region after Z5 and Z6 are superimposed at different ratios according to the embodiments of this application;

[0029] Figure 7 This is a schematic diagram of the off-axis three-reflector system provided according to an embodiment of this application;

[0030] Figure 8 This is a schematic diagram of the imaging quality of an off-axis three-mirror system provided according to an embodiment of this application;

[0031] Figure 9 This is a schematic diagram of fingerprint images of a series of misaligned systems corresponding to a (0°, 0°) field of view of an off-axis three-dimensional mirror system provided according to an embodiment of this application.

[0032] Figure 10 This is a statistical effect diagram of the imaging quality of the optical system provided according to the embodiments of this application;

[0033] Figure 11 This is a schematic diagram illustrating the assembly and adjustment effect according to an embodiment of this application;

[0034] Figure 12 This is a flowchart of a rapid assembly method for optical lenses according to an embodiment of this application;

[0035] Figure 13 This is a diagram showing the effect of the optical system's imaging quality before and after the second misalignment correction according to an embodiment of this application;

[0036] Figure 14This is a zero-field-of-view wavefront aberration photogram of an off-axis three-mirror system with different degrees of freedom misalignment according to an embodiment of this application;

[0037] Figure 15 This is a schematic diagram of the RC system provided according to an embodiment of this application;

[0038] Figure 16 This is a schematic diagram of the imaging quality of the RC deformation system provided according to an embodiment of this application;

[0039] Figure 17 This is a schematic diagram of the structure of a two-mirror system provided according to an embodiment of this application;

[0040] Figure 18 This is a schematic diagram of the imaging quality of the two-mirror system provided according to an embodiment of this application;

[0041] Figure 19 This is a schematic diagram of a fingerprint image of the misalignment system corresponding to a field of view of (0°, 0.6°) according to an embodiment of this application.

[0042] Figure 20 This is a diagram illustrating the difference in imaging quality of the optical system before and after fingerprint image misalignment correction according to an embodiment of this application.

[0043] Figure 21 Waveform pupil diagram of a (0°, 0.6°) field of view provided according to an embodiment of this application;

[0044] Figure 22 This is a schematic diagram illustrating the wavefront aberrations of the same set of input vectors under different basis functions according to embodiments of this application;

[0045] Figure 23 This is a schematic diagram of the RC dual-mirror system structure provided according to an embodiment of this application;

[0046] Figure 24 This is a schematic diagram of fingerprint images of a set of misaligned systems in the field of view (0°, 0.6°) of an RC system provided according to an embodiment of this application;

[0047] Figure 25 This is a schematic diagram of fingerprint images of a set of misaligned systems in the (0°, 0°) field of view of an RC system provided according to an embodiment of this application;

[0048] Figure 26 Example diagram of an optical lens rapid assembly apparatus provided according to embodiments of this application;

[0049] Figure 27 This is a structural schematic diagram of a vehicle provided according to an embodiment of this application. Detailed Implementation

[0050] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0051] With the rapid development of science and technology, high-performance optical systems have been widely used in astronomy, life sciences, artificial intelligence, and other fields, providing more accurate and stable measurement results for scientific research. The imaging quality of these systems directly determines their performance and application value, and assembly and adjustment technology is a key link in ensuring the imaging quality of the system. Therefore, the assembly and adjustment technology of high-performance optical systems has always been a hot topic and a challenge in scientific research and engineering.

[0052] The manufacturing process of high-performance optical systems mainly includes three stages: processing, inspection, and assembly. In the processing and inspection stages, with continuous technological advancements, the processing precision and inspection sensitivity of optical components have reached the nanometer level. However, due to unavoidable errors during processing and assembly, there is often a certain gap between the imaging quality of the optical system before assembly and the nominal design value. Therefore, the role of the assembly stage is to minimize the impact of these errors by precisely adjusting the position and angle of the optical components, so that the system's imaging quality approaches the design value.

[0053] Traditional optical system setup methods primarily rely on manual operation and computer-aided techniques such as sensitivity matrices. However, these methods face numerous challenges in practical applications. First, each element in an optical system has multiple degrees of freedom for setup, and the effects of these degrees of freedom on aberrations are interdependent, making the setup process complex and difficult to predict. Second, when the misalignment of optical elements is large, the sensitivity matrix may not accurately represent the influence of the element's degrees of freedom on aberrations, leading to iterative setup processes. Furthermore, traditional setup methods typically require manual adjustment of the interferometer to obtain aberration information from multiple fields of view, which not only increases the operational difficulty but also prolongs the setup cycle.

[0054] To overcome these challenges, deep learning-based assembly techniques have gained increasing attention in recent years. Deep learning algorithms, trained on large amounts of data, can automatically learn and extract the complex relationship between wavefront aberrations and component positions in an optical system. By inputting a wavefront aberration image of the optical system, the deep learning model can predict the required component position adjustments, thus guiding the assembly process. This method eliminates the need for iterations, significantly reducing the number of adjustments required for optical components and improving assembly efficiency.

[0055] However, despite the achievements of deep learning in optical system assembly and adjustment, several problems and challenges remain. First, for complex optical systems, aberration information from multiple fields of view is still required during assembly and adjustment. This means that in practical applications, a certain degree of manual intervention is still needed to acquire this data. Second, training deep learning models requires a large amount of labeled data, and acquiring and labeling this data is often a tedious and time-consuming task. Furthermore, the generalization ability of deep learning models is also a concern. In practical applications, the structure and parameters of optical systems may change, which could cause the trained model to be unable to adapt to new situations.

[0056] Therefore, the ideal optical system assembly and adjustment technology has the following development trends:

[0057] (1) The assembly and adjustment error becomes smaller and smaller, even negligible. This ensures that the optical system has good imaging quality after assembly and adjustment. For high-performance optical systems, the wavefront aberration of the system after assembly and adjustment should be extremely close to the nominal value. On this basis, the distortion value of the system can even be reduced as much as possible through assembly and adjustment. For systems with extremely high imaging quality requirements, all wavefront aberration Zernike coefficients need to be highly consistent with the design system.

[0058] (2) High assembly and adjustment efficiency. It reduces the number of iterations required to calculate misalignment through aberrations, and can even accurately calculate the current system misalignment without iterations. This reduces the number of adjustments to optical components and aberration measurements. Ideally, optical components can be assembled and adjusted to their accurate positions in a single operation.

