Optimal dislocation angle prediction method of optical system error element based on neural network

Abstract

The invention relates to the technical field of optical system installation and adjustment, in particular to an optimal dislocation angle prediction method of an optical system error element based on a neural network. Comprising the steps of determining structure parameters of an optical system according to specific requirements; actually building an optical system and carrying out adjustment; after the adjustment is completed, recording surface error data of all the optical elements including the angles of the relative coordinate systems of all the optical elements; measuring and recording imaging performance index parameters of the optical system; changing the angle of the relative coordinate system, repeating the steps to obtain the surface error data of the optical element and the performance index parameter data set of the optical system, and inputting the surface error data and the performance index parameter data set into a modular neural network for training to obtain a prediction model; and obtaining surface error data of the optical element according to the requirement of the installation and adjustment task, and inputting the surface error data of the optical element and the performance index of the optical system into the prediction model to obtain the optimal dislocation angle of the optical element. The method has the advantages of low calculation time consumption and reduction of on-site installation and adjustment time.

Description

Technical Field

[0001] The present invention relates to the technical field of optical system alignment, and particularly to an optimal dislocation angle prediction method for error elements of an optical system based on a neural network. Background Art

[0002] Actual optical elements inevitably have manufacturing errors, which will cause the elements to have non-rotational symmetry. Different placement angles in the system will cause different image quality deterioration of the optical system. The existing optical system alignment largely relies on the experience of engineers, and a large amount of time is required to adjust each degree of freedom during the on-site debugging stage to achieve the optimal system performance. The on-site debugging process is not only inefficient, but also has high requirements for engineers' experience, and it is difficult to ensure the stability and consistency of the debugging results.

[0003] At present, many scholars have studied the alignment methods of optical systems. For example, "Research on the Alignment Method of a Telescope System Based on Aberration Correction" reported by Li Li et al. and "Design and Implementation of the Secondary Mirror Alignment in a Hyperbolic Optical System" reported by Yang Haijin et al. gave the framework of the aberration-based alignment method. However, most of the research mainly focuses on the determination of the allowable tolerance range of the misalignment of the ideal optical system. When the actual elements have manufacturing errors, the alignment method still relies on the on-site debugging method, which consumes a lot of time and affects the alignment efficiency of the optical system. Summary of the Invention

[0004] In order to solve the above problems, the present invention provides an optimal dislocation angle prediction method for error elements of an optical system based on a neural network.

[0005] The object of the present invention is to provide an optimal dislocation angle prediction method for error elements of an optical system based on a neural network, which specifically includes the following steps: S1. Obtaining basic system information: Determine the structural parameters of the optical system according to specific requirements; S2. Obtaining element information: Actually build the optical system and perform alignment; after the alignment is completed, record the surface error data of all optical elements; the surface error data includes the angles of each optical element in the relative coordinate system; S3. Obtaining performance information: Measure and record the imaging performance index parameters of the optical system at this time; S4. Change the angle of the relative coordinate system, and repeat steps S2 to S3 to obtain the surface error data of the optical elements and the data of the optical system performance index parameters; S5. Take the angle data of the relative coordinate system and the surface error data of the optical elements as the input data set, and take the optical system performance index parameter data in step S4 as the output data set, and input them into a modular neural network for training to obtain an optical element dislocation angle prediction network model; S6. Set the proportion coefficient of the optical system performance index parameters according to the specific engineering requirements, and use the genetic algorithm for global optimization to obtain the optimal dislocation angle combination when the optical system performance index is the best; S7. According to the requirements of the alignment task, obtain the surface error data of the optical elements, and input the surface error data of the optical elements and the optical system performance index into the optimal dislocation angle prediction network model of the optical elements to obtain the optimal dislocation angle of the optical elements.

[0006] Preferably, the optical system structure parameters include the type of the optical system, as well as the relative positions, placement angles, element surface types, and substrate materials of all optical elements in the optical system.

[0007] Preferably, the angle of the relative coordinate system in step S2 is θ ij , 1 ≤ i ≤ m, 1 ≤ j ≤ n; where m is the number of optical elements in the system and n is the total number of selectable angles; The surface error data also includes the amplitude, phase, and period of the error.

[0008] Preferably, in step S4, steps S2 to S3 are repeated at least N groups, N ≥ 100.

[0009] Preferably, the modular neural network includes a backbone convolutional layer and a dynamic weight layer; the backbone convolutional layer is used to extract the features of the input data; the dynamic weight layer dynamically loads different weights according to the optical system category and is used to map the features to the optical system performance parameter space.

