Optimal dislocation angle prediction method for optical system error elements based on neural network

Abstract

The present application relates to the technical field of optical system adjustment, and particularly relates to an optimal misalignment angle prediction method for optical system error elements based on a neural network. The method comprises the following steps: determining the structure parameters of the optical system according to specific requirements; actually building the optical system and performing adjustment; after the adjustment is completed, recording the surface error data of all optical elements, including the angles of the relative coordinate systems of the optical elements; measuring and recording the imaging performance index parameters of the optical system; changing the angles of the relative coordinate systems, repeating the above steps, obtaining the surface error data of the optical elements and the optical system performance index parameter data set, inputting the modular neural network for training, and obtaining a prediction model; according to the requirements of the adjustment task, obtaining the surface error data of the optical elements, inputting the surface error data of the optical elements and the optical system performance index into the prediction model, and obtaining the optimal misalignment angle of the optical elements. The method has the advantages of less time-consuming calculation and reduced on-site adjustment time.

Description

Technical Field

[0001] This invention relates to the field of optical system assembly and adjustment technology, and in particular to a method for predicting the optimal dislocation angle of optical system error elements based on neural networks. Background Technology

[0002] Manufacturing errors are unavoidable in actual optical components. These errors result in non-rotational symmetry, and different placement angles within the system will lead to varying degrees of image quality degradation. Current optical system assembly and adjustment largely rely on the experience of engineers, requiring significant time during on-site commissioning to fine-tune each degree of freedom to achieve optimal system performance. On-site commissioning is not only inefficient but also demands a high level of engineer experience, making it difficult to guarantee the stability and consistency of the commissioning results.

[0003] Numerous scholars have researched assembly and adjustment methods for optical systems. For example, Li Li et al.'s report, "Research on Assembly and Adjustment Methods of Telescope Systems Based on Aberration Correction," and Yang Haijin et al.'s report, "Design and Implementation of Secondary Mirror Assembly and Adjustment in a Hyperboloid Optical System," presented frameworks for aberration-based assembly and adjustment methods. However, most studies focus on determining the allowable range of misalignment in ideal optical systems. When actual components have manufacturing errors, the assembly and adjustment methods still rely on on-site debugging, which consumes a significant amount of time and affects the assembly and adjustment efficiency of optical systems. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a method for predicting the optimal dislocation angle of optical system error elements based on neural networks.

[0005] The purpose of this invention is to provide a method for predicting the optimal dislocation angle of an optical system error element based on a neural network, specifically including the following steps:

[0006] S1. Acquisition of basic system information: Determine the structural parameters of the optical system according to specific requirements;

[0007] S2. Component Information Acquisition: The optical system is actually built and adjusted; after the adjustment is completed, the surface error data of all optical components is recorded; the surface error data includes the angles of each optical component relative to the coordinate system.

[0008] S3. Performance Information Acquisition: Measure and record the imaging performance parameters of the optical system at this time;

[0009] S4. Change the angle of the relative coordinate system and repeat steps S2~S3 to obtain the surface error data of the optical element and the performance index parameter data of the optical system.

[0010] S5. Using the angle data of the relative coordinate system and the surface error data of the optical element as the input dataset, and the optical system performance index parameter data of step S4 as the output dataset, the modular neural network is trained to obtain the optical element dislocation angle prediction network model.

[0011] S6. Set the proportion coefficients of the optical system performance index parameters according to specific engineering requirements, and use a genetic algorithm for global optimization to obtain the optimal combination of dislocation angles when the optical system performance index is optimal;

[0012] S7. According to the requirements of the assembly and adjustment task, obtain the surface error data of the optical components, input the surface error data of the optical components and the performance indicators of the optical system into the optimal dislocation angle prediction network model of the optical components, and obtain the optimal dislocation angle of the optical components.

[0013] Preferably, the optical system structural parameters include the type of optical system, as well as the relative positions, placement angles, element surface shapes, and substrate materials of all optical elements within the optical system.

[0014] Preferably, the angle relative to the 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;

[0015] The surface error data also includes the magnitude, phase, and period of the error.

[0016] Preferably, in step S4, steps S2 to S3 are repeated for at least N groups, where N ≥ 100.

[0017] Preferably, the modular neural network includes a backbone convolutional layer and a dynamic weight layer; the backbone convolutional layer is used to extract features from 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.

