Splicing richea mang detection method and system based on neural network model training

Abstract

This invention provides a spliced ​​Rickie angle detection method and system based on neural network model training. It combines the sensitivity rules of aberration order and Rickie angle, and through angle differentiation configuration, combined with the reconstruction of edge region weights by the fusion detection model, it can offset the influence of weak edge aberration signals without sensitive blind spots. This significantly reduces the deviation between the detection accuracy of the edge region and the center region, and realizes high-precision detection of large-aperture planar aberrations of all orders. In particular, it significantly improves the aberration detection accuracy of the edge region and solves the problems of large accuracy deviation between the edge region and the center region and insufficient coverage of aberrations of all orders in traditional methods.

Description

Technical Field

[0001] This invention relates to the technical field of plane mirror detection, and discloses a spliced ​​Richcomb detection method and system based on neural network model training. Background Technology

[0002] As a core component of high-power microwave devices and large optical systems, large-aperture plane mirrors directly affect beam transmission accuracy and energy concentration due to their surface aberrations. High-precision detection of aberrations of all orders is required. In particular, due to factors such as uneven microwave reflection and stress concentration in the edge region, the aberration gradient is significantly higher than that in the central region. As a result, although the traditional Richter-Common detection method achieves aberration detection through the principle of multi-angle interference, it is difficult to effectively capture the aberration signal in the edge region due to the weaker aberration signal and larger gradient in the large-aperture mirror detection. Ultimately, this results in a significant deviation in the aberration detection accuracy between the edge region and the central region, which cannot meet the requirements of high-end optical systems for uniform and high-precision detection across the entire aperture. Summary of the Invention

[0003] To address the aforementioned technical problems, the main objective of this invention is to provide a spliced ​​Richcomb detection method and system based on neural network model training.

[0004] This invention first provides a spliced ​​Richmond detection method based on neural network model training, comprising:

[0005] Based on the physical parameters of the large-aperture plane mirror to be tested and the range of aberration order of the target detection, three Rickey angles are output. Each Rickey angle corresponds to a different aberration order, and the angle between any two Rickey angles satisfies the preset orthogonal constraint.

[0006] The large-aperture plane mirror under test is scanned according to the preset sub-aperture division rules, and interference fringe images of each sub-aperture under three Rickey angles are acquired.

[0007] The interference fringe image is input into the trained fusion detection model, which outputs aberrations at different aberration orders. The fusion detection model includes an aberration fusion layer, which is used to combine a sensitivity matrix to reconstruct the pixel feature weights of the edge regions in the interference fringe image. The sensitivity matrix is ​​obtained based on sensitivity calculation for each Richter angle under a standard lens.

[0008] Optionally, the three Rickey angles include a first Rickey angle corresponding to low-order aberrations, a second Rickey angle corresponding to intermediate-order aberrations, and a third Rickey angle corresponding to high-order aberrations; the step of outputting three Rickey angles based on the acquired physical parameters of the large-aperture plane mirror under test and the target detection aberration order range includes:

[0009] Construct an aberration order-Ritchey angle sensitivity correlation model based on the physical parameters of the large-aperture flat mirror to be measured and the range of target detection aberration orders.

[0010] Determine the range of aberration orders corresponding to each Ritchey angle according to the aberration order-Ritchey angle sensitivity correlation model.

[0011] Calculate three Ritchey angles that satisfy the preset orthogonality constraint, and output the three Ritchey angles.

[0012] Optionally, the aberration order-Ritchey angle sensitivity correlation model satisfies that the spatial frequency of the i-th order aberration is obtained by multiplying a proportionality constant related to the aberration type as the first multiplication factor and the ratio of the aberration order to the aperture of the large-aperture flat mirror to be measured as the second multiplication factor; the sensitivity of the i-th order aberration to the Ritchey angle is obtained by multiplying the spatial frequency of the i-th order aberration by the order coefficient power of the Ritchey angle sine function.

[0013] Optionally, the determination of the range of aberration orders corresponding to each Ritchey angle includes:

[0014] Set the low-order aberration order range as 1 to k2, the middle-order aberration order range as k2 + 1 to m, and the high-order aberration order range as m + 1 to p, where k2 < m < p and all are positive integers.

[0015] Calculate the total sensitivity of each aberration order range at different Ritchey angles, and match the Ritchey angles with the highest total sensitivity to the corresponding aberration order ranges respectively to obtain the range of aberration orders corresponding to each Ritchey angle.

[0016] Optionally, the preset orthogonality constraint is that the included angle between any two Ritchey angles is greater than or equal to 60°, and the overlap rate of the sensitive aberration order ranges corresponding to any two Ritchey angles is less than or equal to 30%.

[0017] The calculation of three Ritchey angles that satisfy the preset orthogonality constraint includes:

[0018] Based on the aberration order-Ritchey angle sensitivity correlation model, preliminarily calculate three initial Ritchey angles that respectively match the low-order aberration, middle-order aberration, and high-order aberration.

[0019] If the included angle between any two initial Ritchey angles is less than 60° or the overlap rate of the sensitive aberration order ranges is greater than 30%, then iteratively fine-tune the initial Ritchey angles until the preset orthogonality constraint is satisfied, and output the adjusted three Ritchey angles.

[0020] Optionally, the scanning of the large-aperture flat mirror to be measured according to the preset sub-aperture division rule and the acquisition of the interference fringe images of each sub-aperture at three Ritchey angles include:

[0021] A two-dimensional coordinate system is established with the center of the large-aperture plane mirror to be tested as the origin, and the mirror surface is divided into N×N square sub-apertures, wherein the side length of the sub-aperture in the edge region is smaller than the side length of the sub-aperture in the center region.

[0022] The sub-aperture scanning path is set to a serpentine path, and the scanning overlap rate of the sub-apertures in the edge region is higher than that of the sub-apertures in the center region.

[0023] The scanning device is moved sequentially along the scanning path, and three Rickey angles are switched synchronously at each sub-aperture position. Interference fringe images under each Rickey angle are acquired and associated with the sub-aperture coordinates.