[0059] (3) Simple setup and adjustment. By simply acquiring a single-field wavefront aberration image of the optical system, the wavefront aberrations of different fields of view can be adjusted to levels close to the nominal values. It is even possible to correct system distortion using wavefront aberration images.

[0060] (4) Minimize manual intervention during assembly and adjustment to achieve automated assembly and adjustment. Reducing the number of times system aberrations are measured manually and component positions are adjusted creates the possibility for automated assembly and adjustment. This is beneficial for the mass production of high-performance optical systems with high efficiency.

[0061] The following description, with reference to the accompanying drawings, outlines a method, apparatus, and mechanical device for rapid assembly of optical lenses according to embodiments of this application. Addressing the problem mentioned in the background art that the wavefront aberration of the assembled system cannot reach a level very close to the nominal value, this application provides a rapid assembly method for optical lenses. In this method, the one-to-one correspondence between fingerprint images and the misalignment state of the optical system is utilized. Compared to the traditional wavefront aberration Zernike coefficient, fingerprint images more richly reveal the misalignment information of the optical system. By relying solely on fingerprint images from a single field of view, the misalignment of the optical system can be accurately calculated, greatly simplifying the calculation process. When constructing the training dataset, a large number of misaligned systems are generated based on the nominal system using the Monte Carlo method, and the wavefront aberration under a specific field of view is extracted to generate fingerprint images. These fingerprint images, along with the corresponding component misalignment values, constitute the training dataset, which is efficiently trained using a ResNet network. The misalignment values ​​calculated by the network are accurately compensated to the misaligned system, resulting in a significant improvement in the imaging quality of the assembled system, achieving a high degree of consistency with the nominal system and approaching the diffraction limit. This solves the problems faced in the manufacturing and assembly of high-performance optical systems, such as complex manual operations, limited system processing capabilities, difficulty in acquiring multi-field aberration information, and limitations in the application of new technologies.

[0062] Specifically, Figure 1 This is a schematic flowchart of a rapid assembly method for optical lenses provided in an embodiment of this application.

[0063] like Figure 1 As shown, this rapid assembly method for optical lenses is applied to mechanical equipment, and the rapid assembly method for optical lenses includes the following steps:

[0064] In step S101, the wavefront aberration image of the target optical system in the first target field of view is acquired.

[0065] The first target field of view may include a single target field of view or a multiple target field of view. The wavefront aberration image can describe the degree of distortion of the wavefront shape after the light passes through the optical system and directly reflects the imaging quality of the optical system. The wavefront aberration image may be at least one of a single image, a complete image, and a cropped image.

[0066] It is understood that the embodiments of this application provide an important data foundation for subsequent performance evaluation and assembly work by acquiring the wavefront aberration image of the target optical system in the first target field of view.

[0067] It should be noted that a single image can be an independent, complete image; a complete image can be an image without any missing or damaged parts, containing all the necessary information; a cropped image can be a portion of an image extracted from another image source.

[0068] In step S102, a fingerprint image is extracted from the wavefront aberration image of the first target field of view.

[0069] Among them, the fingerprint image can be an image obtained by superimposing multiple images with different intensities and rotation angles, providing information related to the amount of misalignment in wavelet aberration.

[0070] It is understood that the embodiments of this application simplify the assembly and adjustment process by extracting fingerprint images from the wave aberration images of the first target field of view. By accurately analyzing and processing the fingerprint images, the misalignment of each component in the optical system can be accurately determined, avoiding the repeated iterations and trial-and-error process in traditional assembly and adjustment methods, reducing the number of iterations and manual operations, and improving the accuracy and efficiency of assembly and adjustment.

[0071] It should be noted that the fingerprint image focuses only on the difference in waveform aberrations between the misaligned system and the nominal system. Although the nominal system may have larger waveform aberration coefficients, the changes in these coefficients during misalignment may be relatively small. By subtracting the nominal system coefficients from the fingerprint image, subtle changes before and after system misalignment are revealed more clearly, thus helping the network to more accurately calculate the component misalignment. This processing method not only improves the accuracy of the analysis but also provides a more reliable basis for subsequent assembly and adjustment.

[0072] In this embodiment of the application, the expression for the fingerprint image is:

[0073]

[0074] Where, α j Z is the weight corresponding to the j-th Zernike coefficient. j Z is the value of the j-th Zenig coefficient of the mismatched system. j0 It is the value of the j-th Zernike coefficient of the nominal system, I j It is the normalized image of the j-th Zernike polynomial over the circular domain.

[0075] It's important to note that a fingerprint image can contain misalignment information for every degree of freedom (DOF) of every element. A given Zernike coefficient may be sensitive to changes in some DDFs but less sensitive to changes in others. In other words, adding this Zernike coefficient to the fingerprint image allows it to contain misalignment information for some DDFs. By taking the union of all the sensitive coefficients, we can ensure that the fingerprint image contains misalignment information for all DDFs. This further guarantees that the network can accurately calculate the misalignment for each DDF from the fingerprint image.

[0076] Specifically, for each degree of freedom of adjustment, the three Zernike coefficients that are most sensitive to change are first selected. The adjustment sensitivity of an optical system refers to the degree of change in wavelet aberrations of the system when a certain degree of freedom of an element changes slightly relative to the nominal system. When the misalignment of the optical element is small, each Zernike wavelet aberration coefficient of the misaligned system is approximately linearly related to the misalignment. Here, ΔX represents the misalignment of each degree of freedom of the optical system, and ΔZ represents the difference between the Zernike coefficients of the nominal system and the misaligned system under the misalignment state of ΔX. In this case, the sensitivity matrix A can approximately represent the sensitivity of each Zernike aberration to different degrees of freedom.

[0077]

[0078] After generating the sensitivity matrix A of the optical system, for each degree of freedom, select the three most sensitive coefficients. Specifically, for the j-th column of A... Select the three elements with the largest absolute values. s j1 s j2 s j3 The corresponding coefficients are the three wave aberration coefficients that are most sensitive to the j-th degree of freedom. The set of these three coefficients is called S. j , Therefore, by taking the union of the sets of coefficients that are more sensitive for each degree of freedom, we obtain the set of coefficients S that are more sensitive for the current system.

[0079] Coefficients not in S tend to exhibit smaller changes before and after optical system misalignment, typically two to three orders of magnitude smaller than the changes in more sensitive coefficients. Adding coefficients not in S to the fingerprint image results in minimal changes. Furthermore, these minute changes are easily masked by measurement noise. In contrast, a fingerprint image composed of coefficients in S contains sufficient information about optical system misalignment; therefore, coefficients in S are chosen for generating the fingerprint image.

