Multi-spectral wavefront aberration restoration method and system based on double roof prisms

Abstract

The invention relates to a multi-spectral wavefront aberration restoration method and system based on double roof prisms, and the method comprises the following steps: taking a sub-pupil image as the input of a classification model, obtaining classification feature data, and carrying out the data expansion of the classification feature data, the sub pupil image is obtained by inputting incident wavefront aberration into a simulation light path of the double-ridge prism wavefront sensor, the double-ridge prism wavefront sensor comprises two groups of double-ridge prisms, each group of double-ridge prisms comprises two ridge prisms which are vertically and oppositely arranged, the characteristics of the ridge prisms are equal, and the characteristics of the double-ridge prisms are equal. The two groups of double-roof prisms are made of a material combination for eliminating chromatic aberration; splicing the sub pupil images and the classification feature data after data expansion to serve as input of a regression model, and obtaining a Zernike coefficient matrix corresponding to incident wavefront aberration; and according to the Zernike coefficient matrix corresponding to the incident wavefront aberration and the Zernike polynomial, obtaining a recovered wavefront aberration. Compared with the prior art, the method has the advantage that the wavefront aberration recovery accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of wavefront detection technology, and in particular to a multi-spectral wavefront aberration restoration method and system based on a double-roof prism. Background Art

[0002] Aberrations in optical systems can lead to problems such as blurred imaging and reduced resolution. By restoring the original wavefront aberration, the performance of the optical system can be accurately evaluated, and corresponding correction measures can be taken to optimize the design of the optical system, thereby improving the imaging quality. Wavefront sensors play a vital role in adaptive optical systems. They provide wavefront information for wavefront reconstruction and correction by measuring the phase distortion of the dynamic incident wavefront in real time. Among them, the four-sided pyramid wavefront sensor has the advantages of simple optical path, high sensitivity, and suitability for weak light source detection. It has been widely used in many fields. The four-sided pyramid wavefront sensor uses the principle of two-dimensional knife-edge spectrometry to split the light in two perpendicular directions, dividing the incident light into four beams and mapping them on the CCD. The wavefront slope S of the incident wavefront aberration in the x and y directions is calculated through the light intensity distribution information of the four sub-pupil images. x and S y Finally, the output wavefront slope signal is restored to the phase distribution of the original wavefront aberration through the pattern method wavefront aberration restoration algorithm.

[0003] This wavefront aberration restoration based on the four-sided pyramid wavefront sensor has four main limitations: First, in order to achieve high measurement accuracy without causing excessive light energy loss, the parameters of the single four-sided pyramid prism must meet the following standards: the base angle is about 1-3°, the edge sharpness is within 5μm, and the vertex size is within 20μm. The above high-precision requirements make the actual processing of the four-sided pyramid difficult; Second, when the aberration is too large and exceeds the linear interval of restoration, the output signal and the Zernike coefficient will not be able to meet a good linear relationship; Third, due to the different wavelengths of the incident light, the refraction angle after entering the same prism is also different, so the four-sided pyramid prism also has a chromatic aberration effect problem; Fourth, the current common chromatic aberration solution for the four-sided pyramid prism is to combine materials with different refractive indices and dispersion characteristics to effectively balance the refraction of light of different wavelengths, thereby reducing chromatic aberration. However, this method is only effective within a specific band, and the effect of eliminating chromatic aberration in other bands is relatively limited. Therefore, the above problems lead to a significant increase in the nonlinear error of the pattern method wavefront aberration restoration algorithm when the incident wavefront aberration is large, and when the incident wavefront aberration involves multiple spectra, the chromatic aberration elimination effect of the tetrahedral optical path system is not obvious, the imaging quality on the detection plane is reduced, and the restoration accuracy is further reduced. In summary, there is an urgent need for a wavefront aberration restoration method that can effectively eliminate the chromatic aberration effect under multiple spectra and reduce the nonlinear error when the incident wavefront aberration is large. Summary of the invention

[0004] The objective of the present invention is to provide a multi-spectral wavefront aberration restoration method and system based on a double-roof prism, which overcomes the defects of the above-mentioned existing technologies and improves the accuracy of wavefront aberration restoration.