[0018] Preferably, the training optimization process in step S5 is as follows:

[0019] S501. Perform weighted retrieval based on the optical system feature vector. The weight index k is determined by the following formula:

[0020]

[0021] Where x is the characteristic vector of the optical system. It is a commonly used function in mathematics and computer science. Its core function is to find the input parameter value that makes a function reach its maximum value, or to return the index of the maximum element in a dataset.

[0022] S502. Input the input dataset into the modular neural network; the backbone convolutional layer extracts the features F of the input data:

[0023]

[0024] Here, F is the input data feature extracted by the backbone convolutional layer, and it is also the input of the subsequent dynamic weight layer; Conv represents the convolution function.

[0025] The feature F output from the backbone convolutional layer is flattened into a vector f, and then dynamically weighted... PT k Mapping to the optical system performance parameter space, calculate the predicted angle of the component relative to the coordinate system; the calculation formula is as follows:

[0026]

[0027] in, For activation function, Bias term, For optical system performance indicators;

[0028] 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, expressed as follows:

[0029] ;

[0030] In the formula, This represents the performance indicators of the actual optical system. This represents the predicted performance index of the optical system.

[0031] Preferably, in 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 shows no improvement for 5 consecutive rounds, the learning rate is multiplied by a decay factor of 0.5, and the minimum learning rate threshold is set to 10. -6 .

[0032] Preferably, in step S6, the expression for the proportion coefficient of the optical system performance index parameter is as follows:

[0033] ;

[0034] Where T is a custom evaluation index for the optical system. arrive The percentage 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 Strell ratio;

[0035] The process of a genetic algorithm includes:

[0036] S601. Population Initialization: Assuming the number of optical elements in the optical system to be predicted is U, the initial population is set as follows: The initial dislocation angle data satisfies: ;

[0037] S602. Population Selection: The population in generation t is denoted as... Parent individuals are selected from the population for reproduction using a roulette wheel selection method or tournament selection method based on fitness ranking.

[0038] S603. Crossover: Performs a crossover operation on the selected parent pairs to generate child pairs.

[0039] ;

[0040] in x (p) and x (q) For the selected parent individual, x (C) For offspring individuals;

[0041] S604. Mutation: Applying a small perturbation to the offspring individuals generates mutated individuals:

[0042] ;

[0043] in, Indicates Gaussian noise; The standard deviation is denoted as .

[0044] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0045] This invention presents a method for determining the optimal dislocation angle of components relative to the coordinate system in a system, enabling a significant amount of debugging work to be completed at the simulation level, thus greatly reducing on-site assembly and debugging time. Although the method of this invention requires a large amount of dataset collection in the early stages, once the neural network computational model is built, predicting the optimal dislocation angle of optical system error components is very fast. Therefore, the neural network-based prediction of the optimal dislocation angle of optical system error components proposed in this invention has a significant advantage in terms of computational time. Attached Figure Description

[0046] Figure 1 This is a flowchart of an optimal dislocation angle prediction method for optical system error elements based on neural networks, provided by an embodiment of the present invention.

[0047] Figure 2 This is a feature map of error information for two optical elements provided according to an embodiment of the present invention.

[0048] Figure 3 This is a flowchart of the modular neural network weight selection process provided according to an embodiment of the present invention.

[0049] Figure 4 This is a flowchart of the genetic algorithm operation provided according to an embodiment of the present invention. Detailed Implementation

[0050] In the following description, embodiments of the invention will be described with reference to the accompanying drawings. In the description below, the same modules are denoted by the same reference numerals. Where the same reference numerals are used, their names and functions are also the same. Therefore, their detailed description will not be repeated.

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the 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 merely illustrative of the invention and do not constitute a limitation thereof.

[0052] This invention provides a neural network-based method for predicting the optimal dislocation angle of error components in optical systems. It aims to provide a method for solving the optimal placement angle of a component based on simulation. Utilizing the relationship between optical system performance parameters and the placement angles of each component, the network is trained based on a dataset of parameters from previous real-world systems. When faced with the task of predicting the placement angles of various optical components in similar systems, it can quickly obtain the optimal placement angles for each component, allowing a significant amount of debugging work to be completed at the simulation level, greatly reducing on-site assembly and adjustment time. Specifically, it includes the following steps:

[0053] S1. Acquisition of basic system information: Determine the structural parameters of the optical system according to specific requirements;

[0054] In a specific embodiment, the optical system structural parameters include the type of optical system (such as off-axis three-lens reflex, Cassegrain type, etc.), as well as the relative positions, placement angles, surface shapes, substrate materials, etc. of all optical elements in the optical system; the above information will provide important reference for subsequent optimal crossover angle prediction, ensuring that the constructed neural network can accurately reflect the actual characteristics of the optical system.