[0010] Preferably, the training optimization process in step S5 is as follows: S501. Perform weight retrieval according to the optical system feature vector, and the weight index k is determined by the following formula:

[0011] where x is the optical system feature vector, is a function commonly used in mathematics and computer science, and its core role is to find the input parameter value that makes a certain function obtain the maximum value, or return the index of the maximum value element in the dataset; S502. Input the input dataset into the modular neural network; the backbone convolutional layer extracts the features F of the input data:

[0012] where F is the input data feature extracted by the backbone convolutional layer and is also the input of the subsequent dynamic weight layer, and Conv represents the convolution function; Flatten the feature F output by the backbone convolutional layer into a vector f, and through the dynamically loaded weight PT kMap to the performance parameter space of the optical system, and calculate the predicted value of the angle of the component relative to the coordinate system; the calculation formula is as follows:

[0013] Among them, is the activation function, bias term, is the performance index of the optical system; S503. The difference between the actual optical system performance index and the optical system performance index obtained by network prediction is used as the loss function, which is expressed as follows: ; In the formula, represents the actual optical system performance index, represents the predicted optical system performance index.

[0014] Preferably, in the training and optimization process of step S5, the adam algorithm is used as the optimization algorithm, and the initial learning rate is set to 0.001; when the loss of the validation set has not improved for 5 consecutive rounds, the learning rate is multiplied by the decay factor 0.5, and the minimum learning rate threshold is set to 10 -6 .

[0015] Preferably, in step S6, the expression formula of the proportion coefficient of the optical system performance index parameters is as follows: ; Among them, T is the custom optical system evaluation index, to is the custom evaluation index proportion coefficient, n is the total number of evaluation indexes, MTF represents the modulation transfer function, PSF represents the point spread function, and SR represents the Strehl ratio; The process of the genetic algorithm includes: S601. Initialize the population: Assume that the number of optical elements in the optical system to be predicted is U, then the initialized population is set to: ; The initialized dislocation angle data satisfies: ; S602. Population selection: The t-th generation population is denoted as , and the roulette wheel method or tournament selection method based on fitness ranking is used to select parental individuals from the population for reproduction; S603. Crossover: Perform crossover operations on the selected parental individual pairs to generate offspring individuals: ; Among them x (p) and x (q) are the selected parental individuals, x(C) is the offspring individual; S604. Mutation: Apply a small perturbation to the offspring individual to generate a mutated individual: ; wherein, represents Gaussian noise; is the standard deviation.

[0016] Compared with the prior art, the present invention can achieve the following beneficial effects: The present invention provides a method for solving the optimal dislocation angle of the relative coordinate system of components in a system, enabling a large amount of debugging work to be completed at the simulation level, significantly reducing the on-site installation and adjustment time. Although the amount of data set collection in the early stage of the method of the present invention is relatively large, once the neural network calculation model is built, predicting the optimal dislocation angle of the error components of the optical system is very fast. Therefore, the prediction of the optimal dislocation angle of the error components of the optical system based on the neural network proposed by the present invention has obvious advantages in terms of calculation time consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flowchart of a method for predicting the optimal dislocation angle of an error component of an optical system based on a neural network according to an embodiment of the present invention.

[0018] Figure 2 is a characteristic diagram of the error information of two optical components according to an embodiment of the present invention.

[0019] Figure 3 is a flowchart of modular neural network weight selection according to an embodiment of the present invention.

[0020] Figure 4 is a flowchart of the genetic algorithm operation according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In the following, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, the same modules are denoted by the same reference numerals. In the case of the same reference numerals, their names and functions are also the same. Therefore, their detailed descriptions will not be repeated.

[0022] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and do not constitute a limitation to the present invention.

[0023] The present invention provides an optimal dislocation angle prediction method for optical system error elements based on a neural network, aiming to give a method for solving the optimal placement angle of an element based on simulation, utilize the relationship between the numerical values of optical system indicators and the placement angles of each element, and train the network based on a data set of actual system parameters in the past; when facing the prediction task of the placement angles of each optical element in the same type of system, it can quickly obtain the optimal placement angles of each element, enabling a large number of debugging tasks to be completed at the simulation level and greatly reducing the on-site installation and adjustment time; specifically including the following steps: S1. Obtaining basic system information: Determine the structural parameters of the optical system according to specific requirements; In a specific embodiment, the structural parameters of the optical system include the type of the optical system (such as off-axis three-mirror type, Cassegrain type, etc.), as well as the relative positions, placement angles, surface profiles, substrate materials, etc. of all optical elements in the optical system; the above information will provide an important reference basis for the subsequent prediction of the optimal stagger angle, ensuring that the constructed neural network can accurately reflect the actual characteristics of the optical system.