[0005] The objective of the present invention can be achieved through the following technical solutions:

[0006] A multi-spectral wavefront aberration restoration method based on a double-roof prism includes the following steps:

[0007] Taking the sub-pupil image as the input of the classification model to obtain classification feature data, and performing data expansion on the classification feature data. The sub-pupil image is obtained by inputting the incident wavefront aberration into the simulation optical path of a double-roof prism wavefront sensor. The double-roof prism wavefront sensor includes two groups of double-roof prisms. Each group of double-roof prisms includes two roof prisms placed vertically opposite to each other. The characteristics of the roof prisms are equal. The materials of the two groups of double-roof prisms are a material combination for eliminating chromatic aberration;

[0008] After splicing the sub-pupil image and the classification feature data after data expansion, using them as the input of the regression model to obtain the Zernike coefficient matrix corresponding to the incident wavefront aberration;

[0009] According to the Zernike coefficient matrix corresponding to the incident wavefront aberration and the Zernike polynomial, the restored wavefront aberration is obtained.

[0010] Furthermore, the double-roof prism wavefront sensor includes four effective sub-apertures. The detection area of each effective sub-aperture is M×M pixels. The sub-pupil images within the four groups of detection areas are combined into a 2M×2M pixel image as the sub-pupil image input to the regression model.

[0011] Furthermore, the material combination for eliminating chromatic aberration includes multiple types among H-BaK6 and JGS1, and BK7 and SF11.

[0012] Furthermore, the exit refraction angles of the two groups of double-roof prisms are:

[0013]

[0014] In the formula, θ 7 (λ) is the exit refraction angle, λ is the incident light wavelength, n 1 (λ) is the refractive index of the first group of prisms, n 2 (λ) is the refractive index of the second group of prisms, α 1 is the base angle of the first group of double-roof prisms, α 2 is the base angle of the second group of double-roof prisms, D is the diameter of the incident pupil, and f is the focal length of the optical system.

[0015] Furthermore, the chromatic aberration correction effect of the double-roof prism wavefront sensor matches the ratio of the base angles of the two groups of double-roof prisms, and the ratio of the base angles of the two groups of double-roof prisms shows a constant relationship as follows:

[0016] C = [C 1 C 2 C 3 C 4 …C K

[0017] In the formula, C is the constant relationship, K is the label of different optical band ranges, C 1 C 2 C 3 C 4 …C K is the ratio of the base angles of the two groups of double-roof prisms under different optical band ranges;

[0018] The ratio of the base angles of the two groups of double-roof prisms is:

[0019]

[0020] In the formula, λ 1 is the minimum wavelength of the optical band, λ 2 is the maximum wavelength of the optical band, α 1 is the base angle of the first group of double-roof prisms, α 2 is the base angle of the second group of double-roof prisms, n 1 (λ 1 ) is the refractive index after the minimum wavelength enters the first group of prisms, n 1 (λ 2 ) is the refractive index after the maximum wavelength enters the first group of prisms, n 2 (λ 1 ) is the refractive index after the minimum wavelength enters the second group of prisms, n 2 (λ 2 ) is the refractive index after the maximum wavelength enters the second group of prisms.

[0021] Furthermore, the classification model includes a convolutional layer, a pooling layer, a fully connected layer, a Dropout layer, and a classification layer connected in sequence.

[0022] Furthermore, the regression model includes a feature extraction group and a regression fitting group connected in sequence. The feature extraction group includes a convolutional layer and a pooling layer, and is used to gradually extract the sub-pupil image and the classification feature data after data expansion. The regression fitting group is used to fuse the classification feature data extracted by the feature extraction group.

[0023] Furthermore, the Zernike coefficient matrix is:

[0024] A​K = [a K1 a K2 …a KN

[0025] Wherein, A K is the Zernike coefficient matrix in a specific wavelength band of the spectrum, and a K1 a K2 …a KN are the Zernike coefficients of the 1-Nth order in a specific wavelength band.