[0024] Optionally, the scanning overlap rate of the edge region sub-aperture being higher than that of the center region sub-aperture includes:

[0025] The scanning overlap rate of the sub-aperture in the central region is set to 30%~40%, and the scanning overlap rate of the sub-aperture in the edge region is set to 40%~50%, where the edge region is defined by the sub-aperture center coordinates (x, y) satisfying... The region is defined by D, where D is the diameter of the large-aperture plane mirror to be tested.

[0026] Optionally, the sensitivity matrix is ​​obtained by performing sensitivity calculations on each Richter angle under a standard lens, including:

[0027] A standard mirror is placed in the detection optical path, and the true values ​​of each order of aberration of the standard mirror are known.

[0028] Interference fringe images of the standard mirror were acquired at the three Rickey angles, and the measured values ​​of each order of aberration were obtained after preprocessing.

[0029] The sensitivity coefficient of each Ridge angle to the i-th order aberration is calculated as the ratio of the absolute value of the difference between the measured value of each order aberration and the true value of each order aberration at the corresponding Ridge angle to the true value of each order aberration.

[0030] The sensitivity coefficients are arranged according to the aberration order - Rickey angle dimension to obtain the sensitivity matrix.

[0031] Optionally, the fusion detection model further includes a sub-aperture spatial feature extraction layer;

[0032] The interference fringe image is input into the trained fusion detection model, and the fusion detection model outputs aberrations at different aberration orders, including:

[0033] The interference fringe image is converted into wavefront phase data and correlated with sub-aperture coordinates to obtain model input data;

[0034] The sub-aperture spatial feature extraction layer extracts features from the input data, wherein the wavefront phase data of the sub-aperture in the edge region is extracted using a convolution kernel with a size greater than a set threshold.

[0035] The aberration fusion layer combines the sensitivity matrix to perform weighted fusion of the extracted features, and the pixel feature weights of the edge regions in the weighted fusion are dynamically adjusted through a spatial weight function.

[0036] Output the aberration quantization values ​​for different aberration orders.

[0037] Optionally, the spatial weighting function is obtained based on the edge enhancement coefficient, the sub-aperture center coordinates, and the aperture of the large-aperture plane mirror to be tested;

[0038] The aberration fusion layer, in conjunction with the sensitivity matrix, performs weighted fusion of the extracted features, including:

[0039] Based on the sensitivity matrix, determine the basic weight values ​​for each Rickey angle for different aberration orders;

[0040] The final fusion weight is obtained by multiplying the base weight value by the spatial weight function value of the corresponding sub-aperture.

[0041] The features at different Rickey angles are weighted and summed according to the final fusion weights to obtain the aberration quantization values ​​at different aberration orders.

[0042] The present invention further provides a spliced ​​Richcomb detection system based on neural network model training, the Richcomb detection system comprising:

[0043] The Ridge angle generator is used to generate three detection light fields, each with a different Ridge angle aberration order and which are orthogonal to each other;

[0044] The sub-aperture scanning device divides the large-aperture plane mirror to be tested into sub-apertures and scans them according to a preset path.

[0045] The acquisition device captures interference fringe images of the sub-aperture under three detection light fields;

[0046] The processing device is used to input the interference fringe image into a trained fusion detection model. The fusion detection model outputs aberrations at different aberration orders. The fusion detection model includes an aberration fusion layer, which is used to combine a sensitivity matrix to reconstruct the pixel feature weights of the edge regions in the interference fringe image. The sensitivity matrix is ​​obtained based on sensitivity calculations for each Richter angle under a standard lens.

[0047] The beneficial effects of this invention are:

[0048] This invention provides a spliced ​​Rickie angle detection method and system based on neural network model training. It combines the sensitivity rules of aberration order and Rickie angle, and through angle differentiation configuration, combined with the reconstruction of edge region weights by the fusion detection model, it can offset the influence of weak edge aberration signals without sensitive blind spots. This significantly reduces the deviation between the detection accuracy of the edge region and the center region, and realizes high-precision detection of large-aperture planar aberrations of all orders. In particular, it significantly improves the aberration detection accuracy of the edge region and solves the problems of large accuracy deviation between the edge region and the center region and insufficient coverage of aberrations of all orders in traditional methods. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0050] Figure 1 This is a flowchart of the spliced ​​Richmond detection method based on neural network model training according to the present invention;

[0051] Figure 2 This is a structural diagram of the spliced ​​Richcomb detection system based on neural network model training according to the present invention.

[0052] Reference numerals: 1. Rickey angle generator; 2. Sub-aperture scanning device; 3. Processing device. Detailed Implementation

[0053] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0054] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0055] This invention provides a spliced ​​Richmond detection method based on neural network model training, such as... Figure 1 As shown, it includes:

[0056] 101: Based on the physical parameters of the large-aperture plane mirror to be tested and the range of aberration order of the target detection, three Rickey angles are output. Each Rickey angle corresponds to a different aberration order, and the angle between any two Rickey angles satisfies the preset orthogonal constraint.

[0057] 102: Scan the large-aperture plane mirror under test according to the preset sub-aperture division rules and collect interference fringe images of each sub-aperture under the three Rickey angles;

[0058] 103: Input the interference fringe image into the trained fusion detection model. The fusion detection model outputs aberrations at different aberration orders. The fusion detection model includes an aberration fusion layer. The aberration fusion layer is used to combine the sensitivity matrix to reconstruct the pixel feature weights of the edge region in the interference fringe image. The sensitivity matrix is ​​obtained based on sensitivity calculation for each Rickey angle under a standard lens.

[0059] It should be noted that, in this embodiment of the application, the large-aperture plane mirror under test is specifically a planar reflecting element with a large physical size in the optical system, and its surface aberration directly affects the transmission and focusing performance of the light beam. The aberration order is a level based on the spatial distribution characteristics of the aberration. For example, low-order aberrations correspond to overall deviations in surface shape, while high-order aberrations correspond to local minor distortions. Different orders of aberrations have different effects on the performance of the optical system.

[0060] This application research found that existing technologies for detecting large-aperture plane mirrors generally only consider overall aberrations. However, in reality, higher-order and intermediate-order aberrations in the edge regions of large-aperture plane mirrors also significantly affect the performance of the optical system. Existing technologies, on the one hand, ignore the influence of higher-order and intermediate-order aberrations, and on the other hand, cannot simultaneously combine the aberrations of the three orders, resulting in low detection accuracy.