[0024] S2. Obtaining element information: Actually build the optical system and perform installation and adjustment; after the installation and adjustment are completed, record the surface error data of all optical elements; the surface error data includes the angle θ ij (1 ≤ i ≤ m, 1 ≤ j ≤ n) of each optical element in the relative coordinate system, where m is the number of optical elements in the optical system and n is the total number of selectable angles; Figure 2 The error information characteristics of two optical elements are shown in Specifically, the surface error data also includes the amplitude, phase, period, etc. of the error.

[0025] S3. Obtaining performance information: Measure and record the imaging performance index parameters of the optical system at this time.

[0026] S4. Change the angle θ ij of the relative coordinate system, and repeat steps S2 to S3 at least N groups (N ≥ 100) to obtain the surface error data of the optical elements and the data of the optical system performance index parameters; in a specific embodiment, the results of index analysis after importing the error information at different angles into the optical system are shown in Figure 3 .

[0027] S5. Use the angle data of the relative coordinate system and the surface error data of the optical elements as the input data set, and use the optical system performance index parameter data in step S4 as the output data set, and input them into a modular neural network for training to obtain an optical element dislocation angle prediction network model; Specifically, a modular neural network is a modular network that can select different weight data for network prediction according to different structural parameters of the optical system. Specifically, in the training part, the input-output data sets obtained under different optical system structures are trained, and then the network weights obtained from the training are saved. In the next prediction process, without adjusting the network structure, only by inputting the feature vector of the optical system structure, the modular application for different optical systems can be realized. It should be noted that for the modular neural network, its backbone network can be replaced by structures such as ResNet and Transformer, and the dynamic weight layer can also adopt a fully connected layer or an attention mechanism, as long as the core design of "shared feature extraction + dynamic mapping" remains unchanged; The modular neural network includes a backbone convolutional layer and a dynamic weight layer; the backbone convolutional layer, which can also be called a shared convolutional layer, is the same for different types of optical systems and is used to extract the feature F of the input data J:

[0028] Among them, F is the input data feature extracted by the backbone convolutional layer and is also the input of the dynamic weight layer; Conv represents the convolution function; The dynamic weight layer dynamically loads different weights according to the optical system category and is used to map the features to the optical system performance parameter space. Specifically, the output F of the backbone convolutional layer is flattened into a vector f, and through the dynamically loaded weights PT k it is mapped to the optical system performance parameter space:

[0029] Among them, is the activation function (ReLU function), the bias term, is the optical system performance prediction value.

[0030] The training optimization process is as follows: S501. Perform weight retrieval according to the optical system feature vector, and the weight index k is determined by the following formula:

[0031] Among them, x is the optical system feature vector, is a function commonly used in mathematics and computer science, and its core role is to find the input parameter value that makes a certain function obtain the maximum value, or to return the index of the maximum value element in the data set; S502. Input the input data set into the modular neural network; the backbone convolutional layer extracts the feature F of the input data:

[0032] Among them, F is the input data feature extracted by the backbone convolutional layer and is also the input of the subsequent dynamic weight layer. Conv represents the convolution function; Flatten the feature F output by the backbone convolutional layer into a vector f, and map it to the optical system performance parameter space through the dynamically loaded weight PT k to calculate the predicted value of the angle of the component relative to the coordinate system. The calculation formula is as follows:

[0033] Among them, is the activation function (ReLU function), the bias term, is the optical system performance index; S503. The difference between the actual optical system performance index and the optical system performance index obtained by network prediction is used as the loss function, which is expressed as follows: ; In the formula, represents the actual optical system performance index, represents the predicted optical system performance index; Use the adam algorithm as the optimization algorithm, and set the initial learning rate to 0.001; when the validation set loss has not improved for 5 consecutive rounds, multiply the learning rate by the decay factor 0.5, and set the minimum learning rate threshold to 10 -6 . In addition, add an L2 regularization term (λ = 1e−4) to the weights to prevent overfitting. Set the batch size to 32 and the number of training iterations to 100,000 times. Number the trained weight files of different optical systems and record them as PT k , ; where k is the number of the optical system and W is the number of types of optical systems; Adopt one-hot encoding to convert the optical system category into a computable vector form; for different optical systems, in a specific embodiment, take the Cassegrain and off-axis three-mirror systems as examples, and use one-hot encoding to process them to obtain the optical system category features in vector form. For example, [1,0] represents the off-axis three-mirror system and [0,1] represents the Cassegrain system. Corresponding the encoded optical system feature vector with the corresponding modular neural network weight file to ensure that the network can distinguish the process characteristics of different systems. The flowchart for selecting the modular neural network weights is shown in Figure 3 as shown.