[0026] Furthermore, the restored wavefront aberration is:

[0027]

[0028] Wherein, is the restored wavefront aberration, a Ki is the Zernike coefficient of the ith order in a specific wavelength band, ρ represents the distance between the point (x, y) and the origin (0, 0) in polar coordinates, θ represents the counterclockwise angle between ρ and the x-axis, and Z i (ρ, θ) represents the ith order Zernike polynomial, is the radial function, n is the radial order of the Zernike polynomial, m is the angular frequency of the Zernike polynomial, and k is the summation index.

[0029] According to another aspect of the present invention, a multi-spectral wavefront aberration restoration system based on a double-roof prism is provided, including:

[0030] A data expansion module, configured to use the sub-pupil image as the input of a classification model to obtain classification feature data, and perform data expansion on the classification feature data. The sub-pupil image is obtained by inputting the incident wavefront aberration into the simulation optical path of a double-roof prism wavefront sensor. The double-roof prism wavefront sensor includes two groups of double-roof prisms. Each group of double-roof prisms includes two vertically opposite roof prisms. The characteristics of the roof prisms are equal, and the materials of the two groups of double-roof prisms are a material combination for eliminating chromatic aberration;

[0031] A Zernike coefficient matrix acquisition module, configured to splice the sub-pupil image and the data-expanded classification feature data as the input of a regression model to obtain a Zernike coefficient matrix corresponding to the incident wavefront aberration;

[0032] A wavefront aberration restoration module, configured to obtain the restored wavefront aberration according to the Zernike coefficient matrix corresponding to the incident wavefront aberration and the Zernike polynomial.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] ​1. The present invention extracts band features by combining a classification network, enhances the ability of a regression model to utilize effective information in specific bands on the basis of extracting sub-pupil image information, realizes the accurate prediction of multi-spectral band Zernike coefficients by a single regression model, reduces the model complexity, improves the generalization of the restoration algorithm, avoids non-linear errors, and improves the wavefront aberration restoration accuracy.

[0035] 2. The present invention constructs a double-roof prism wavefront sensor through two groups of double-roof prisms, uses a material combination that eliminates chromatic aberration as the material of the two groups of double-roof prisms, converts the wavefront aberration into a sub-pupil image through the simulation optical path of the double-roof prism wavefront sensor, realizes the wavefront segmentation in the orthogonal direction, maintains the optical path symmetry, reduces the aberration, and greatly eliminates the influence of chromatic aberration on the restoration of wavefront aberration. Brief Description of the Drawings

[0036] Figure 1 is a schematic flow chart of a multi-spectral wavefront aberration restoration method based on a double-roof prism proposed by the present invention;

[0037] Figure 2 is a schematic diagram of the simulation optical path of a double-roof prism wavefront sensor;

[0038] Figure 3 is a schematic diagram of the geometric model of two groups of double-roof prisms;

[0039] Figure 4 is a schematic diagram of the training steps of a classification model and a regression model;

[0040] Figure 5 is a schematic diagram of the structures of a classification model and a regression model. Detailed Embodiment

[0041] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives detailed implementation manners and specific operation processes, but the protection scope of the present invention is not limited to the following embodiments.

[0042] Embodiment 1

[0043] This embodiment provides a multi-spectral wavefront aberration restoration method based on a double-roof prism, as Figure 1 shown, including the following steps:

[0044] S1. Take the sub-pupil image as the input of the classification model to obtain classification feature data, and perform data expansion on the classification feature data.

[0045] The sub-pupil image is obtained by inputting the incident wavefront aberration into the simulation optical path of the double-roof prism wavefront sensor. The simulation optical path of the double-roof prism wavefront sensor is as Figure 2As shown, E1 is the incident light plane, (ε,η) is the coordinate system of the incident light plane, (u,v) is the focal plane coordinate system of the double-roof prism, (x,y) is the detection plane coordinate system, f1 is the focal length of lens L1, f2 is the focal length of lens L2, E2 is the focal plane of the double-roof prism, E3 is the front focal plane of lens L2, and I1, I2, I3, and I4 are the light intensities of the four sub-pupil images on the detection plane (x,y).