[0055] S2. Component Information Acquisition: The optical system is actually assembled and adjusted; after assembly, the surface error data of all optical components is recorded; the surface error data includes the angle θ of each optical component relative to the coordinate system. ij (1≤i≤m, 1≤j≤n), 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 the two optical elements are shown in the figure;

[0056] Specifically, surface error data also includes the magnitude, phase, and period of the error.

[0057] S3. Performance Information Acquisition: Measure and record the imaging performance parameters of the optical system at this time.

[0058] S4. Change the angle θ relative to the coordinate system. ij Repeat steps S2-S3 at least N times (N≥100) to obtain surface error data of the optical element and performance parameter data of the optical system; in a specific embodiment, the results of index analysis after importing error information from different angles into the optical system are shown in [the figure]. Figure 3 .

[0059] S5. Using the angle data of the relative coordinate system and the surface error data of the optical element as the input dataset, and the optical system performance index parameter data of step S4 as the output dataset, the modular neural network is trained to obtain the optical element dislocation angle prediction network model.

[0060] Modular neural networks are, in essence, modular networks that can select different weights for prediction based on the structural parameters of different optical systems. Specifically, in the training phase, the input and output datasets obtained under different optical system structures are used for training, and then the trained network weights are saved. In the next prediction process, there is no need to adjust the network structure; only the feature vector of the optical system structure needs to be input to achieve modular application for different optical systems. It should be noted that for modular neural networks, the backbone network can be replaced with ResNet, Transformer, or other structures, and the dynamic weight layer can also adopt a fully connected layer or attention mechanism, as long as the core design of "shared feature extraction + dynamic mapping" remains unchanged.

[0061] Modular neural networks consist of backbone convolutional layers and dynamic weight layers; the backbone convolutional layers, also known as shared convolutional layers, are the same across different types of optical systems and are used to extract features F from the input data J.

[0062]

[0063] Where F is the input data feature extracted by the backbone convolutional layer, and it is also the input to the dynamic weight layer; Conv represents the convolution function.

[0064] The dynamic weighting layer dynamically loads different weights based on the optical system category to map features to the optical system performance parameter space. Specifically, the output F of the backbone convolutional layer is flattened into a vector f, and the dynamically loaded weights are then used to map the features to the optical system performance parameter space. PT k Mapped to the optical system performance parameter space:

[0065]

[0066] in, This is the activation function (ReLU function). Bias term, These are predicted values ​​for the optical system performance.

[0067] The training optimization process is as follows:

[0068] S501. Perform weighted retrieval based on the optical system feature vector. The weight index k is determined by the following formula:

[0069]

[0070] Where x is the characteristic vector of the optical system. It is a commonly used function in mathematics and computer science. Its core function is to find the input parameter value that makes a function reach its maximum value, or to return the index of the maximum element in a dataset.

[0071] S502. Input the input dataset into the modular neural network; the backbone convolutional layer extracts the features F of the input data:

[0072]

[0073] Here, F is the input data feature extracted by the backbone convolutional layer, and it is also the input of the subsequent dynamic weight layer; Conv represents the convolution function.

[0074] The feature F output from the backbone convolutional layer is flattened into a vector f, and then dynamically weighted... PT k Mapping to the optical system performance parameter space, calculate the predicted angle of the component relative to the coordinate system; the calculation formula is as follows:

[0075]

[0076] in, This is the activation function (ReLU function). Bias term, For optical system performance indicators;

[0077] 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, expressed as follows:

[0078] ;

[0079] In the formula, This represents the performance indicators of the actual optical system. This represents the predicted performance index of the optical system.