[0061] This application provides a spliced ​​Rickie angle detection method based on neural network model training. Its core concept lies in combining the sensitivity rules of aberration order and Rickie angle. By configuring angle differences and reconstructing the weights of edge regions using a fusion detection model, the influence of weak edge aberration signals can be offset without sensitive blind spots. This significantly reduces the deviation between the detection accuracy of edge regions and the central region, achieving high-precision detection of large-aperture planar aberrations of all orders. In particular, it significantly improves the aberration detection accuracy of edge regions, solving the problems of large accuracy deviation between edge regions and central regions and insufficient coverage of aberrations of all orders in traditional methods.

[0062] First, based on the physical parameters of the large-aperture plane mirror to be tested, such as aperture and material (which can be directly obtained through measurement tools), and the target detection aberration order range (such as the upper and lower limits of the specific aberration order to be detected, which can be preset according to the performance requirements of the optical system), three Ridge angles are output.

[0063] In this embodiment, the Rickey angle is specifically the incident angle between the light field and the mirror under test, and its magnitude affects the sensitivity to different orders of aberrations. Each Rickey angle corresponds to a different aberration order, and the angle between any two Rickey angles satisfies a preset orthogonal constraint. This preset orthogonal constraint can prevent excessive overlap of the sensitivity ranges of different Rickey angles to aberrations, ensuring that each order of aberration can be captured independently and effectively.

[0064] Secondly, the large-aperture plane mirror under test is scanned according to a preset sub-aperture division rule, and interference fringe images of each sub-aperture at three Rickey angles are acquired. In this embodiment, the sub-aperture specifically refers to multiple small-sized regions into which the large-aperture plane mirror is divided. Since the large-aperture plane mirror cannot be fully covered by a single detection, this embodiment achieves full-mirror detection by dividing and scanning the sub-apertures.

[0065] Interference fringe images are images formed by the interference of a reference light after the detection light field is reflected by a mirror. The fringe pattern contains aberration information and can be acquired by imaging devices, such as CCD cameras. This application does not limit this.

[0066] Finally, the interference fringe image is input into the trained fusion detection model, which outputs aberrations at different aberration orders. The fusion detection model is a trained neural network model containing an aberration fusion layer. This layer combines the sensitivity matrix to reconstruct the pixel feature weights of edge regions in the interference fringe image. It should be noted that the sensitivity matrix is ​​obtained by calculating the sensitivity of each Ridge angle under a standard mirror. The standard mirror is a high-precision mirror with known true values ​​for each aberration order. The sensitivity of different Ridge angles to each aberration order is calculated using detection data on the standard mirror, thus forming the matrix.

[0067] In an optional embodiment, the three Rickey angles include a first Rickey angle corresponding to low-order aberrations, a second Rickey angle corresponding to mid-order aberrations, and a third Rickey angle corresponding to high-order aberrations; the step of outputting the three Rickey angles based on the acquired physical parameters of the large-aperture plane mirror under test and the target detection aberration order range includes:

[0068] Based on the physical parameters of the large-aperture plane mirror under test and the range of target detection aberration order, an aberration order-Ritchie angle sensitivity correlation model is constructed.

[0069] Based on the aforementioned aberration order-Riche angle sensitivity correlation model, determine the range of aberration orders corresponding to each Riche angle;

[0070] Calculate the three Rickey angles that satisfy the preset orthogonal constraints, and output the three Rickey angles.

[0071] In this embodiment, the three Rickey angles include a first Rickey angle corresponding to lower-order aberrations, a second Rickey angle corresponding to middle-order aberrations, and a third Rickey angle corresponding to higher-order aberrations. The classification of Rickey angles can be determined based on the differences in sensitivity of aberration order to Rickey angles; lower-order aberrations are more sensitive to smaller incident angles, while higher-order aberrations are more sensitive to larger incident angles.

[0072] The process of outputting three Rickey angles based on the obtained physical parameters of the large-aperture plane mirror under test and the target detection aberration order range includes the following steps:

[0073] First, based on the physical parameters of the large-aperture plane mirror under test and the range of target detection aberration orders, an aberration order-Ritchie angle sensitivity correlation model is constructed. This model is used to describe the sensitivity of different orders of aberrations to different Ritchie angles. It is understood that, in the embodiments of this application, the spatial characteristics of aberrations are physically correlated with the Ritchie angle; higher-order aberrations exhibit more drastic spatial changes and are therefore more sensitive to larger Ritchie angles.

[0074] Secondly, based on the aforementioned aberration order-Ritchie angle sensitivity correlation model, the aberration order range corresponding to each Ritchie angle is determined. In this step, it is necessary to clarify the aberration order most suitable for detection for each Ritchie angle. Specifically, it can be found through model calculation that a certain Ritchie angle has the highest sensitivity to aberrations of orders 1 to 3, and then it is assigned to the lower-order aberration range.

[0075] Finally, the three Rickey angles that satisfy the preset orthogonality constraint are calculated and output. Specifically, a Rickey angle is selected from each aberration order, and then an iterative operation is performed. If they are not pairwise orthogonal, other Rickey angles are selected again until the Rickey angles that are pairwise orthogonal in each of the three aberration orders are selected.

[0076] In an optional embodiment, the aberration order-Ritchie angle sensitivity correlation model satisfies the following: the spatial frequency of the i-th order aberration is obtained by multiplying the first multiplication factor by taking the proportionality constant related to the aberration type as the first multiplication factor and the ratio of the aberration order to the aperture of the large-aperture plane mirror under test as the second multiplication factor; the sensitivity of the i-th order aberration to the Ritchie angle is obtained by multiplying the spatial frequency of the i-th order aberration by the power of the coefficient of the order of the Ritchie angle sine function.

[0077] Specifically, the spatial frequency of the i-th order aberration. Where k1 is a proportionality constant related to the aberration type, i is the aberration order, and D is the aperture of the large-aperture plane mirror to be tested; the i-th order aberration corresponds to the Rickey angle. Sensitivity Where n is the order coefficient. Specifically, n can be set to 1 for low-order aberrations, 1.5 for medium-order aberrations, and 2 for high-order aberrations. Of course, this application does not impose any restrictions on the value of n, which can be set according to specific circumstances.