[0034] Considering that the influence of the element arrangement angles of different optical systems on the performance indicators of the optical system varies, and at the same time, the types of common optical systems are limited. Therefore, in this application, different optical systems are classified, and modular neural networks are trained for different types of optical systems. The network weights are dynamically selected or adjusted according to different optical systems to ensure the prediction accuracy of the neural network.

[0035] In this application, a deep neural network (DNN) can be used as a module in the modular neural network. It should be noted that other types of neural networks can also be used to achieve the same training effect.

[0036] S6. Set the proportion coefficient of the optical system performance index parameters according to the specific engineering requirements, and use the genetic algorithm (GA) for global optimization to obtain the best dislocation angle combination when the optical system performance index is optimal. At this time, it is the required best dislocation angle. The flowchart is shown in Figure 4 ; Specifically, set the optical system performance evaluation index according to the specific engineering requirements, and customize the proportion parameters of the required performance evaluation index. For an optical system, the performance evaluation indexes include modulation transfer function (MTF), point spread function (PSF), Strehl ratio (SR), etc. For different engineering tasks, the performance indexes that need to be focused on are also different. Therefore, before predicting the best dislocation angle, first customize the proportion parameters of the required performance evaluation index. The expression is as follows: ; Among them, T is the customized optical system evaluation index, to are the customized proportion coefficients of the evaluation indexes, n is the total number of evaluation indexes, MTF represents the modulation transfer function, PSF represents the point spread function, and SR represents the Strehl ratio.

[0037] After completing the index definition, it is necessary to use the genetic algorithm to obtain the best dislocation angle combination that makes the optical system performance optimal. The genetic algorithm is a global optimization algorithm that simulates natural selection and genetic mechanisms and is suitable for complex problems that are non-differentiable, multi-modal, and multi-constrained. The process includes: S601. Initialize the population: Assume that the number of optical elements in the optical system to be predicted is U, then the initialized population is set as: ; The initialized dislocation angle data satisfies: ; Among them, x j represents the dislocation angle of the jth element; In order to enable the genetic algorithm to obtain the optimal dislocation angle, the fitness function is defined to be the same as the performance evaluation index T; S602. Population Selection: Denote the population in the t-th generation as , and use the roulette wheel method or tournament selection method based on fitness ranking to select parental individuals from the population for reproduction; if the roulette wheel method is used, the probability that the i-th individual is selected is: ; In the formula, is the fitness function of X (i,t) , and X (i,t) is the i-th individual in the t-th generation; is the fitness function of X (j,t) , and X (j,t) is the j-th individual in the t-th generation; S603. Crossover: Perform a crossover operation on the selected parental individual pairs to generate offspring individuals: ; where x (p) and x (q) are the selected parental individuals, x (C) is the offspring individual; α represents the mutation coefficient; S604. Mutation: Apply a small perturbation to the offspring individuals to generate mutated individuals: ; where represents Gaussian noise; is the standard deviation, which is used to control the perturbation amplitude and enhance the global search ability; Set the maximum number of iterations of the model to 50 generations.

[0038] S7. According to the requirements of the alignment task (i.e., when the optical element with manufacturing errors is to be installed in a specific system), obtain the surface error data of the optical element, and input the surface error data of the optical element and the optical system performance indicators into the optimal dislocation angle prediction network model of the optical element, then the optimal dislocation angle of the optical element can be obtained.

[0039] Specifically, when a new task arrives, the feature vector of the optical system type obtained is recognized by the database, and then the modular neural network will use the network weights corresponding to the optical system type as the weights for this output to make predictions. Although the amount of data collection in the early stage of this method is large, once the neural network calculation model is built, predicting the optimal dislocation angle of the optical system error element is very fast. Therefore, the prediction of the optimal dislocation angle of the optical system error element based on the neural network proposed in this patent has obvious advantages in terms of calculation time consumption and understandability.

[0040] It should be understood that the various forms of processes shown above can be used, with steps reordered, added or deleted. For example, the steps described in the disclosure of the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and no limitation is imposed herein.