[0046] The double-roof prism wavefront sensor includes an incident pupil plane, two sets of double-roof prisms, and a detection plane. The geometric models of the two sets of double-roof prisms are as Figure 3 shown. Each set of double-roof prisms includes two roof prisms placed vertically opposite to each other. The characteristics of the roof prisms are equal. The two sets of double-roof prisms are arranged in reverse to achieve wavefront splitting in the orthogonal direction, maintain optical path symmetry, and reduce aberration. The materials of the two sets of double-roof prisms are material combinations for eliminating chromatic aberration. The materials of the two sets of double-roof prisms include various material combinations of H-BaK6 and JGS1, and BK7 and SF11 for eliminating chromatic aberration. In this embodiment, the material of the first set of double-roof prisms is H-BaK6, and the material of the second set of double-roof prisms is JGS1. Among them, the applicable range of H-BaK6 is from 380nm to 2400nm, and the applicable range of JGS1 is from 185nm to 2400nm. The detection plane of the double-roof prism wavefront sensor includes four effective sub-apertures. The detection area of each effective sub-aperture is M×M pixels. The sub-pupil images in the four detection areas are combined into a 2M×2M pixel image as the sub-pupil image input to the regression model. θ 1 to θ 6 are all the angles after the incident light is refracted by the roof prism. θ 7 is the final exit refraction angle. The exit refraction angles of the two sets of double-roof prisms are as follows:

[0047]

[0048] In the formula, θ 7 (λ) is the exit refraction angle, λ is the incident light wavelength, n 1 (λ) is the refractive index of the first set of double-roof prisms, n 2 (λ) is the refractive index of the second set of double-roof prisms, α 1 is the base angle of the first set of double-roof prisms, α 2 is the base angle of the second set of double-roof prisms, D is the diameter of the incident pupil, and f is the focal length of the optical system.

[0049] In this embodiment, the diameter D of the incident pupil is 1.75mm, and the focal length f of the optical system is 150mm. According to the above formula, to ensure that there is no overlap between the sub-pupil images, it is necessary to satisfy: θ 7 (λ) > 0.47°.

[0050] The chromatic aberration correction effect of the double-roof prism wavefront sensor matches the ratio of the base angles of the two sets of double-roof prisms, and the ratio of the base angles of the two sets of double-roof prisms shows a constant relationship as:

[0051] C = [C 1 C 2 C 3 C 4 …C K

[0052] In the formula, C is the constant relationship, K is the label of different optical band ranges, C 1 C 2 C 3 C 4 …C K is the ratio of the base angles of the two sets of double-roof prisms under different optical band ranges;

[0053] The ratio of the base angles of the two sets of double-roof prisms is:

[0054]

[0055] In the formula, λ 1 is the minimum wavelength of the optical band, λ 2 is the maximum wavelength of the optical band, α 1 is the base angle of the first set of double-roof prisms, α 2 is the base angle of the second set of double-roof prisms, n 1 (λ 1 ) is the refractive index after the minimum wavelength enters the first set of prisms, n 1 (λ 2 ) is the refractive index after the maximum wavelength enters the first set of prisms, n 2 (λ 1 ) is the refractive index after the minimum wavelength enters the second set of prisms, n 2 (λ 2 ) is the refractive index after the maximum wavelength enters the second set of prisms.

[0056] When in the visible spectrum: when λ 1 = 500nm, λ 2 = 600nm, When λ 1 = 600nm, λ 2 = 700nm,

[0057] When in the near-infrared spectrum: when λ 1 = 700nm, λ 2 = 800nm, When λ 1 = 800nm, λ 2 = 900nm, When λ​1 = 900 nm, λ 2 = 1000 nm, When λ 1 = 1000 nm, λ 2 = 1100 nm,

[0058] In the mid-infrared spectrum: When λ 1 = 1100 nm, λ 2 = 1700 nm, When λ 1 = 1700 nm, λ 2 = 2200 nm,

[0059] Since it is impossible to calculate each wavelength one by one, the present invention selects the maximum wavelength and the minimum wavelength in a certain wavelength band as extreme cases for analysis. n 1 (λ 1 ) - n 1 (λ 2 ) and n 2 (λ 1 ) - n 2 (λ 2 ) usually change little at different wavelengths. When α 1 , the ratio of α 2 is fixed at a certain constant, the chromatic aberration within a specific wavelength band can be significantly reduced, thereby improving the performance of the optical system, but the chromatic aberration correction effect in other wavelength bands will be weakened. When dividing according to the above wavelength band range, the ratio of α 1 and α 2 is usually between 0.6 and 0.9. Therefore, the multi-spectral range in this embodiment is 500 - 2200 nm.