[0080] In step S103, the misalignment of the optical lenses in the target optical system is determined based on the fingerprint image, and the optical lenses of the target optical system are assembled based on the misalignment.

[0081] Among them, the misalignment can be the deviation between the actual position or parameters of optical elements in an optical system and the ideal state.

[0082] It is understood that by utilizing the difference information between the misalignment amount contained in the fingerprint image and the nominal system wave aberration, the misalignment of each component in the optical system can be identified more accurately. By assembling the optical lens based on the misalignment amount, the number of trial and error and the iteration process in the assembly and adjustment process can be greatly reduced, which not only improves the performance and quality of the optical system, but also reduces the assembly and adjustment cost and time.

[0083] In this embodiment, determining the misalignment of the target optical system based on a fingerprint image and assembling the optical lenses of the target optical system based on the misalignment includes: inputting the fingerprint image into a primary misalignment model, the primary misalignment model outputting a primary misalignment of the target optical system, and performing a first misalignment assembly of the optical lenses of the target optical system based on the primary misalignment; and / or, inputting the fingerprint image of the optical lenses of the target optical system after the first misalignment assembly into a secondary misalignment model, the secondary misalignment model outputting a secondary misalignment of the target optical system, and performing a second misalignment assembly of the optical lenses of the target optical system based on the secondary misalignment.

[0084] The primary misalignment model can be used to analyze and calculate the misalignment of an optical system during the initial assembly process, while the secondary misalignment model can be used to analyze and calculate the misalignment of an optical system after a primary misalignment assembly. The primary misalignment can be the misalignment of each component in the optical system calculated by the primary misalignment model, while the secondary misalignment can be the remaining misalignment of the optical system after a primary misalignment assembly calculated by the secondary misalignment model.

[0085] It is understood that the embodiments of this application use the first offset amount output by the first offset model for the first offset assembly, which can accurately adjust the initial offset of each component in the optical system. By inputting the fingerprint image after the first offset assembly into the second offset model, a more accurate second offset amount can be obtained. The second offset assembly based on the second offset amount can further correct the position or state of each component in the optical system, making it closer to the ideal value, thus achieving efficient and accurate assembly and adjustment of the optical system.

[0086] Specifically, when the system is out of balance, the fingerprint images corresponding to different fields of view exhibit significantly different pattern complexities. For example... Figure 2 As shown, a 1m focal length, F5 RC system was given nine random misalignment values, and fingerprint images corresponding to the (0°, 0°) and (0.67°, 0.67°) fields of view were observed. Figure 3 and Figure 4As shown, the fingerprint images corresponding to the (0°, 0°) field of view are relatively similar, mainly exhibiting rotation and flipping of coma images. However, the fingerprint images corresponding to the (0.67°, 0.67°) field of view have more complex patterns, with twisted and deformed textures, resulting in lower similarity. Fingerprint images with richer and more complex patterns are more conducive to accurately extracting misalignment information.

[0087] The complexity of a fingerprint image is mainly related to the system's first-order aberrations, specifically the 5th to 9th terms of the Zernike coefficients, such as... Figure 5 As shown, the fingerprint image does not contain the first four terms of Zernike decomposition. Furthermore, the changes in higher-order aberrations after the ninth Zernike term before and after system misalignment are typically less than one-tenth of the changes in the first-order aberrations. In other words, higher-order aberrations only alter the details of the fingerprint image, contributing little to the overall pattern.

[0088] In this embodiment of the application, before inputting the fingerprint image into the first misalignment model, the method includes: selecting an image of any second target field of view from the fingerprint image.

[0089] The second target field of view can include multiple sub-fields of view, which can be the optimal field of view, the suboptimal field of view, and the ordinary field of view, etc.

[0090] It is understood that the embodiments of this application select the second target field of view from the fingerprint image, taking into account the impact of different fields of view on fingerprint image processing and recognition performance. When the selected field of view is the optimal field of view, the image complexity of this region is high, containing rich fingerprint feature information, thus the accuracy of the calculated element position will be higher, thereby improving the accuracy and reliability of fingerprint recognition. However, in practical applications, fingerprint images may be affected by various factors, causing the optimal field of view to not always be accurately selected. In this case, a suboptimal or ordinary field of view can be selected for recognition. Although the image complexity of the suboptimal and ordinary fields of view is lower than that of the optimal field of view, and the feature information is relatively less, they still contain certain fingerprint features, which can be used to improve the accuracy of the offset. This realizes the selection of a suboptimal or ordinary field of view as an alternative when the optimal field of view cannot be selected, which not only ensures high-precision recognition under optimal conditions, but also improves recognition performance under unfavorable conditions, thereby enriching the overall performance.

[0091] When the two astigmatisms corresponding to Zernike polynomials Z5 and Z6 are superimposed with arbitrary sizes, their corresponding image in the circular domain is the normalized image corresponding to Z5 multiplied by a coefficient and then rotated by a certain angle, such as... Figure 6 As shown in Table 1 below, assuming the coefficients of polynomials Z5 and Z6 are z5 and z6 respectively, the expression for the corresponding circular domain is:

[0092]

[0093] in, The image at this point is the normalized image corresponding to Z5 multiplied by Rotate counterclockwise The image after that. Similarly, it can be proven that the same conclusion applies to the superposition of two coma aberrations.

[0094] Table 1. Zernike aberration expressions for items 5-9 and their corresponding aberrations.

[0095]

[0096] Therefore, for a specific field of view, the ratio between the corresponding first-order astigmatism, coma, and spherical aberration determines the complexity of the fingerprint image. Among them, the magnitude of astigmatism is... The magnitude of coma is In typical optical systems, spherical aberration accounts for a significantly smaller proportion than astigmatism and coma. Therefore, given a series of misalignment states, the complexity of a fingerprint image can be approximated by I0. a and I c The statistical results of the ratio are used to measure it. When I a and I c When the ratios are almost all much less than 1, the proportion of coma in the image is relatively large. Therefore, fingerprint images have a high degree of similarity and will present similar characteristics. Figure 3 The pattern in the image. When I a and I c When the ratios are almost all much greater than 1, the complexity of the fingerprint image is similarly low. And when I... a and I c When the ratio is closer to 1, the resulting fingerprint image exhibits a richer pattern. Based on this metric, a more complex field of view can be selected for the fingerprint image, allowing the network to learn more information and consequently, the calculated misalignment will be more accurate.