[0080] The Adam algorithm was used as the optimization algorithm, with an initial learning rate of 0.001. When the validation set loss showed no improvement for five consecutive rounds, the learning rate was multiplied by a decay factor of 0.5, and the minimum learning rate threshold was set to 10. -6 In addition, an L2 regularization term (λ=1e−4) was added to the weights to prevent overfitting. The batch size was set to 32, and the number of training iterations was set to 100,000. The weight files trained on different optical systems were numbered and denoted as follows: PT k , Where k is the number of the optical system and W is the number of types of optical systems;

[0081] One-hot encoding is used to transform optical system categories into a computable vector form. For different optical systems, in specific embodiments, taking Cassegrain and off-axis three-lens systems as examples, one-hot encoding is used to process them to obtain optical system category features in vector form, such as [1,0] representing an off-axis three-lens system and [0,1] representing a Cassegrain system. The encoded optical system feature vectors are then matched with the corresponding modular neural network weight files to ensure that the network can distinguish the process characteristics of different systems. The flowchart for selecting modular neural network weights is shown below. Figure 3 As shown.

[0082] Considering that the placement angle of components in different optical systems has different effects on the performance indicators of the optical system, and that the types of commonly used optical systems are limited, this application classifies different optical systems, trains modular neural networks for different categories of optical systems, and dynamically selects or adjusts network weights according to different optical systems to ensure the prediction accuracy of the neural network.

[0083] In this application, deep neural networks (DNNs) can be used as a module in a modular neural network; it should be noted that other types of neural networks can also be used to achieve the same training effect.

[0084] S6. Set the proportion coefficients of the optical system performance parameters according to specific engineering requirements, and use a genetic algorithm (GA) for global optimization to obtain the optimal combination of dislocation angles when the optical system performance is at its best. This combination is the required optimal dislocation angle. See the flowchart below. Figure 4 ;

[0085] Specifically, optical system performance evaluation indicators are set according to specific engineering requirements, and the required performance evaluation indicator percentage parameters are customized.

[0086] For optical systems, performance evaluation metrics include various parameters such as modulation transfer function (MTF), point spread function (PSF), and Strell ratio (SR). The performance metrics required vary depending on the engineering task. Therefore, before predicting the optimal dislocation angle, we first define the required performance evaluation metric weighting parameters, as shown in the following expression:

[0087] ;

[0088] Where T is a custom evaluation index for the optical system. arrive is the percentage 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 Strell ratio.

[0089] After defining the indicators, a genetic algorithm is needed to obtain the optimal combination of dislocation angles that maximizes the performance of the optical system. A genetic algorithm is a global optimization algorithm that simulates natural selection and genetic mechanisms. It is suitable for complex problems that are non-differentiable, multi-peaked, and have multiple constraints. The process includes:

[0090] S601. Population Initialization: Assuming the number of optical elements in the optical system to be predicted is U, the initial population is set as follows: The initial dislocation angle data satisfies: ;where x j This represents the dislocation angle of the j-th element;

[0091] In order for the genetic algorithm to obtain the optimal dislocation angle, the fitness function is defined in the same way as the performance evaluation index T.

[0092] S602. Population Selection: The population in generation t is denoted as... We can use either a roulette wheel selection algorithm or a tournament selection algorithm based on fitness ranking to select parent individuals from the population for reproduction. If we use the roulette wheel selection algorithm, the probability of the i-th individual being selected is:

[0093] ;

[0094] In the formula, For X (i,t) The fitness function, X (i,t) It is the i-th individual in the t-th generation; For X (j,t) The fitness function, X (j,t) It is the j-th individual in the t-th generation;

[0095] S603. Crossover: Performs a crossover operation on the selected parent pairs to generate child pairs.

[0096] ;

[0097] in x (p) and x (q) For the selected parent individual, x (C) For offspring individuals; α represents the coefficient of variation;

[0098] S604. Mutation: Applying a small perturbation to the offspring individuals generates mutated individuals:

[0099] ;

[0100] in, Indicates Gaussian noise; The standard deviation is used to control the amplitude of the disturbance and improve the global search capability.

[0101] The maximum number of iterations for the model is set to 50.

[0102] S7. According to the requirements of the assembly and adjustment task (i.e., when optical components with manufacturing errors are to be installed in a specific system), obtain the surface error data of the optical components, input the surface error data of the optical components and the performance indicators of the optical system into the optimal dislocation angle prediction network model of the optical components, and the optimal dislocation angle of the optical components can be obtained.

[0103] Specifically, when a new task arrives, the feature vector of the acquired optical system type is identified by the database. Then, the modular neural network uses the network weights corresponding to the optical system type as the weights for the current output to make predictions. Although this method requires a large dataset in the early stages, once the neural network computational model is built, predicting the optimal dislocation angle of the optical system error element is very fast. Therefore, the neural network-based prediction of the optimal dislocation angle of the optical system error element proposed in this patent has significant advantages in terms of computational time and ease of understanding.