[0060] The selection of the base angles and materials of the double-roof prisms needs to eliminate the chromatic aberration within a specific wavelength band and also avoid the coincidence of sub-pupil images on the detection plane. In this embodiment, the base angles α 1 of the two groups of double-roof prisms are taken as 15°, and α 2 is taken as 19.89°. After calculation, with this configuration of base angles and prism materials, the chromatic aberration in the wavelength bands of 500 to 600 nm, 800 to 900 nm, and 1700 to 2200 nm can be significantly reduced, but the chromatic aberration elimination effect in other wavelength bands of the above spectra is relatively weak.

[0061] S2. After splicing the sub-pupil image and the classified feature data after data expansion as the input of the regression model, a Zernike coefficient matrix corresponding to the incident wavefront aberration is obtained.

[0062] Randomly generate incident wavefront aberrations with wavelengths respectively in the visible spectrum, near-infrared spectrum, and mid-infrared spectrum. Using the Zernike polynomial function, fit each set of wavefront data to obtain the Zernike coefficient matrix. After inputting the incident wavefront aberration into the double-roof prism wavefront sensor system, sub-pupil images are obtained on the detection plane. In this embodiment, the number of pixels on the detection plane is 1280×1280 pixels. After 4×4 binning operation, the number of pixels on the detection plane is 320×320 pixels, and the size of the four output sub-pupil images is 32×32 pixels. To ensure the integrity of the image information and leave a certain processing margin, the final size of the four intercepted sub-pupil images is 35×35 pixels. After merging the sub-pupil images in the four detection regions, use bicubic interpolation to scale the image and adjust its size to 60×60 pixels to minimize information loss and maintain the details and accuracy of the image.

[0063] Repeat the above steps to generate 10,000 sets of data and assign labels to them according to the wavelength range of the samples.

[0064] In this embodiment, a classification-regression convolutional neural network architecture is adopted. This architecture includes two network models that work together: one for classification tasks and the other for regression tasks. They are trained independently. The training steps of the classification model and the regression model are as Figure 4 shown. Randomly generate six sets of training set samples from the 500 to 2200 nm wavelength band, with two sets of samples each in the visible spectrum, near-infrared spectrum, and mid-infrared spectrum. Therefore, the input data set of the classification model consists of 60×60 pixel sub-pupil images under multiple spectra, and the output data set is divided into six category labels, specifically: 500 to 600 nm (label 1), 600 to 700 nm (label 2) in the visible spectrum; 700 to 900 nm (label 3) and 900 to 1100 nm (label 4) in the near-infrared spectrum; 1100 to 1700 nm (label 5), 1700 to 2200 nm (label 6) in the mid-infrared spectrum.

[0065] The classification model is as Figure 5As shown, it includes three convolutional layers, three pooling layers, two fully connected layers, one Dropout layer, and one classification layer. At the input layer, the sub-pupil image data is input through a single channel with a size of [60, 60, 1]. The first convolutional layer uses a 3x3 convolutional kernel to generate 32 feature maps, and introduces non-linearity through the ReLU activation function. Then, it is downsampled through a 2x2 max pooling layer. The second convolutional layer uses a 3x3 convolutional kernel to generate 64 feature maps, followed by another 2x2 pooling layer for further downsampling. The third convolutional layer uses a 3x3 convolutional kernel to generate 128 feature maps, and increases non-linearity through the ReLU activation function. Finally, it is dimension-reduced through a 2x2 max pooling layer. Then, the flattened features are fused through a fully connected layer with a size of 128 and processed through the ReLU activation function. To prevent overfitting, a 50% Dropout layer is used. The final output layer contains 6 neurons, and the output is converted into a probability distribution of six categories through the Softmax activation function to verify the classification accuracy.