[0097] In this embodiment, before inputting the fingerprint image into the first misalignment model, the method further includes: generating multiple misalignment systems based on the designed nominal system; calculating the wavefront aberration of each misalignment system under a preset field of view, and generating a fingerprint image of each misalignment system based on the wavefront aberration of each misalignment system under the preset field of view; generating a training dataset for the first misalignment based on the fingerprint image of each misalignment system and the misalignment amount of the optical lens; and training the neural network using the training dataset for the first misalignment until the training stopping condition is met to obtain the first misalignment model.

[0098] The nominal system can be an optical system in an ideal state manufactured according to design specifications and requirements. The misaligned system can be an optical system in which the position or state of the optical element deviates from the nominal state due to manufacturing errors, assembly errors or other factors. Wavelength aberration can be the deviation between the actual light and the ideal light, usually expressed as a multiple of the wavelength. The training dataset for the first assembly can be a dataset containing fingerprint images of multiple misaligned systems and the corresponding optical lens misalignment.

[0099] It is understood that, by generating multiple misalignment systems and calculating their wavefront aberrations, this embodiment of the application can simulate various misalignment situations that may occur during actual manufacturing and assembly. The fingerprint images and corresponding optical lens misalignment amounts together constitute a training dataset for training the neural network model. When the training meets the stopping condition, the resulting misalignment model can predict the lens misalignment amount of a given misalignment system. This method automates and intelligently adjusts optical systems, improves the accuracy and efficiency of adjustment, reduces reliance on manual operation, and provides strong support for the manufacturing and adjustment of optical systems.

[0100] For example, design an off-axis three-lens mirror system with a field of view of (1.757°, 1.318°), a focal length of 500mm, and an aperture of F / 4. The structural diagram is shown below. Figure 7 As shown, the image quality is as follows Figure 8 As shown, its wavefront aberration RMS value is 0.01373λ. This system has a compact structure, wide-spectrum chromatic aberration-free operation, no central obstruction, high transmittance with no absorption, and offers ample design freedom and good cost control, making it widely used.

[0101] A training set of 100,000 data points was created. During the misalignment process, the primary mirror was used as the baseline, and the secondary and tertiary mirrors were adjusted across 10 degrees of freedom (DOF). For each mirror, the eccentricity misalignment range in the x, y, and z directions was ±0.25 mm, and the tilt misalignment along the x and y axes was ±0.025 degrees. Since the components of this off-axis three-mirror system are rotationally symmetric, the tilt misalignment along the z-axis was not considered. A uniform probability density function was used for each DOF when generating the dataset.

[0102] In this embodiment, the misalignment is generated using a uniform distribution for each degree of freedom. After generating a series of misaligned systems based on the probability distribution of each degree of freedom, the wavefront aberration of the (0°, 0°) field of view of each system is selected to generate a fingerprint image. An example of the fingerprint image is shown below. Figure 9 As shown. The network used for training was ResNet34. The total training time was 58 minutes, completed on a single-core GPU NVIDIA GeForce RTX 3090. After training, it only takes 4ms to predict the offset of an out-of-balance system.

[0103] To test the performance of the network in correcting the misalignment system, 10,000 new misalignment systems were generated, while maintaining consistency with the dataset used for training the network. Figure 10 The figure shows the statistical effect of optical system imaging quality before and after correction of misalignment using fingerprint images. When the optical system is misaligned, the imaging quality deteriorates significantly, and the average wavefront aberration RMS value of the misaligned system increases to 0.1618λ. The system after misalignment prediction and reconfiguration using a network exhibits highly consistent imaging quality, with the average wavefront aberration RMS value decreasing to 0.01460λ, representing a 90.97% improvement in average imaging quality compared to the misaligned system. The absolute error between the average wavefront aberration RMS value of the reconfigured system and the nominal system is 0.0009λ.

[0104] like Figure 11 As shown, the adjusted system not only has a wavelet aberration RMS value very close to that of the nominal system, but its wavelet aberration distribution is also similar to that of the nominal system. Off-axis three-mirror system misalignment information can be extracted from fingerprint images through learning, and the misaligned system can be adjusted to a position with imaging quality close to that of the nominal system.

[0105] It should be noted that, as Figure 12 As shown, training the neural network begins with designing a nominal system that meets the requirements in CodeV. After determining the degrees of freedom to be tuned and their corresponding offset ranges, a uniform probability distribution model is used for each degree of freedom to generate a large amount of offset data. Each set of data, ΔX, represents a set of random offset values ​​for all degrees of freedom that need to be tuned. For each set of offset data, the Zernike coefficients corresponding to the current offset system and the change in the nominal system, ΔZ, are recorded and used together with ΔX to form the training label values. Then, the wavefront aberration fingerprint image required for training is encoded based on ΔZ and provided to the network for training along with the label values.

[0106] After network training is complete, a new misaligned system is randomly generated. At this point, the misalignment ΔX0 of the misaligned system is unknown to the network. The current misaligned system's ΔF is obtained and a fingerprint image is generated, which is then input into the network. The network-predicted misalignment ΔX′ is fed back to the system, and the corrected system's misalignment ΔX is then calculated. update Let ΔX0 - ΔX be the image quality of the corrected system, representing the network's corrective capability.

[0107] In this embodiment, before inputting the fingerprint image of the optical lens of the target optical system after a first misalignment assembly into the secondary misalignment model, the method further includes: inputting the training dataset of the first misalignment assembly into the primary misalignment model, the primary misalignment model outputting the primary misalignment values ​​of multiple misalignment systems, performing a first misalignment assembly on the multiple misalignment systems based on the primary misalignment values; generating a training dataset for the second misalignment assembly based on the fingerprint images and misalignment values ​​of the multiple misalignment systems after the first misalignment assembly, and training the primary misalignment model using the training dataset until the training stopping condition is met to obtain the secondary misalignment model.

[0108] The training termination condition may include at least one of the following: the training duration reaches the target duration and the training loss value is less than the target value.

[0109] It is understood that the embodiments of this application utilize a neural network model for a stepwise optimization assembly and adjustment process. First, a primary misalignment model is used to predict the misalignment amount and perform the first misalignment assembly. Then, based on the assembled data, a training dataset for the second assembly and adjustment is generated, and the model is trained to obtain a secondary misalignment model. This significantly improves the assembly and adjustment accuracy, efficiency, and stability of the optical system, providing effective technical support for the manufacturing and assembly and adjustment of the optical system. It is suitable for large-scale manufacturing scenarios and helps to reduce manufacturing costs and improve production efficiency.