[0078] In this embodiment of the application, the aberration order-Ritchie angle sensitivity correlation model is used to quantify the sensitivity of different orders of aberrations to different Ritchie angles, and it is constructed based on the trigonometric function relationship between the spatial frequency of the aberration and the Ritchie angle.

[0079] It should be noted that, in this embodiment, the spatial frequency of the i-th order aberration is specifically a characterization index of how quickly the aberration changes spatially. The higher the order of the aberration, the higher the spatial frequency, which can be calculated using parameters such as the aberration order and the mirror aperture. In the model, the spatial frequency of the i-th order aberration... It is related to the aberration order i and the aperture D of the large-aperture plane mirror under test, and also includes a proportionality constant k1 related to the aberration type. For example, different types of aberrations, such as astigmatism and coma, have different correlation coefficients between their spatial frequency and order, which will not be elaborated in this application.

[0080] In the embodiments of this application, in the aberration order-Ritchie angle sensitivity correlation model, Spatial frequency represents the rate at which aberrations change in mirror space. The higher the order of the aberration, the denser the details of the surface distortion; that is, higher-order aberrations have more local micro-bulges and micro-depressions, and the more drastic the spatial changes. The larger the aperture D, the wider the spatial distribution range of aberrations of the same order, and the relatively lower the spatial frequency.

[0081] It should be noted that, in this embodiment, aberrations with higher spatial frequencies theoretically require more sensitive detection conditions (such as large Rickey angles) to be captured. Therefore, the spatial frequency... As one of the factors affecting sensitivity.

[0082] This is the quantized value of the incident effect at the Rickie angle. It is the angle between the light field and the normal of the mirror being tested. The angle of incidence (Riche angle) is used to characterize the intensity of a light beam's response to local changes in a mirror surface. The larger the angle of incidence (Riche angle), the higher the light beam's perception of local surface undulations during mirror reflection. Follow It increases and increases, among which, , The larger, The larger the value, the stronger the beam's response to aberrations, and the higher its sensitivity.

[0083] n is the order coefficient. Specifically, the spatial variation of low-order aberrations (such as defocus) is gentle, and the signal intensity is affected slowly by the incident angle. Therefore, n takes a lower value. along with gentle growth to avoid excessive amplification of the sensitivity of low-order aberrations; the spatial variation of high-order aberrations (such as cross coma) is drastic, and the signal intensity is affected faster by the incident angle. Therefore, n takes a higher value. along with rapid growth to enable a large Rayleigh angle to significantly enhance the sensitivity to high-order aberrations; the spatial variation of medium-order aberrations is between that of low-order aberrations and high-order aberrations.

[0084] In an optional embodiment, the determining the aberration order range corresponding to each Rayleigh angle includes:

[0085] Setting the low-order aberration order range as 1 to k2, the medium-order aberration order range as k2 + 1 to m, and the high-order aberration order range as m + 1 to p, where k2 < m < p and all are positive integers;

[0086] Calculating the total sensitivity of each aberration order range at different Rayleigh angles, and matching the Rayleigh angle with the highest total sensitivity to the corresponding aberration order range to obtain the aberration order range corresponding to each Rayleigh angle.

[0087] In the embodiment of the present application, the determining the aberration order range corresponding to each Rayleigh angle is to clarify the detection division of labor for each Rayleigh angle and ensure that all-order aberrations can be efficiently captured.

[0088] Specifically, this process includes: First, setting the low-order aberration order range as 1 to k2, the medium-order aberration order range as k2 + 1 to m, and the high-order aberration order range as m + 1 to p, where k2, m, and p are positive integers and k2 < m < p. It should be noted that the division of these ranges can be adjusted according to actual detection requirements, and the present application is not limited thereto.

[0089] Second, calculating the total sensitivity of each aberration order range at different Rayleigh angles. The total sensitivity refers to the sum of the sensitivities of a certain Rayleigh angle to all aberrations within a certain aberration order range, and through this sum, the overall sensitivity of this Rayleigh angle to the aberrations in this range can be judged. Finally, matching the Rayleigh angle with the highest total sensitivity to the corresponding aberration order range, so as to obtain the aberration order range corresponding to each Rayleigh angle, ensuring that the aberrations in each range can be detected by the most sensitive Rayleigh angle.

[0090] Exemplarily, for the sake of easy understanding, the present application provides an example for auxiliary explanation. It should be understood that the numerical values involved in the following example are numerical examples assumed for the convenience of understanding and do not represent the specific data in actual detection. The present application does not elaborate herein.

[0091] Assuming the target detection aberration order ranges from 1 to 10, it is divided as follows:

[0092] Low-order aberrations range: 1st to 3rd order (i.e., k2=3);

[0093] Intermediate aberration range: 4th to 6th order (i.e., m=6);

[0094] Higher-order aberrations range: 7th to 10th order (i.e., p=10);

[0095] Suppose there are three Ricky Points to be evaluated: (small angle), (medium angle), (wide angle).

[0096] According to the sensitivity formula Where lower-order aberrations n=1, middle-order aberrations n=1.5, and higher-order aberrations n=2, The sensitivity of each Ridge angle to aberrations of each order increases with the aberration order i. As shown in Table 1, the sensitivity of each Ridge angle to aberrations of each order can be calculated.

[0097] Table 1. Sensitivity of different Ridge angles to aberrations of various orders.