[0066] The performance of the classification model was tested with samples in the visible spectrum from 500 to 600 nm (label 1) and from 600 to 700 nm (label 2); samples in the near-infrared spectrum from 700 to 900 nm (label 3) and from 900 to 1100 nm (label 4); and samples in the mid-infrared spectrum from 1100 to 1700 nm (label 5) and from 1700 to 2200 nm (label 6).

[0067] For the visible spectrum, for samples with label 1, the model successfully classified 1535 samples correctly, and only 21 samples were misclassified; for samples with label 2, the model successfully classified 1441 samples correctly, and only 44 samples were misclassified.

[0068] For the near-infrared spectrum, for samples with label 3, the model successfully classified all samples correctly; for samples with label 4, the model successfully classified 1474 samples correctly, and only 8 samples were misclassified.

[0069] For the mid-infrared spectrum, for samples with label 5, the model successfully classified 1506 samples correctly, and only 2 samples were misclassified; for samples with label 6, the model successfully classified 1470 samples correctly, and only 1 sample was misclassified.

[0070] In a regression task, the goal is to predict continuous values rather than discrete classification labels. Since the classification results after Softmax activation are discrete classification labels (labels 1-6) and cannot directly represent meaningful continuous values, they are not suitable as the input to the regression network directly. Therefore, in the present invention, the MATLAB activations function is used to extract the high-level classification features of the second fully connected layer of the classification model. The high-level classification features are expanded to a size of 60×60×6 through the repmat function and stitched with the corresponding 60×60×1 sub-pupil image to form 60×60×7 multi-channel input data as the input to the final regression model; the output data is a 36th-order Zernike coefficient matrix.

[0071] The Zernike coefficient matrix is as follows:

[0072] A K =[a K1 a K2 …a K36

[0073] In the formula, A K is the Zernike coefficient matrix in a specific wavelength band of the spectrum, and a K1 a K2 …a K36 are the 1st to 36th order Zernike coefficients in a specific wavelength band.

[0074] The regression model is as Figure 5 shown. The input layer consists of the sub-pupil image and the classification feature data after data expansion, and forms [60, 60, 7] multi-channel input data through channel stitching. The regression model includes three groups of convolutional layers and pooling layers, which gradually extract and fuse the image and classification features. Each convolutional layer uses a 3x3x7 convolutional kernel to generate 32, 64, and 128 feature maps respectively. Each layer is followed by a ReLU activation function to introduce non-linearity, and downsampling is performed through a 2x2 max pooling layer to gradually reduce the spatial dimension. The output of the convolutional module is flattened and enters a fully connected layer with 128 neurons for preliminary feature fusion, and then combined with the classification features. The fused information is further processed through another fully connected layer with 128 neurons. Finally, the model outputs the regression value through a fully connected layer with 36 output nodes. The classification features obtained through the classification model directly affect the feature extraction path through stitching at the input stage, and are deeply fused during the convolutional and fully connected processes, providing category information guidance for the regression model and improving the adaptability and prediction accuracy of the regression model for multi-category data.

[0075] S3. Obtain the restored wavefront aberration according to the Zernike coefficient matrix corresponding to the incident wavefront aberration and the Zernike polynomial.

[0076] ​The restored wavefront aberration is:

[0077]

[0078] wherein, is the restored wavefront aberration, a Ki is the i-th order Zernike coefficient under a specific wavelength band, ρ represents the distance between the point (x, y) and the origin (0, 0) in polar coordinates, θ represents the counterclockwise angle between ρ and the x-axis, and Z i (ρ, θ) represents the i-th order Zernike polynomial, is the radial function, n is the radial order of the Zernike polynomial, m is the angular frequency of the Zernike polynomial, and k is the summation index.

[0079] The test regression model is used to restore the accuracy of multi-order large aberrations in the above six cases. A number of test samples are randomly selected, and the sample images and the feature labels generated by the classification model are input into the regression model. After predicting the Zernike coefficients, they are converted into wavefronts, and the residual wavefronts are calculated. The effectiveness of the regression model in the present invention is verified by the restored wavefronts, the residual wavefronts, and the Zernike coefficient errors.