[0110] The optical system in this embodiment, after one adjustment, has an imaging quality that is quite close to the nominal system, which can meet the requirements of most adjustment work. If the imaging quality of the adjusted system is required to be very close to that of the nominal system, a second adjustment can be performed. The second adjustment performs transfer learning on the already trained network. To generate the training dataset for the second adjustment, the network trained in the first training is used to adjust each of the 100,000 misaligned systems generated in the first training. Compared with the nominal system, the adjusted system still has a certain amount of misalignment. Therefore, the system after the first adjustment can be regarded as the misaligned system for the second training. The fingerprint image of the system after the first adjustment and the corresponding misalignment are obtained to generate the dataset for the second training, and then the second training is performed.

[0111] In this embodiment of the application, the training loss value is calculated using a network loss function, which is:

[0112]

[0113] Where m is the number of degrees of freedom in the assembly and adjustment, and x i x is the actual misalignment in the i-th degree of freedom. i ′ is the network prediction misalignment of the i-th degree of freedom.

[0114] Specifically, when the misalignment is small, the aberrations of the optical system are approximately linearly related to the misalignment. Therefore, the network loss function is modified by adding a regularization condition based on the setup sensitivity. In this case, network training can be completed in just 18 minutes. After training, it still only takes 4ms to predict the misalignment of an misaligned system. Figure 13 As shown, the statistical effect of the optical system's imaging quality before and after the second correction of the misalignment is presented.

[0115] After the second adjustment, the average wavefront aberration RMS value of the optical system was further reduced to 0.01392λ. The absolute error between the average wavefront aberration RMS value of the adjusted system and the nominal system was 0.0002λ, and the relative error was only 1.4%. The imaging quality of the adjusted system was very close to that of the nominal system.

[0116] Simultaneously, after the second reassembly, each wavefront aberration coefficient of the reassembly system is very close to that of the nominal system. The first to fourth wavefront aberration coefficients can be eliminated by image plane focusing; the wavefront aberrations after the ninth coefficient are much smaller than those of the fifth to ninth coefficients. Therefore, the standard deviation of the fifth to ninth Zernike wavefront aberration coefficients can represent the consistency of the wavefront aberration distribution among the 10,000 optical systems used to test the network reassembly performance. As shown in Table 2 below, after the second reassembly, the standard deviation of the fifth to ninth Zernike wavefront aberration coefficients of the optical system is greatly reduced. This indicates that the wavefront aberration distribution of the system after the second reassembly is highly consistent with that of the nominal system.

[0117] Table 2 Standard deviations of Zernike wavelet aberration coefficients (items 5-9) after assembly and adjustment.

[0118]

[0119]

[0120] It should be noted that using Loss MSE As a loss function, it can generally predict the offset of an imbalanced system relatively accurately and correct the imbalanced system to a position close to the nominal system. However, Loss MSE It directly represents the positional differences of each degree of freedom between the assembled system and the nominal system, but cannot directly represent the differences in their imaging quality. For example, as... Figure 14 As shown, these are wavefront aberration pupil diagrams for the off-axis three-mirror system at different misalignment positions. All three misaligned systems have only a single degree of freedom misalignment, and the misalignment amounts are numerically identical, meaning their loss... MSE They are the same. However, the imaging quality of their corresponding systems differs significantly. Furthermore, Loss MSEIn the calculation process, tilt and eccentricity misalignment are treated as dimensionless values, resulting in the loss of dimensional information. In practical optical systems, when the tilt and eccentricity misalignment of an element are numerically the same, the wave aberration change caused by the tilt misalignment is often much greater than the change caused by the eccentricity misalignment. Therefore, simply using the loss... MSE As a loss function of the network, the imaging quality of the system after adjustment may still be somewhat different from that of the nominal system.

[0121] If the imaging quality of the adjusted system is required to be very close to that of the nominal system, regularization conditions can be set during training based on the adjusted sensitivity to further reduce the difference in imaging quality between the two. Generally, the sensitivities of different degrees of freedom in a system are not the same. When the misalignment of different degrees of freedom is the same, the wavelet aberration of the system is not the same.

[0122] Therefore, by using the assembly sensitivity of each degree of freedom, the difference between the Zernike aberration coefficients of the assembled system and the nominal system can be approximately estimated, thereby constructing the regularization term of the loss function.

[0123] The loss function at this point can be written as Loss * =Loss MSE +βLoss A ,Right now:

[0124] Loss A =MSE(A(△X-△X′))

[0125] △X-△X′ represents the positional deviation of each component in the assembled system from the nominal system after feeding back the network-predicted misalignment to the misaligned system. Since the positional deviation of the assembled system is small, A(△X-△X′) can be approximately represented as the difference between the Zernike aberration coefficients of the assembled system and the nominal system. Using Loss... A The regularization term in the loss function aims to make the network's predicted misalignment as accurate as possible, while also ensuring that the Zernike aberrations of the adjusted system and the nominal system are as close as possible, meaning that their imaging quality is as close as possible.

[0126] Generally speaking, with Loss MSE As a loss function, it allows the imaging quality of the adjusted system to be very close to that of the nominal system. This is because Loss MSEIdeally, the predicted misalignment should closely approximate the actual misalignment, meaning the assembled system's position will typically be very close to the nominal system, resulting in image quality very close to the nominal system. However, the relationship between system image quality and the misalignment of a degree of freedom is not entirely linear. It's possible that a slight shift in the misalignment of a component's degree of freedom away from the nominal system can actually bring the system's image quality closer to the nominal system. A The objective is directly related to the system's imaging quality, and the direct expectation is that the image quality of the assembled system will be closer to that of the nominal system. Therefore, with Loss... A As a loss function, the image quality of the adjusted system is usually better than that of the system with the loss function. MSE The system derived from the loss function is closer to the nominal system, but the prediction accuracy of component offsets may be slightly reduced. During network training, it is often advisable to first use the loss function... MSE As a loss function. When the network's prediction of component offset is relatively accurate, a certain weighted loss is added to the loss function based on experience. A Networks trained in this way can produce systems where the component positions are close to those of the nominal system, and the imaging quality is also typically close to that of the nominal system.

