Light-weight structure design method and device of optical lens

Abstract

The invention belongs to the technical field of lenses, and discloses a lightweight structure design method and device of an optical lens. The method comprises the following steps: acquiring lens customization demand data, wherein the lens customization demand data comprises lens optical performance demand data, lens physical performance demand data and customer demand budget; processing the lens customization demand data by using a feature extraction method to obtain a customization demand structural body set; constructing a lens customization demand data tree according to the customization demand structural body set; obtaining a lens customization material set according to the lens customization demand data tree; according to the lens customization material set and the lens customization demand data tree, carrying out lightweight adjustment and optimization on the lens; according to the scheme, through intelligent lens demand analysis, optimization modeling and self-adaptive adjustment and optimization, lightweight adjustment and optimization of the lens are realized while individual demands of customers are met, and the flexibility and accuracy of lightweight design of the lens are improved.

Description

Technical Field

[0001] The present invention relates to the field of lens technology, and more specifically, to a lightweight structural design method and device for an optical lens. Background Art

[0002] The current design of lightweight lens structures usually relies on experience-driven material selection and structural optimization methods, lacking a systematic analysis of customer customization requirements, which makes it difficult for material selection solutions to strike a balance between optical performance, mechanical strength, and manufacturing costs. In addition, existing lens design methods often only take all requirements into consideration, ignoring the priority and hierarchy of lens requirements, resulting in insufficient flexibility in lens material selection, making it difficult to achieve precise optimization of lens materials while meeting the key optical properties of the lens, thus affecting the overall performance and production feasibility of the lens. At the same time, traditional optimization algorithms fail to fully consider the multi-level demand relationships of lenses during the lens lightweighting process, and are unable to achieve adaptive adjustment of the amount of lens material used, making it difficult for lens design to balance lightweighting with multi-dimensional performance requirements such as optics and mechanics.

[0003] Therefore, how to provide an intelligent and precise lens material selection and lens lightweight structure optimization method according to the customized needs of lenses, ensure that the lens can achieve the maximum lightweight while meeting the optical, mechanical and other performance requirements, and improve the utilization rate of lens materials and reduce the production cost of lenses, has become a technical problem that needs to be urgently solved in the optical industry. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a lightweight structure design method for an optical lens, comprising:

[0005] Acquire lens customization demand data, wherein the lens customization demand data includes lens optical performance demand data, lens physical performance demand data and customer demand budget;

[0006] Using feature extraction methods to process lens customization demand data, obtaining a customized demand structure set;

[0007] Construct a lens customization requirement data tree according to the customization requirement structure set;

[0008] Obtain a lens customized material set according to the lens customized demand data tree;

[0009] The lens is lightweight and optimized based on the lens customized material set and lens customized demand data tree.

[0010] Furthermore, the method for lightweight tuning of the lens according to the lens customized material set and the lens customized demand data tree includes:

[0011] Step500: Denote the number of lens customization materials in the lens customization material set as DZSL. Input the lens customization requirement data tree and DZSL lens customization materials into the material usage model respectively to obtain the lens material usage intervals corresponding to the DZSL lens customization materials. Initialize a lens material population containing H lens material candidate solutions. Let the maximum number of iterations be ZDCS, let zdcs = 1, and the value range of zdcs is from 1 to ZDCS.

[0012] Step501: Evaluate the lens material fitness values corresponding to the H lens material candidate solutions in the lens material population.

[0013] Step502: Sort the H lens material fitness values in descending order, and select the first G lens material fitness values to construct an elite lens material population.

[0014] Step503: Perform a mutation operation on the elite lens material population according to a preset method to obtain F mutant lens material candidate solutions and add them to the elite lens material population. Overwrite the elite lens material population to the lens material population, and let H = G + F.

[0015] Step504: Let zdcs = zdcs + 1. If zdcs is less than or equal to ZDCS, continue to execute Step501 to Step503; if zdcs is greater than ZDCS, select the lens material candidate solution with the largest lens material fitness value from the lens material population as the optimal lens material lightweight solution.

[0016] Furthermore, the method of performing a mutation operation on the elite lens material population according to a preset method to obtain F mutant lens material candidate solutions includes:

[0017] Preset a first fitness threshold and a second fitness threshold, where the first fitness threshold is less than the second fitness threshold. Divide the elite lens material population according to the first fitness threshold and the second fitness threshold to obtain a high material fitness set, a medium material fitness set, and a low material fitness set.

[0018] Perform random cross - overs within the high material fitness set to obtain lens material candidate solutions and add them to the high material fitness set.

[0019] Perform random cross - overs between the lens material candidate solutions in the medium material fitness set and the lens material candidate solutions in the high material fitness set to obtain lens material candidate solutions and add them to the medium material fitness set. Perform random cross - overs within the medium material fitness set to obtain Select several lens material candidates and add them to the fitness set of materials;

[0020] Randomly cross the lens material candidates in the low-fitness set of materials with those in the high-fitness set of materials to obtain several lens material candidates and add them to the low-fitness set of materials; perform random crossover in the low-fitness set of materials to obtain several lens material candidates and add them to the low-fitness set of materials;

[0021] When is reached, stop the random crossover operation to obtain F mutant lens material candidates.