[0080] When the classification output of the test set samples is label 1, the residual wavefront after the regression model restores the multi-order Zernike large aberration with RMS = 1.06μm and PV = 6.03μm is: RMS = 0.0118μm;

[0081] When the label is 2, the residual wavefront after the regression model restores the multi-order Zernike large aberration with RMS = 1.23μm and PV = 6.98μm is: RMS = 0.0206μm;

[0082] When the label is 3, the residual wavefront after the regression model restores the multi-order Zernike large aberration with RMS = 2.88μm and PV = 25.9μm is: RMS = 0.046μm;

[0083] When the label is 4, the residual wavefront after the regression model restores the multi-order Zernike large aberration with RMS = 3.08μm and PV = 23.1μm is: RMS = 0.0639μm;

[0084] When the label is 5, the residual wavefront after the regression model restores the multi-order Zernike large aberration with RMS = 1.01μm and PV = 6.13μm is: RMS = 0.0291μm;

[0085] When the label is 6, the residual wavefront after the regression model restores the multi-order Zernike large aberration with RMS = 2.89μm and PV = 10.7μm is: RMS = 0.0566μm.

[0086] After multiple calculations, the above restoration error is controlled within about 3%. Therefore, the convolutional neural network in the present invention has a relatively stable restoration effect on large aberration under different wavelength bands.

[0087] Embodiment 2

[0088] This embodiment provides a multi-spectral wavefront aberration restoration system based on a double-roof prism, including:

[0089] A data expansion module, which is used to take the sub-pupil image as the input of the classification model to obtain classification feature data, and perform data expansion on the classification feature data. The sub-pupil image is obtained by inputting the incident wavefront aberration into the simulation optical path of a double-roof prism wavefront sensor. The double-roof prism wavefront sensor includes two groups of double-roof prisms. Each group of double-roof prisms includes two roof prisms placed vertically opposite to each other. The characteristics of the roof prisms are equal. The materials of the two groups of double-roof prisms are a material combination for eliminating chromatic aberration;

[0090] A Zernike coefficient matrix acquisition module, which is used to splice the sub-pupil image and the classification feature data after data expansion as the input of the regression model to obtain a Zernike coefficient matrix corresponding to the incident wavefront aberration;

[0091] A wavefront aberration restoration module, which is used to obtain the restored wavefront aberration according to the Zernike coefficient matrix corresponding to the incident wavefront aberration and the Zernike polynomial.

[0092] The rest is the same as in Embodiment 1.

[0093] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the protection scope determined by the claims.

Claims

1. A multi-spectral wavefront aberration restoration method based on double roof prism, characterized in that: The following steps are involved: The sub-pupil image is used as an input of a classification model to obtain classification feature data, and data expansion is performed on the classification feature data, wherein the sub-pupil image is obtained by inputting an incident wavefront aberration into a simulated optical path of a double roof prism wavefront sensor, wherein the double roof prism wavefront sensor includes two groups of double roof prisms, each group of the double roof prisms includes two roof prisms vertically opposite to each other, the properties of the roof prisms are equal, and the materials of the two groups of double roof prisms are a combination of materials that eliminate chromatic aberration; The sub-pupil image and the classification feature data after data expansion are spliced ​​as input of a regression model to obtain a Zernike coefficient matrix corresponding to the incident wavefront aberration; The restored wavefront aberration is obtained according to the Zernike coefficient matrix and Zernike polynomial corresponding to the incident wavefront aberration.

2. The multi-spectral wavefront aberration restoration method based on double roof prism according to claim 1, characterized in that: The double roof prism wavefront sensor includes four effective sub-apertures, each of which has a detection area of ​​M×M pixels. The four groups of sub-pupil images in the detection area are merged into a 2M×2M pixel image as a sub-pupil image input into the regression model.

3. The multi-spectral wavefront aberration restoration method based on double roof prism according to claim 1, characterized in that: The material combination for eliminating color difference includes multiple combinations of H-BaK6 and JGS1 and BK7 and SF11.