[0127] The rapid assembly method for optical lenses proposed in this application utilizes the one-to-one correspondence between fingerprint images and the misalignment state of the optical system. Compared to the traditional wavefront aberration Zernike coefficient, fingerprint images more richly reveal the misalignment information of the optical system. By relying solely on fingerprint images from a single field of view, the misalignment of the optical system can be accurately calculated, greatly simplifying the calculation process. When constructing the training dataset, a large number of misaligned systems are generated based on the nominal system using the Monte Carlo method, and their wavefront aberrations under specific fields of view are extracted to generate fingerprint images. These fingerprint images, along with the corresponding component misalignments, constitute the training dataset, which is then efficiently trained using a ResNet network. The misalignment calculated by the network is accurately compensated into the misaligned system, resulting in a significant improvement in the imaging quality of the assembled system, achieving a high degree of consistency with the nominal system and approaching the diffraction limit. This solves the problems encountered in the manufacturing and assembly of high-performance optical systems, such as complex manual operations, limited system processing capabilities, difficulties in acquiring multi-field aberration information, and limitations in the application of new technologies.

[0128] The following section will describe the rapid assembly of the optical lenses for the dual-mirror telescope system, including the following steps:

[0129] Double-mirror telescope systems, also known as RC optical systems, offer advantages such as large aperture, long focal length, wide spectral range, easy weight reduction, and strong temperature adaptability. However, RC systems consisting of only two mirrors do not easily achieve a large field of view with excellent image quality. To broaden the field of view and correct off-axis aberrations, a lens group or mirror is usually added after the RC two-mirror system, forming a deformable RC system.

[0130] like Figure 15 As shown, an RC distorting system with a focal length of 1m and a field of view of (0.6°, 0.6°) has an imaging quality as follows: Figure 16 As shown, in an RC distorting system, the adjustment sensitivity of the secondary mirror is typically much higher than that of other components. A tilt misalignment of 0.02° along the x-direction in the secondary mirror increases the system's wavefront aberration RMS value to 0.04λ; while in the lens group, a tilt misalignment of 0.15° along the x-direction in the second lens only leads to an increase in wavefront aberration to 0.04λ. This excessively high sensitivity of the secondary mirror makes it difficult to adjust the RC distorting system to a position where the image quality is close to the nominal value, even after repeated iterations.

[0131] Therefore, when only the primary and secondary mirrors are installed, and no subsequent components are installed, the position of the secondary mirror must be very close to the nominal system to reduce the difficulty of subsequent assembly and adjustment of the RC deformation system. For example... Figure 17 As shown, Figure 15 The lens group in the image is removed, while the surface shape and position of the primary and secondary mirrors remain unchanged, creating a two-mirror system with image quality as follows: Figure 18 As shown. Using this two-mirror system as the nominal system for assembly, the performance of extracting the secondary mirror misalignment from the fingerprint image of the two-mirror system was studied.

[0132] A training set containing 100,000 data points was created. During the misalignment process, the primary mirror was used as the reference, and the five degrees of freedom of the secondary mirror were adjusted. The eccentricity misalignment range of the secondary mirror in the x, y, and z directions was ±0.2 mm, and the tilt misalignment along the x and y axes was ±0.025 degrees. Since the secondary mirror of the RC system is rotationally symmetric, the tilt misalignment along the z-axis does not need to be considered. In this embodiment, the misalignment amount of each degree of freedom is generated in a uniform distribution manner. After generating a series of misalignment systems according to the probability distribution of each degree of freedom, the wavefront aberration of the (0°, 0.6°) field of view of each system is selected to generate a fingerprint image, as shown in the figure. Figure 19 As shown. The total training time for the network was 45 minutes, completed on a single-core GPU NVIDIA GeForce RTX 3090. After training, it only takes 4ms to predict the offset of an out-of-balance system.

[0133] To test the performance of the network in correcting the misalignment system, 10,000 new misalignment systems were generated, while maintaining consistency with the dataset used for training the network. Figure 20 As shown, when the nominal aberration of the dual-mirror system is large, the average wavefront aberration RMS value of the misaligned system further increases to 0.5659λ, an increase of 0.04λ compared to the nominal system's wavefront aberration RMS value. The degradation in image quality will be further amplified after lens installation. The system with misalignment predicted by a network and then adjusted exhibits highly concentrated image quality, with an average wavefront aberration RMS value of 0.52537λ, and a relative error of only -0.005% compared to the nominal system's wavefront aberration RMS value. Furthermore, the misalignment of the secondary mirror in the adjusted dual-mirror system is very small. The statistical values ​​of the secondary mirror misalignment in 10,000 adjusted systems are shown in Table 3 below.

[0134] Table 3 Initial misalignment range, mean and standard deviation of misalignment for each degree of freedom after setup.

[0135]

[0136] like Figure 21 As shown, even if the nominal system has significant aberrations, the wavefront aberration distribution of the system adjusted using this method can be similar to that of the nominal system. The current system misalignment is extracted from the fingerprint image of the RC two-mirror system and then adjusted. After adjustment, the secondary mirror position is very close to that of the nominal system, and the wavefront aberration distribution is also very similar. This greatly facilitates the adjustment of the lens group or mirror behind the secondary mirror.

[0137] Learning from Zernike coefficients of wavelet aberrations allows for a relatively accurate calculation of the misalignment of an off-balanced system, thus guiding assembly. However, for neural networks, it's unknown which aberration each Zernike coefficient corresponds to and what type of misalignment it represents. In other words, using coefficient vectors as training data causes the network to lose information about the basis functions of the Zernike decomposition. Figure 22 As shown, for the same set of coefficient vectors, the wavelet aberrations represented by the vectors differ depending on the corresponding decomposition method. The lack of basis function information in the Zernike decomposition increases the uncertainty in the neural network learning process. Furthermore, for multi-element assembly problems, it is usually necessary to build a large-scale fully connected neural network, and it is difficult to determine the number of layers and the number of neurons in each layer. At the same time, to increase the wavelet aberration images included in the dataset, multi-field coefficient information is often required, which is difficult to achieve in actual assembly processes. Moreover, even when using multi-field wavelet aberration coefficients for learning, the imaging quality of the assembled system usually still cannot approach the nominal value.

[0138] In this embodiment, a single-field fingerprint image is used as training data for a neural network. The fingerprint image extracts information related to the offset from wave aberrations in a specific field of view, characterizing the offset location of the current offset system. The fingerprint image possesses uniqueness, rich patterns, and low noise. Each fingerprint image corresponds one-to-one with an offset system at different offset locations; the fingerprint images differ for systems at different offset locations. The fingerprint image is composed of a series of base images superimposed with different intensities and rotation angles, resulting in a rich pattern. Simultaneously, the fingerprint image removes information in the wave aberrations that is easily affected by noise, retaining only the information most closely related to the offset.