[0098]

[0099] Calculate the sum of the sensitivities of the three Rickey angles in the low-order (1~3), mid-order (4~6), and high-order (7~10) ranges, specifically including:

[0100] Sum of sensitivities:

[0101] Lower order range: 0.8 + 0.7 + 0.6 = 2.1;

[0102] Intermediate range: 0.3 + 0.2 + 0.1 = 0.6;

[0103] Higher order range: 0.1 + 0.05 + 0.03 + 0.02 = 0.2;

[0104] Sum of sensitivities:

[0105] Lower order range: 0.6 + 0.5 + 0.4 = 1.5;

[0106] Intermediate range: 0.7 + 0.8 + 0.6 = 2.1;

[0107] Higher order range: 0.4 + 0.3 + 0.2 + 0.1 = 1.0;

[0108] Sum of sensitivities:

[0109] Low-order range: 0.3 + 0.2 + 0.1 = 0.6;

[0110] Intermediate range: 0.5 + 0.6 + 0.7 = 1.8;

[0111] Higher order range: 0.8 + 0.9 + 0.95 + 0.85 = 3.5;

[0112] Compare the sum of the sensitivity of the three Rickey angles in each aberration range, and match the Rickey angle with the highest sum to that range. Specifically, this includes:

[0113] Lower order range (orders 1-3): The total sensitivity (2.1) is the highest → Corresponding to lower-order aberrations;

[0114] Intermediate range (4th to 6th order): The total sensitivity (2.1) is the highest → Corresponding to intermediate-order aberrations;

[0115] Higher order range (7th to 10th order): The total sensitivity (3.5) was the highest → Corresponding to higher-order aberrations;

[0116] Therefore, it can be determined that Corresponding to lower-order aberrations, Corresponding to intermediate-order aberrations, Corresponding to higher-order aberrations.

[0117] In an optional embodiment, the preset orthogonal constraint is that the included angle between any two Rickey angles is greater than or equal to 60°, and the overlap rate of the sensitive aberration order ranges corresponding to any two Rickey angles is less than or equal to 30%.

[0118] The calculation of the three Ritchie angles that satisfy the preset orthogonal constraint includes:

[0119] Based on the aforementioned aberration order-Ritchie angle sensitivity correlation model, three initial Ritchie angles are initially calculated to match low-order aberrations, medium-order aberrations, and high-order aberrations, respectively.

[0120] If the angle between any two initial Rickey angles is less than 60° or the overlap rate of the sensitive aberration order range is greater than 30%, the initial Rickey angles are iteratively fine-tuned until the preset orthogonal constraint is met, and the three adjusted Rickey angles are output.

[0121] In this embodiment, the preset orthogonal constraint is used to ensure that the detection ranges of the three Rickey angles are independent of each other, avoiding mutual interference of aberration signals. Specifically, the constraint includes two aspects: first, the included angle between any two Rickey angles must meet certain conditions; second, the overlap rate of the sensitive aberration order ranges corresponding to any two Rickey angles must be controlled within a certain range.

[0122] It should be noted that, in this embodiment of the application, the included angle of the Rickey angle is specifically the angular difference between the three Rickey angles. A larger included angle can reduce the overlap of their aberration-sensitive ranges. In this embodiment of the application, the overlap rate of the sensitive aberration order range is specifically the proportion of the overlapping part in the sensitive aberration order range corresponding to two Rickey angles to the total range. A lower overlap rate can ensure that the aberration order detected by each Rickey angle is relatively independent.

[0123] The calculation of three Rickey angles that satisfy the preset orthogonality constraint specifically includes: First, based on the aberration order-Rickey angle sensitivity correlation model, three initial Rickey angles are initially calculated to match low-order, mid-order, and high-order aberrations, respectively. These initial values ​​are determined based on the highest sensitivity of each Rickey angle to the target aberration order. Second, if the included angle or the overlap rate of the sensitive aberration order range between any two initial Rickey angles does not satisfy the preset orthogonality constraint, the initial Rickey angles are iteratively fine-tuned, that is, the size of the Rickey angles is gradually adjusted, and the included angle and overlap rate are recalculated after each adjustment until the orthogonality constraint condition is met, and finally the three adjusted Rickey angles are output.

[0124] In an optional embodiment, the step of scanning the large-aperture plane mirror under test according to a preset sub-aperture division rule and acquiring interference fringe images of each sub-aperture at three Rickey angles includes:

[0125] A two-dimensional coordinate system is established with the center of the large-aperture plane mirror to be tested as the origin, and the mirror surface is divided into N×N square sub-apertures, wherein the side length of the sub-aperture in the edge region is smaller than the side length of the sub-aperture in the center region.

[0126] The sub-aperture scanning path is set to a serpentine path, and the scanning overlap rate of the sub-apertures in the edge region is higher than that of the sub-apertures in the center region.

[0127] The scanning device is moved sequentially along the scanning path, and three Rickey angles are switched synchronously at each sub-aperture position. Interference fringe images under each Rickey angle are acquired and associated with the sub-aperture coordinates.

[0128] In this embodiment, scanning the large-aperture plane mirror under test according to the preset sub-aperture division rules and acquiring interference fringe images of each sub-aperture at three Rickey angles is to obtain aberration information of the entire aperture, especially for targeted optimization of the edge region.

[0129] The specific process includes: First, establishing a two-dimensional coordinate system with the center of the large-aperture plane mirror to be tested as the origin. This coordinate system is used to locate the position of each sub-aperture. The mirror surface is divided into N×N square sub-apertures, where the side length of the edge region sub-apertures is smaller than the side length of the center region sub-apertures. It should be noted that the aberrations in the edge region are more complex, and smaller sub-apertures can improve local detection accuracy.

[0130] Specifically, a sub-aperture scanning path can be set to ensure continuous coverage of the sub-apertures and reduce scanning omissions. This application does not impose restrictions on the scanning path setting. At the same time, the scanning overlap rate of the sub-apertures in the edge region is higher than that in the center region. In this embodiment, the overlap rate is specifically the proportion of overlapping areas of adjacent sub-apertures. A higher overlap rate can enhance the continuity of aberration information in the edge region.

[0131] Finally, the scanning device is moved sequentially along the scanning path, simultaneously switching three Ridge angles at each sub-aperture position, acquiring interference fringe images at each Ridge angle, and associating the sub-aperture coordinates. Simultaneous switching of Ridge angles ensures consistent detection conditions for the same sub-aperture at different Ridge angles, while associating the sub-aperture coordinates provides a positional basis for subsequent spatial stitching of aberrations.

[0132] In an optional embodiment, the scanning overlap rate of the edge region sub-aperture being higher than that of the center region sub-aperture includes:

[0133] The scanning overlap rate of the sub-aperture in the central region is set to 30%~40%, and the scanning overlap rate of the sub-aperture in the edge region is set to 40%~50%, where the edge region is defined by the sub-aperture center coordinates (x, y), satisfying the following conditions: The region is defined by D, where D is the diameter of the large-aperture plane mirror to be tested.