4. The multi-spectral wavefront aberration restoration method based on double roof prism according to claim 1, characterized in that: The outgoing refraction angles of the two groups of double roof prisms are: Wherein, θ7(λ) is the exit refraction angle, λ is the wavelength of the incident light, n1(λ) is the refractive index of the first group of prisms, n2(λ) is the refractive index of the second group of prisms, α1 is the base angle of the first group of double roof prisms, α2 is the base angle of the second group of double roof prisms, D is the diameter of the incident light pupil, and f is the focal length of the optical system.

5. The multi-spectral wavefront aberration restoration method based on double roof prism according to claim 1, characterized in that: The chromatic aberration correction effect of the double roof prism wavefront sensor matches the bottom angle ratio of the two groups of double roof prisms. The bottom angle ratio of the two groups of double roof prisms presents a constant relationship: C=[C1C2C3C4…C K ] Where C is a constant relationship, K is a label of different optical band ranges, C1C2C3C4…C K is the ratio of the bottom angles of two sets of double-roof prisms in different optical band ranges; The ratio of the bottom angles of the two sets of double-roof prisms is: Wherein, λ1 is the minimum wavelength of the optical band, λ2 is the maximum wavelength of the optical band, α1 is the base angle of the first group of double roof prisms, α2 is the base angle of the second group of double roof prisms, n1(λ1) is the refractive index after the minimum wavelength enters the first group of prisms, n1(λ2) is the refractive index after the maximum wavelength enters the first group of prisms, n2(λ1) is the refractive index after the minimum wavelength enters the second group of prisms, and n2(λ2) is the refractive index after the maximum wavelength enters the second group of prisms.

6. The multi-spectral wavefront aberration restoration method based on double roof prism according to claim 1, characterized in that: The classification model includes a convolutional layer, a pooling layer, a fully connected layer, a Dropout layer and a classification layer which are connected in sequence.

7. The multi-spectral wavefront aberration restoration method based on double roof prism according to claim 1, characterized in that: The regression model includes a feature extraction group and a regression fitting group connected in sequence, the feature extraction group includes a convolution layer and a pooling layer, which are used to gradually extract the sub-pupil image and the classification feature data after data expansion, and the regression fitting group is used to fuse the classification feature data extracted by the feature extraction group.

8. The multi-spectral wavefront aberration restoration method based on double roof prism according to claim 1, characterized in that: The Zernike coefficient matrix is: A K =[a K1 a K2 …a KN ] In the formula, A K is the Zernike coefficient matrix at a specific band in the spectrum, a K1 a K2 …a KN is the 1-N order Zernike coefficient in a specific band.

9. The multi-spectral wavefront aberration restoration method based on double roof prism according to claim 1, characterized in that: The restored wavefront aberration is: In the formula, To restore the wavefront aberration, a Ki is the i-th order Zernike coefficient in a specific band, ρ represents the distance between the point (x, y) and the origin (0, 0) in polar coordinates, θ represents the angle between ρ and the x-axis in the counterclockwise direction, and Z i (ρ,θ) represents the i-th order Zernike polynomial, is the radial function, n is the radial order of the Zernike polynomial, m is the angular frequency of the Zernike polynomial, and k is the summation index.

10. A multi-spectral wavefront aberration restoration system based on a double roof prism, characterized in that: include: A data expansion module, used for taking a sub-pupil image as an input of a classification model to obtain classification feature data, and performing data expansion on the classification feature data, wherein the sub-pupil image is obtained by inputting an incident wavefront aberration into a simulated optical path of a double roof prism wavefront sensor, wherein the double roof prism wavefront sensor comprises two groups of double roof prisms, each group of the double roof prisms comprises two roof prisms vertically opposite to each other, the properties of the roof prisms are equal, and the materials of the two groups of the double roof prisms are a combination of materials that eliminate chromatic aberration; A Zernike coefficient matrix acquisition module is used to splice the sub-pupil image and the classification feature data after data expansion as input of the regression model to obtain a Zernike coefficient matrix corresponding to the incident wavefront aberration; The wavefront aberration restoration module is used to obtain the restored wavefront aberration according to the Zernike coefficient matrix and Zernike polynomial corresponding to the incident wavefront aberration.

Images