[0139] Based on fingerprint images, the network can accurately calculate the current system offset. On one hand, during assembly and adjustment, experienced engineers can estimate the offset by using images measured with an interferometer. Using fingerprint images as input makes the neural network's learning process more similar to manual offset assessment, meaning the neural network directly extracts offset information from the image. On the other hand, fingerprint images simultaneously contain coefficient information from wavelet aberration decomposition and information from the decomposition basis functions, providing the neural network with richer information.

[0140] Therefore, fingerprint images allow for the rapid and accurate calculation of optical system misalignment, enabling the system to be aligned close to the nominal system. The entire calculation process requires no iteration and is time-efficient. Furthermore, only aberration information from a single field of view is needed to calculate the misalignment. This means that during alignment, it is unnecessary to change the interferometer position to measure wavefront aberrations in different fields of view, significantly reducing the difficulty of alignment. The image quality of the aligned optical system is very close to that of the nominal system; even every wavefront aberration coefficient of the aligned system can approach that of the nominal system. This method reduces manual intervention during alignment, greatly shortening the alignment time.

[0141] Choosing an appropriate field of view is beneficial for improving the network's tuning performance. When the system is misaligned, some fields of view produce fingerprint images with richer patterns, while others produce fingerprint images with simpler patterns. Selecting fingerprint images corresponding to fields of view with richer patterns will further improve training results. Figure 23 As shown, during the training of the two-mirror system, the setup is performed based on the fingerprint image of the (0°, 0.6°) field of view, as follows: Figure 24 As shown, the system performance is significantly higher than the result obtained by adjusting the fingerprint image at (0°, 0°). Figure 25As shown, the fingerprint image of the central field of view of the RC system is mainly composed of a normalized image of coma rotated by a certain angle, while other aberration images contain relatively little information. However, the fingerprint images of the edge fields of view of the RC system have richer patterns. Some fingerprint images resemble astigmatic images after distortion, while others are unlike any aberration image. This is because, after the RC system is decentered, the aberrations in the central field of view are still mainly primary coma, with other types of aberrations being almost an order of magnitude smaller than coma. After RC misalignment, astigmatism in its edge fields of view increases significantly. This means that the edge fields of view of the RC system have multiple aberrations related to the current misalignment, providing richer information for the network to learn and resulting in more accurate misalignment predictions. The above analysis suggests that selecting fields of view with poorer imaging quality for analysis during the setup process may help improve prediction performance. This is because, near the nominal system, there are many solutions with different misalignment locations but similar imaging quality. The fingerprint images corresponding to these solutions in fields of view with better imaging quality are likely to be similar because there are fewer types of aberrations in those fields. In fields of view with poor imaging quality, there are often multiple aberrations. The correspondence between each aberration and the amount of misalignment is usually different, which results in fingerprint images at each misalignment location having richer shapes rather than being too similar. In some cases of fabricated misalignment, selecting a field of view beyond the nominal system's field of view to extract fingerprint images can yield even more accurate results in determining the amount of misalignment.

[0142] In summary, the embodiments of this application include systems with lenses, systems with a large number of components, and systems with special mechanisms. Due to the special nature of these systems, although they can be directly assembled and adjusted, certain techniques can be applied during the assembly and adjustment process to further improve the performance. For RC deformable systems, the secondary mirror has very high sensitivity; at the same time, the lens group has lower sensitivity and corrects some aberrations caused by secondary mirror misalignment. Directly calculating the misalignment of the secondary mirror and the lens group simultaneously makes it difficult to make the image quality of the assembled system very close to that of the nominal system. Therefore, the lens group can be installed first without installing it. Based on the fingerprint images of the primary and secondary mirrors, the secondary mirror can be adjusted to a position with a smaller misalignment. Then, the lens group can be installed, and based on the fingerprint image at this time, the misalignment of the current secondary mirror and the lens group can be calculated and compensated. In this way, not only is the misalignment of the secondary mirror reduced after assembly and adjustment, but the image quality of the system is also closer to that of the nominal system. The entire assembly and adjustment process only requires measuring two wavelet aberration images, which does not increase the workload too much. At the same time, when installing an optical system, the positions of the primary and secondary mirrors are usually determined first, and then the positions of subsequent components are determined. The assembly and adjustment process is consistent with the traditional component installation process.

[0143] Furthermore, in the assembly and adjustment of actual optical systems, there is a certain degree of robustness to component surface shape errors. In a real optical system, each component carries a certain surface shape error, leading to a discrepancy between actual measurements and wavelet aberrations obtained from software simulations. However, even during training, when the dataset is generated from a simulation system containing only ideal surface shapes, the network can still predict the current system's misalignment based on fingerprint images acquired from the actual system with surface shape errors. After one assembly and adjustment, the imaging quality of the actual system often improves significantly. At this point, the assembled system still differs from the nominal system's imaging quality, primarily due to surface shape errors. A better approach is to measure the surface shape error of each component before assembly and adjustment, import it into the software-designed model, and optimize the model. Such a system can be considered the nominal system in actual assembly and adjustment. Training based on this nominal system and predicting the misalignment will result in the assembled actual system's imaging quality being closer to the nominal system.

[0144] Next, the rapid assembly apparatus for optical lenses according to embodiments of this application is described with reference to the accompanying drawings.

[0145] Figure 26 This is a block diagram of an optical lens rapid assembly device according to an embodiment of this application.

[0146] like Figure 26 As shown, the optical lens rapid assembly device 10 includes: an acquisition module 100, an extraction module 200, and a processing module 300.

[0147] The acquisition module 100 is used to acquire the wavefront aberration image of the target optical system in the first target field of view; the extraction module 200 is used to extract the fingerprint image from the wavefront aberration image of the first target field of view, wherein the fingerprint image is an image obtained by superimposing multiple images with different intensities and rotation angles, providing information related to the misalignment in the wavefront aberration; the processing module 300 is used to determine the misalignment of the optical lens in the target optical system based on the fingerprint image, and assemble the optical lens of the target optical system based on the misalignment.

[0148] It should be noted that the foregoing explanation of the embodiment of the rapid assembly method for optical lenses also applies to the rapid assembly device for optical lenses in this embodiment, and will not be repeated here.