[0134] In this embodiment, the scanning overlap rate of the sub-aperture in the edge region is higher than that in the center region in order to enhance the aberration detection accuracy in the edge region. This is because the aberration gradient in the edge region is larger, requiring more redundant information to ensure detection accuracy.

[0135] In an optional embodiment, the sensitivity matrix is ​​obtained by performing sensitivity calculations on each Richter angle under a standard lens, including:

[0136] A standard mirror is placed in the detection optical path, and the true values ​​of each order of aberration of the standard mirror are known.

[0137] Interference fringe images of the standard mirror were acquired at the three Rickey angles, and the measured values ​​of each order of aberration were obtained after preprocessing.

[0138] The sensitivity coefficient of each Ridge angle to the i-th order aberration is calculated as the ratio of the absolute value of the difference between the measured value and the true value of each order aberration at the corresponding Ridge angle to the true value of each order aberration. Specifically, the sensitivity coefficient S(i,j) of each Ridge angle to the i-th order aberration is calculated as |measured value(i,j) - true value(i)| / true value(i), where j is the Ridge angle index.

[0139] The sensitivity coefficients are arranged according to the aberration order - Rickey angle dimension to obtain the sensitivity matrix.

[0140] It should be noted that, in this embodiment, the central region specifically refers to the region near the center of the mirror, where the aberration distribution is relatively gentle; the edge region specifically refers to the region near the edge of the mirror, which can be defined by the relationship between the sub-aperture center coordinates and the mirror aperture. For example, when the sub-aperture center coordinates meet certain geometric conditions, it can be determined that it belongs to the edge region.

[0141] The scanning overlap rate of the sub-apertures in the central region and the scanning overlap rate of the sub-apertures in the edge region can be set according to the actual detection requirements. The overlap rate of the edge region is higher. The purpose is to reduce the error of edge aberration detection by using the overlap information of adjacent sub-apertures and ensure the continuity of aberration distribution.

[0142] In this embodiment of the application, the sensitivity matrix is ​​used to quantify the sensitivity of different Richter angles to aberrations of each order, providing a basis for aberration fusion of the fusion detection model, and its acquisition is based on the calibration experiment of the standard mirror.

[0143] The specific process includes: First, a standard mirror is placed in the detection optical path. The standard mirror is a high-precision plane mirror with known surface shape accuracy, and its true values ​​of each order of aberration can be obtained in advance through a higher-precision detection method.

[0144] Secondly, interference fringe images of the standard mirror are acquired at the three Rickey angles respectively. After preprocessing these images, for example, noise reduction and phase unwrapping, effective aberration information is extracted to obtain the measured values ​​of each order of aberration at each Rickey angle.

[0145] Then, the sensitivity coefficient of each Rickey angle to the i-th order aberration is calculated. This coefficient is obtained by the deviation between the measured value and the true value, and reflects the sensitivity of the Rickey angle to the i-th order aberration. The smaller the deviation, the higher the sensitivity.

[0146] Finally, the sensitivity coefficients are arranged according to the aberration order - Ridge angle dimension, that is, the rows represent the aberration order and the columns represent the Ridge angle, forming a sensitivity matrix, which can be directly used as prior knowledge for the fusion detection model.

[0147] In an optional embodiment, the fusion detection model further includes a sub-aperture spatial feature extraction layer;

[0148] The interference fringe image is input into the trained fusion detection model, and the fusion detection model outputs aberrations at different aberration orders, including:

[0149] The interference fringe image is converted into wavefront phase data and correlated with sub-aperture coordinates to obtain model input data;

[0150] The sub-aperture spatial feature extraction layer extracts features from the input data, wherein the wavefront phase data of the sub-aperture in the edge region is extracted using a convolution kernel with a size greater than a set threshold.

[0151] The aberration fusion layer combines the sensitivity matrix to perform weighted fusion of the extracted features, and the pixel feature weights of the edge regions in the weighted fusion are dynamically adjusted through a spatial weight function.

[0152] Output the aberration quantization values ​​for different aberration orders.

[0153] In this embodiment, the fusion detection model is used to extract and fuse aberration information from the interference fringe image. It includes a sub-aperture spatial feature extraction layer and an aberration fusion layer, which can achieve accurate output of aberrations of all orders.

[0154] The process of inputting the interference fringe image into the trained fusion detection model and outputting aberrations at different aberration orders specifically includes: First, converting the interference fringe image into wavefront phase data. The wavefront phase data is a quantified form of the aberrations contained in the interference fringe image, which can be calculated from the interference fringe image using a phase extraction algorithm, for example, the Fourier transform method. Simultaneously, associating the wavefront phase data with the sub-aperture coordinates, the model input data is obtained, ensuring that the aberration information corresponds to the spatial position.

[0155] Secondly, the sub-aperture spatial feature extraction layer extracts features from the input data. This layer extracts spatial features of aberrations through convolution operations, specifically using large-size convolution kernels to extract features from the wavefront phase data of sub-apertures in edge regions. It should be noted that large-size convolution kernels can capture a wider range of spatial correlation information, making them suitable for complex aberration features in edge regions.

[0156] Then, the aberration fusion layer combines the sensitivity matrix to perform weighted fusion of the extracted features. In the weighted fusion, the pixel feature weights of the edge regions are dynamically adjusted through a spatial weight function, with the aim of increasing the contribution of aberration features in the edge regions and offsetting the influence of their weak signals.

[0157] Finally, the aberration quantization values ​​for different aberration orders are output. These values ​​are the specific magnitudes of each aberration order and can be directly used to evaluate the surface accuracy of the mirror.

[0158] In an optional embodiment, the spatial weighting function is determined based on the edge enhancement coefficient, the sub-aperture center coordinates, and the aperture of the large-aperture plane mirror under test. Specifically, the spatial weighting function is: Where (x,y) are the coordinates of the sub-aperture center, D is the aperture of the large-aperture plane mirror to be tested, and k3 is the edge enhancement coefficient. The spatial weight value of the sub-aperture in the edge region is higher than that in the center region. It is an exponential function with the natural constant as its base;

[0159] The aberration fusion layer, in conjunction with the sensitivity matrix, performs weighted fusion of the extracted features, including:

[0160] Based on the sensitivity matrix, determine the basic weight values ​​for each Rickey angle for different aberration orders;

[0161] The final fusion weight is obtained by multiplying the base weight value by the spatial weight function value of the corresponding sub-aperture.