[0041] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub - combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An optimal dislocation angle prediction method for an optical system error element based on a neural network, characterized in that: Specifically, it includes the following steps: S1. Obtaining basic system information: Determine the structural parameters of the optical system according to specific requirements; S2. Obtaining component information: Actually build the optical system and perform alignment and adjustment; after the alignment and adjustment are completed, record the surface error data of all optical components; the surface error data includes the angles of each optical component in the relative coordinate system; S3. Obtaining performance information: Measure and record the imaging performance index parameters of the optical system at this time; S4. Change the angle of the relative coordinate system, and repeat steps S2 to S3 to obtain the surface error data of the optical components and the optical system performance index parameter data; S5. Use the angle data of the relative coordinate system and the surface error data of the optical components as the input data set, and use the optical system performance index parameter data in step S4 as the output data set, and input it into the modular neural network for training to obtain the optical component dislocation angle prediction network model; S6. Set the proportion coefficient of the optical system performance index parameters according to specific engineering requirements, and use the genetic algorithm for global optimization to obtain the best dislocation angle combination when the optical system performance index is the best; S7. According to the requirements of the alignment and adjustment task, obtain the surface error data of the optical components, and input the surface error data of the optical components and the optical system performance index into the optical component optimal dislocation angle prediction network model to obtain the optimal dislocation angle of the optical components.

2. The optimal dislocation angle prediction method for an optical system error element based on a neural network according to claim 1, characterized in that: The structural parameters of the optical system include the type of the optical system, as well as the relative positions, placement angles, element surface types, and substrate materials of all optical components in the optical system.

3. The optimal dislocation angle prediction method for an optical system error element based on a neural network according to claim 1, characterized in that: The angle of the relative coordinate system in the step S2 is θ ij , where 1 ≤ i ≤ m and 1 ≤ j ≤ n; Where m is the number of optical components in the system, and n is the total number of selectable angles; The surface error data also includes the amplitude, phase, and period of the error.

4. The optimal dislocation angle prediction method for an optical system error element based on a neural network according to claim 1, characterized in that: In step S4, steps S2 to S3 are repeated at least N groups, N≥100.

5. The optimal dislocation angle prediction method for an optical system error element based on a neural network according to claim 1, characterized in that: The modular neural network includes a backbone convolutional layer and a dynamic weight layer; the backbone convolutional layer is used to extract the features of the input data; the dynamic weight layer dynamically loads different weights according to the optical system category and is used to map the features to the optical system performance parameter space.

6. The optimal dislocation angle prediction method for an optical system error element based on a neural network according to claim 5, characterized in that: The training and optimization process in step S5 is as follows: S501. Perform weight retrieval according to the optical system feature vector, and the weight index k is determined by the following formula: where x is the optical system feature vector, is a function commonly used in mathematics and computer science. Its core function is to find the input parameter value that maximizes a certain function, or to return the index of the maximum value element in a dataset; S502. Input the input data set into the modular neural network; the backbone convolutional layer extracts the features F of the input data: Where F is the input data feature extracted by the backbone convolutional layer and is also the input of the subsequent dynamic weight layer, and Conv represents the convolution function; Flatten the feature F output by the backbone convolutional layer into a vector f, and map it to the optical system performance parameter space through dynamically loaded weights PT k Calculate the predicted value of the angle of the component relative to the coordinate system; the calculation formula is as follows: Among them, is the activation function, is the bias term, is the optical system performance index; S503. The difference between the actual optical system performance index and the optical system performance index predicted by the network is used as the loss function, which is expressed as follows: ； In the formula, represents the performance index of the actual optical system, represents the predicted performance index of the optical system.

7. The optimal dislocation angle prediction method for an optical system error element based on a neural network according to claim 6, characterized in that: During the training optimization process of step S5, the adam algorithm is used as the optimization algorithm, and the initial learning rate is set to 0.001; when the validation set loss has not improved for 5 consecutive rounds, the learning rate is multiplied by the decay factor 0.5, and the minimum learning rate threshold is set to 10 -6 .

8. The optimal dislocation angle prediction method for an optical system error element based on a neural network according to claim 1, characterized in that: In step S6, the expression of the proportion coefficient of the optical system performance index parameters is as follows: ； where T is a custom evaluation index for the optical system, to is the proportion coefficient of the custom evaluation index, n is the total number of evaluation indexes, MTF represents the modulation transfer function, PSF represents the point spread function, and SR represents the Strehl ratio; The process of the genetic algorithm includes: S601. Initialize the population: Assume that the number of optical elements in the optical system to be predicted is U, then the initialized population is set as: ; The initialized dislocation angle data satisfies: ; S602. Population selection: Denote the population in the t-th generation as , and use the roulette wheel method or tournament selection method based on fitness ranking to select parental individuals from the population for reproduction; S603. Crossover: Perform crossover operations on the selected parent individuals to generate offspring individuals: ； Among them x (p) and x (q) are the selected parent individuals, x (C) are the offspring individuals; S604. Mutation: Apply a small perturbation to the offspring individuals to generate mutated individuals: ； Among them, represents Gaussian noise; is the standard deviation.

Images