[0149] The rapid assembly device for optical lenses proposed in this application utilizes the one-to-one correspondence between fingerprint images and the misalignment state of the optical system. Compared to the traditional wavefront aberration Zernike coefficient, fingerprint images more richly reveal the misalignment information of the optical system. By relying solely on fingerprint images from a single field of view, the misalignment of the optical system can be accurately calculated, greatly simplifying the calculation process. When constructing the training dataset, a large number of misaligned systems are generated based on the nominal system using the Monte Carlo method, and wavefront aberrations under specific fields of view are extracted to generate fingerprint images. These fingerprint images, along with the corresponding component misalignments, constitute the training dataset, which is then efficiently trained using a ResNet network. The misalignment calculated by the network is accurately compensated into the misaligned system, resulting in a significant improvement in the imaging quality of the assembled system, achieving a high degree of consistency with the nominal system and approaching the diffraction limit. This solves the problems encountered in the manufacturing and assembly of high-performance optical systems, such as complex manual operations, limited system processing capabilities, difficulties in acquiring multi-field aberration information, and limitations in the application of new technologies.

[0150] Figure 27 A schematic diagram of the structure of a mechanical device provided in an embodiment of this application. The mechanical device may include:

[0151] The memory 2701, the processor 2702, and the computer program stored on the memory 2701 and capable of running on the processor 2702.

[0152] When the processor 2702 executes the program, it implements the rapid assembly method for optical lenses provided in the above embodiments.

[0153] Furthermore, the mechanical equipment also includes:

[0154] Communication interface 2703 is used for communication between memory 2701 and processor 2702.

[0155] The memory 2701 is used to store computer programs that can run on the processor 2702.

[0156] The memory 2701 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0157] If the memory 2701, processor 2702, and communication interface 2703 are implemented independently, then the communication interface 2703, memory 2701, and processor 2702 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 27 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0158] Optionally, in a specific implementation, if the memory 2701, processor 2702, and communication interface 2703 are integrated on a single chip, then the memory 2701, processor 2702, and communication interface 2703 can communicate with each other through an internal interface.

[0159] The processor 2702 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0160] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0161] Furthermore, the terms "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. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0162] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0163] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0164] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A method for rapid assembly of optical lenses, characterized in that, The method is applied to mechanical equipment and includes the following steps: Acquire the wavefront aberration image of the target optical system in the first target field of view; A fingerprint image is extracted from the wavefront aberration image of the first target field of view, wherein the fingerprint image is an image obtained by superimposing multiple images with different intensities and rotation angles, providing information related to the amount of misalignment in the wavefront aberration; The misalignment of the optical lenses in the target optical system is determined based on the fingerprint image, and the optical lenses of the target optical system are assembled based on the misalignment. The expression for the fingerprint image is: in, It is the first The weights corresponding to the Xiang Zernick coefficients, It is the first disordered system The value of the Xiang Zernike coefficient, It is the nominal system number The value of the Xiang Zernike coefficient, S is the normalized image of the j-th Zernike polynomial over the circular domain, and S is the set of the three wave aberration coefficients most sensitive to each degree of freedom.

2. The rapid assembly method for optical lenses according to claim 1, characterized in that, The first target field of view includes a single target field of view or a multiple target field of view, and the wave aberration image is at least one of a single image, a complete image, and a cropped image.

3. The rapid assembly method for optical lenses according to claim 1, characterized in that, The step of determining the misalignment of the target optical system based on the fingerprint image, and assembling the optical lenses of the target optical system based on the misalignment, includes: The fingerprint image is input into a primary misalignment model, which outputs a primary misalignment value for the target optical system. Based on this primary misalignment value, the optical lenses of the target optical system are assembled for the first misalignment; and / or, The fingerprint image of the optical lens of the target optical system after a first misalignment assembly is input into the second misalignment model. The second misalignment model outputs the second misalignment amount of the target optical system. Based on the second misalignment amount, the optical lens of the target optical system is assembled for a second misalignment.

4. The rapid assembly method for optical lenses according to claim 1, characterized in that, Before inputting the fingerprint image into a single misalignment model, the following steps are included: Select an image of any second target field of view from the fingerprint image.

5. The rapid assembly method for optical lenses according to claim 4, characterized in that, Before inputting the fingerprint image into the first-order misalignment model, the following steps are also included: Multiple dysfunctional systems are generated based on the completed nominal system design; Calculate the wavefront aberration of each misaligned system under a preset field of view, and generate a fingerprint image of each misaligned system based on the wavefront aberration of each misaligned system under the preset field of view; The first training dataset for assembly is generated based on the fingerprint image of each misaligned system and the misalignment of the optical lens. The neural network is trained using the training dataset obtained from the first setup until the training stopping condition is met, at which point the first misalignment model is obtained.

6. The rapid assembly method for optical lenses according to claim 5, characterized in that, Before inputting the fingerprint image after the optical lenses of the target optical system have undergone a first misalignment assembly into the second misalignment model, the method further includes: The training dataset of the first assembly is used to input the first misfit model, the first misfit model outputs the first misfit of multiple misfit systems, and the first misfit assembly is performed on the multiple misfit systems based on the first misfit; Based on the fingerprint images and misalignment values ​​of the multiple misaligned systems after the first misalignment assembly, a training dataset for the second misalignment assembly is generated. The first misalignment model is trained using the training dataset until the training stopping condition is met, at which point the second misalignment model is obtained.

7. The rapid assembly method for optical lenses according to claim 5 or 6, characterized in that, The training termination conditions include at least one of the following: the training duration reaches the target duration and the training loss value is less than the target value. The training loss value is calculated using a network loss function, which is: in, It refers to the number of degrees of freedom in assembly and adjustment. It is the actual misalignment of the i-th degree of freedom. It is the network prediction misalignment of the i-th degree of freedom.

8. A rapid optical lens assembly apparatus, suitable for control using the rapid optical lens assembly method according to any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire the wavefront aberration image of the target optical system in the first target field of view; The extraction module is used to extract a fingerprint image from the wave aberration image of the first target field of view, wherein the fingerprint image is an image obtained by superimposing multiple images with different intensities and rotation angles, providing information related to the amount of misalignment in the wave aberration; The processing module is used to determine the misalignment of the optical lenses in the target optical system based on the fingerprint image, and to assemble the optical lenses of the target optical system based on the misalignment.

9. A mechanical device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the rapid assembly method for optical lenses as described in any one of claims 1-7.

Images