[0162] The features at different Rickey angles are weighted and summed according to the final fusion weights to obtain the aberration quantization values ​​at different aberration orders.

[0163] In this embodiment, the spatial weighting function is used to dynamically adjust the aberration feature weights of sub-apertures at different locations, especially strengthening the weights of edge regions to ensure that edge aberrations are fully considered.

[0164] It should be noted that the edge enhancement coefficient in this function can be adjusted according to the strength of edge aberrations to adapt to different detection scenarios, and this application is not limited to this.

[0165] Regarding the aberration fusion layer, which combines the aforementioned sensitivity matrix to perform weighted fusion of extracted features, the process specifically includes: First, determining the base weight value for each Ridge angle to different aberration orders based on the sensitivity matrix; Ridge angles with higher sensitivity have larger base weight values ​​corresponding to their sensitive aberration orders. Second, multiplying the base weight value by the spatial weight function value of the corresponding sub-aperture to obtain the final fusion weight. This weight considers both the sensitivity of the Ridge angle and the spatial position of the sub-aperture, ensuring that sensitive aberration features in edge regions are enhanced. Finally, weighting and summing the features at different Ridge angles according to the final fusion weight yields the aberration quantization values ​​for different aberration orders, achieving accurate aberration fusion.

[0166] This invention also provides a spliced ​​Richcomb detection system based on neural network model training, such as... Figure 2 As shown, the Richcomm detection system includes:

[0167] The Ridge angle generator 1 is used to generate three detection light fields, each with a different Ridge angle aberration order and which are orthogonal to each other;

[0168] Sub-aperture scanning device 2 divides the large-aperture plane mirror under test into sub-apertures and scans them according to a preset path; acquisition device captures interference fringe images of the sub-apertures under three detection light fields;

[0169] Processing device 3 is used to input the interference fringe image into a trained fusion detection model. The fusion detection model outputs aberrations at different aberration orders. The fusion detection model includes an aberration fusion layer. The aberration fusion layer is used to combine a sensitivity matrix to reconstruct the pixel feature weights of the edge regions in the interference fringe image. The sensitivity matrix is ​​obtained based on sensitivity calculation for each Richter angle under a standard lens.

[0170] In this embodiment of the application, a spliced ​​Richcomb detection system based on neural network model training is provided. This system is used to implement the above detection method. Its components work together to ensure the efficiency and accuracy of aberration detection.

[0171] The Rickie angle detection system includes a Rickie angle generator 1, a sub-aperture scanning device 2, an acquisition device, and a processing device 3. The Rickie angle generator 1 generates three detection light fields, each with a Rickie angle corresponding to a different aberration order and being orthogonal to each other. It can achieve accurate control of the Rickie angle through an angle adjustment mechanism, such as a rotary table, ensuring that the light field is incident on the mirror under test at a preset angle.

[0172] The sub-aperture scanning device 2 divides the large-diameter plane mirror to be tested into sub-apertures and scans them according to a preset path. It includes a displacement mechanism that can drive the detection component to move along the preset path to realize point-by-point detection of the sub-apertures.

[0173] The acquisition device is used to capture interference fringe images of three sub-apertures under three detection light fields. It may include an imaging device, exemplarily, such as a camera, that converts optical signals into electrical signals to form a digital image.

[0174] Processing device 3 is used to input the interference fringe image into the trained fusion detection model and output the aberrations at different aberration orders, specifically as follows: Figure 2 As shown, it includes aberration order one, aberration order two, and aberration order three. The device may include a computing unit, exemplarily such as a computer or a dedicated chip, to run the fusion detection model and perform functions such as data preprocessing, feature extraction, and aberration fusion. It should be noted that the fusion detection model in processing device 3 includes an aberration fusion layer, which combines a sensitivity matrix to reconstruct the pixel feature weights of the edge region. The sensitivity matrix is ​​obtained in the same way as described above, calculated based on calibration experiments using a standard mirror.

[0175] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A spliced ​​Richmond detection method based on neural network model training, characterized in that, Including: According to the obtained physical parameters of the large-aperture plane mirror to be measured and the target detection aberration order range, output three Ritchey angles. The aberration orders corresponding to each Ritchey angle are different, and the included angle between any two Ritchey angles satisfies a preset orthogonality constraint; Scan the large-aperture plane mirror to be measured according to the preset sub-aperture division rule, and collect the interference fringe images of each sub-aperture at the three Ritchey angles; Input the interference fringe images into a trained fusion detection model. The fusion detection model outputs the aberrations at different aberration orders. The fusion detection model includes an aberration fusion layer, and the aberration fusion layer is used to combine the sensitivity matrix to reconstruct the pixel feature weights in the edge region of the interference fringe images. The sensitivity matrix is obtained based on the sensitivity calculation for each Ritchey angle under a standard mirror; Obtaining the sensitivity matrix based on the sensitivity calculation for each Ritchey angle under a standard mirror includes: Place the standard mirror in the detection optical path, and the true values of each order of aberration of the standard mirror are known; Collect the interference fringe images of the standard mirror at the three Ritchey angles respectively, and obtain the measurement values of each order of aberration for each Ritchey angle after preprocessing; Calculate the sensitivity coefficient of each Ritchey angle to the i-th order of aberration as the ratio of the absolute value of the difference between the measured values of each order of aberration and the true values of each order of aberration to the true values of each order of aberration at the corresponding Ritchey angle; Arrange the sensitivity coefficients in the aberration order - Ritchey angle dimension to obtain the sensitivity matrix.