[0104] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0105] The specific embodiments described above do not constitute a limitation on the scope of protection of this 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 this invention should be included within the scope of protection of this invention.

Claims

1. A method for predicting the optimal dislocation angle of an optical system error element based on a neural network, characterized in that: Specifically, the steps include the following: S1. Acquisition of basic system information: Determine the structural parameters of the optical system according to specific requirements; S2. Component Information Acquisition: The optical system is actually built and adjusted; after the adjustment is completed, the surface error data of all optical components is recorded; the surface error data includes the angles of each optical component relative to the coordinate system. The angle relative to the coordinate system 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; S3. Performance Information Acquisition: Measure and record the imaging performance parameters of the optical system at this time; S4. Change the angle of the relative coordinate system and repeat steps S2~S3 to obtain the surface error data of the optical element and the performance index parameter data of the optical system. S5. Using the angle data of the relative coordinate system and the surface error data of the optical element as the input dataset, and the optical system performance index parameter data of step S4 as the output dataset, the modular neural network is trained to obtain the optical element dislocation angle prediction network model. S6. Set the proportion coefficients of the optical system performance index parameters according to specific engineering requirements, and use a genetic algorithm for global optimization to obtain the optimal combination of dislocation angles when the optical system performance index is optimal; the expressions for the proportion coefficients of the optical system performance index parameters are as follows: ； Where T is a custom evaluation index for the optical system. arrive The percentage 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 Strell ratio; The process of a genetic algorithm includes: S601. Population Initialization: Assuming the number of optical elements in the optical system to be predicted is U, the initial population is set as follows: The initial dislocation angle data satisfies: ; S602. Population Selection: The population in generation t is denoted as... Parent individuals are selected from the population for reproduction using a roulette wheel selection method or tournament selection method based on fitness ranking. S603. Crossover: Performs a crossover operation on the selected parent pairs to generate child pairs. ； in x (p) and x (q) For the selected parent individual, x (C) For offspring individuals; S604. Mutation: Applying a small perturbation to the offspring individuals generates mutated individuals: ； in, Indicates Gaussian noise; Standard deviation; S7. According to the requirements of the assembly and adjustment task, obtain the surface error data of the optical components, input the surface error data of the optical components and the performance indicators of the optical system into the optimal dislocation angle prediction network model of the optical components, and obtain the optimal dislocation angle of the optical components.

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

3. The optimal dislocation angle prediction method for optical system error elements based on neural networks according to claim 1, characterized in that: In step S2, the surface error data also includes the magnitude, phase, and period of the error.

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

5. The optimal dislocation angle prediction method for optical system error elements based on neural networks 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 features from the input data; the dynamic weight layer dynamically loads different weights according to the optical system category, which is used to map the features to the optical system performance parameter space.

6. The optimal dislocation angle prediction method for optical system error elements based on neural networks according to claim 5, characterized in that: The training and optimization process in step S5 is as follows: S501. Perform weighted retrieval based on the optical system feature vector. The weight index k is determined by the following formula: Where x is the characteristic vector of the optical system. It is a commonly used function in mathematics and computer science. Its core function is to find the input parameter value that makes a function reach its maximum value, or to return the index of the maximum element in a dataset. S502. Input the input dataset into the modular neural network; the backbone convolutional layer extracts the features F of the input data: Here, F is the input data feature extracted by the backbone convolutional layer, and it is also the input of the subsequent dynamic weight layer; Conv represents the convolution function. The feature F output from the backbone convolutional layer is flattened into a vector f, and then dynamically weighted... PT k Mapping to the optical system performance parameter space, calculate the predicted angle of the component relative to the coordinate system; the calculation formula is as follows: in, For activation function, Bias term, For optical system performance indicators; 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, expressed as follows: ； In the formula, This represents the performance indicators of the actual optical system. This represents the predicted performance index of the optical system.

7. The optimal dislocation angle prediction method for optical system error elements based on neural networks according to claim 6, characterized in that: In 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 shows no improvement for 5 consecutive rounds, the learning rate is multiplied by a decay factor of 0.5, and the minimum learning rate threshold is set to 10. -6 .

Images