2. The spliced ​​Richmond detection method based on neural network model training according to claim 1, characterized in that, The three Ritchey angles include a first Ritchey angle corresponding to low-order aberration, a second Ritchey angle corresponding to medium-order aberration, and a third Ritchey angle corresponding to high-order aberration; the output of the three Ritchey angles according to the obtained physical parameters of the large-aperture plane mirror to be measured and the target detection aberration order range includes: Based on the physical parameters of the large-aperture plane mirror to be measured and the target detection aberration order range, construct an aberration order - Ritchey angle sensitivity correlation model; According to the aberration order - Ritchey angle sensitivity correlation model, determine the aberration order range corresponding to each Ritchey angle respectively; Calculate three Ritchey angles that satisfy the preset orthogonality constraint, and output the three Ritchey angles.

3. The spliced ​​Richmond detection method based on neural network model training according to claim 2, characterized in that, The aberration order - Ritchey angle sensitivity correlation model satisfies: the spatial frequency of the i-th order of aberration is obtained by multiplying a proportionality constant related to the aberration type as the first multiplication factor and the ratio of the aberration order to the aperture of the large-aperture plane mirror to be measured as the second multiplication factor; the sensitivity of the i-th order of aberration to the Ritchey angle is obtained by multiplying the spatial frequency of the i-th order of aberration by the order coefficient power of the sine function of the Ritchey angle.

4. The spliced ​​Richmond detection method based on neural network model training according to claim 2, characterized in that, The determination of the aberration order range corresponding to each Ritchey angle respectively includes: Set the low-order aberration order range as 1~k2, the medium-order aberration order range as k2 + 1~m, and the high-order aberration order range as m + 1~p, where k2 < m < p and all are positive integers; Calculate the total sensitivity of each aberration order range at different Ritchey angles, and match the Ritchey angle with the highest total sensitivity to the corresponding aberration order range respectively to obtain the aberration order range corresponding to each Ritchey angle.

5. The spliced ​​Richmond detection method based on neural network model training according to claim 2, characterized in that, The preset orthogonal constraint is that the angle between any two Rickey angles is greater than or equal to 60°, and the overlap rate of the sensitive aberration order ranges corresponding to any two Rickey angles is less than or equal to 30%. The calculation of the three Ritchie angles that satisfy the preset orthogonal constraint includes: Based on the aforementioned aberration order-Ritchie angle sensitivity correlation model, three initial Ritchie angles are initially calculated to match low-order aberrations, medium-order aberrations, and high-order aberrations, respectively. If the angle between any two initial Rickey angles is less than 60° or the overlap rate of the sensitive aberration order range is greater than 30%, the initial Rickey angles are iteratively fine-tuned until the preset orthogonal constraint is met, and the three adjusted Rickey angles are output.

6. The spliced ​​Richmond detection method based on neural network model training according to claim 1, characterized in that, The process of scanning the large-aperture plane mirror under test according to the preset sub-aperture division rules and acquiring interference fringe images of each sub-aperture at three Rickey angles includes: A two-dimensional coordinate system is established with the center of the large-aperture plane mirror to be tested as the origin, and the mirror surface is divided into N×N square sub-apertures, wherein the side length of the sub-aperture in the edge region is smaller than the side length of the sub-aperture in the center region. The sub-aperture scanning path is set to a serpentine path, and the scanning overlap rate of the sub-apertures in the edge region is higher than that of the sub-apertures in the center region. The scanning device is moved sequentially along the scanning path, and three Rickey angles are switched synchronously at each sub-aperture position. Interference fringe images under each Rickey angle are acquired and associated with the sub-aperture coordinates.

7. The spliced ​​Richmond detection method based on neural network model training according to claim 6, characterized in that, The scanning overlap rate of the edge region sub-aperture being higher than that of the center region sub-aperture includes: The scanning overlap rate of the sub-aperture in the central region is set to 30%~40%, and the scanning overlap rate of the sub-aperture in the edge region is set to 40%~50%. The edge region is the region where the center coordinates (x,y) of the sub-aperture satisfy x²+y²>(D / 2)²×0.8, where D is the aperture of the large-aperture plane mirror to be tested.

8. The spliced ​​Richmond detection method based on neural network model training according to claim 1, characterized in that, The fusion detection model also includes a sub-aperture spatial feature extraction layer; The interference fringe image is input into the trained fusion detection model, and the fusion detection model outputs aberrations at different aberration orders, including: The interference fringe image is converted into wavefront phase data and correlated with sub-aperture coordinates to obtain model input data; The sub-aperture spatial feature extraction layer extracts features from the input data, wherein the wavefront phase data of the sub-aperture in the edge region is extracted using a convolution kernel with a size greater than a set threshold. The aberration fusion layer combines the sensitivity matrix to perform weighted fusion of the extracted features, and the pixel feature weights of the edge regions in the weighted fusion are dynamically adjusted through a spatial weight function. Output the aberration quantization values ​​for different aberration orders.

9. The spliced ​​Richmond detection method based on neural network model training according to claim 8, characterized in that, The spatial weighting function is obtained based on the edge enhancement coefficient, the sub-aperture center coordinates, and the aperture of the large-aperture plane mirror to be tested; The aberration fusion layer, in conjunction with the sensitivity matrix, performs weighted fusion of the extracted features, including: Based on the sensitivity matrix, determine the basic weight values ​​for each Rickey angle for different aberration orders; The final fusion weight is obtained by multiplying the base weight value by the spatial weight function value of the corresponding sub-aperture. The features at different Rickey angles are weighted and summed according to the final fusion weights to obtain the aberration quantization values ​​at different aberration orders.

10. A spliced ​​Richmond detection system based on neural network model training, used to implement the spliced ​​Richmond detection method based on neural network model training as described in any one of claims 1-9, characterized in that, The Ruichi Kangmang detection system includes: The Ridge angle generator is used to generate three detection light fields, each with a different Ridge angle aberration order and which are orthogonal to each other; The sub-aperture scanning device divides the large-aperture plane mirror to be tested into sub-apertures and scans them according to a preset path. The acquisition device captures interference fringe images of the sub-aperture under three detection light fields; The processing device is used to input the interference fringe image into a trained fusion detection model. The fusion detection model outputs aberrations at different aberration orders. The fusion detection model includes an aberration fusion layer, which is used to combine a sensitivity matrix to reconstruct the pixel feature weights of the edge regions in the interference fringe image. The sensitivity matrix is ​​obtained based on sensitivity calculations for each Richter angle under a standard lens.

Images