[0022] Furthermore, the method for constructing the lens customization requirement data tree includes:

[0023] Step200: Preset the height of the lens customization requirement data tree as , is an integer greater than 0, preset customization requirement priority thresholds, and the customization requirement priority thresholds are , ; Let sg = 1, and the value range of sg is from 1 to SG;

[0024] Step201: If sg is equal to 1, then construct the customization requirement structures in the customization requirement structure set whose customization requirement priority values are greater than or equal to into the sg-th layer of the lens customization requirement data tree;

[0025] If sg is greater than 1 and less than or equal to , then construct the customization requirement structures in the customization requirement structure set whose customization requirement priority values are greater than or equal to and less than into the sg-th layer of the lens customization requirement data tree;

[0026] If sg is equal to , then construct the customization requirement structures in the customization requirement structure set whose customization requirement priority values are less than into the sg-th layer of the lens customization requirement data tree;

[0027] Step202: Let sg = sg + 1. If sg is less than or equal to SG, then continue to execute Step201; if sg is greater than SG, then obtain the lens customization requirement data tree and end the current process.

[0028] Furthermore, the method for obtaining the customization requirement structure set includes:

[0029] Step100: Denote the number of customization requirements in the lens customization requirement data as DZ, let dz = 1, and the value range of dz is from 1 to DZ; construct a lens customization requirement set with DZ customization requirements.

[0030] Step101: Input the dz-th customization requirement in the lens customization requirement set into the requirement priority value model to obtain the corresponding customization requirement priority value.

[0031] Step102: Construct a corresponding customization requirement structure with the dz-th customization requirement and the customization requirement priority value.

[0032] Step103: Add the dz-th customization requirement structure to the customization requirement structure set.

[0033] Step104: Let dz = dz + 1. If dz is less than or equal to DZ, continue to execute Step101 to Step103. If dz is greater than DZ, obtain the customization requirement structure set and end the current process.

[0034] Furthermore, the method for obtaining the lens customization material set according to the lens customization requirement data tree includes:

[0035] Step300: The number of material selection layers of the lens customization requirement data tree is CLS, and the value range of CLS is from 1 to , let cls = 1, and the value range of cls is from 1 to CLS; construct CLS requirement structure temporary storage sets.

[0036] Step301: Add customization requirement structures in the lens customization requirement data tree to the cls-th requirement structure temporary storage set, is the number of customization requirement structures for the cls-th layer of customization requirements; if cls is greater than 1, add the customization requirement structures in the (cls - 1)-th requirement structure temporary storage set to the cls-th requirement structure temporary storage set.

[0037] Step302: Input the cls-th requirement structure temporary storage set into the lens material model to obtain the cls-th lens applicable material set.

[0038] Step303: Let cls = cls + 1. If cls is less than or equal to CLS, continue to execute Step301 to Step302; if cls is greater than CLS, obtain CLS lens applicable material sets, and analyze and process the CLS lens applicable material sets according to a preset method to obtain the lens customization material set.

[0039] Further, a method for analyzing and processing a set of CLS lens applicable materials according to a preset method to obtain a lens customized material set includes:

[0040] Step400: Take the intersection of the CLS temporary storage sets of requirement structures to obtain the temporary storage intersection of requirement structures; let cls = 1;

[0041] Step401: Obtain the cls-th temporary storage set of requirement structures; if the cls-th temporary storage set of requirement structures does not contain the customized requirement structures in the temporary storage intersection of requirement structures, then use the cls-th temporary storage set of requirement structures as the cls-th requirement structure de-duplication set and directly execute Step402;

[0042] If the cls-th temporary storage set of requirement structures contains the customized requirement structures in the temporary storage intersection of requirement structures, then delete the customized requirement structures in the temporary storage intersection of requirement structures from the cls-th temporary storage set of requirement structures to obtain the cls-th requirement structure de-duplication set and execute Step402;

[0043] Step402: Let cls = cls + 1. If cls is less than or equal to CLS, then continue to execute Step401; if cls is greater than CLS, then execute Step403;

[0044] Step403: Construct the CLS requirement structure de-duplication sets into a total requirement structure set, count the number of each customized requirement structure in the total requirement structure set, sort the total requirement structure set in descending order according to the number of customized requirement structures, and select the top N customized requirement structures from the sorted total requirement structure set in descending order; and construct the lens customized material set from the top N customized requirement structures and the temporary storage intersection of requirement structures.

[0045] Further, the training method of the material usage model includes:

[0046] Pre-collect a material usage data set, where the material usage data set includes Y groups of material usage data and the lens material usage intervals corresponding to the Y groups of material usage data. Y is a positive integer greater than 0. The material usage data includes lens customized requirement data trees and lens customized materials; divide the material usage data set into a training set and a validation set, where the training set is used to train the material usage model and the validation set is used to evaluate the generalization performance of the material usage model;

[0047] During the training process of the material usage model, minimizing the cross-entropy loss function is used as the optimization objective. The early stopping strategy is used to monitor the performance of the validation set. By continuously adjusting the network parameters, the model performance is optimized. When the prediction accuracy on the validation set reaches the preset accuracy, the training is stopped. The material usage model is trained using a deep neural network based on a multi-layer perceptron.

[0048] The material usage data is converted into a high-dimensional feature vector. The input layer of the material usage model receives the high-dimensional feature vector, extracts the non-linear relationship in the material usage data through the hidden layer. Finally, the output layer of the material usage model calculates the probability distribution of the lens material usage interval through the softmax activation function, and outputs the lens material usage interval corresponding to the maximum probability as the final prediction result.

[0049] The lens material model is obtained by the same training method as the material usage model.

[0050] Furthermore, the training method of the demand priority value model includes:

[0051] Pre-collect a demand priority value data set, which includes Q groups of demand priority value data and the customized demand priority values corresponding to the Q groups of demand priority value data. Q is a positive integer greater than 0. The demand priority value data includes customized demands. The demand priority value data set is divided into a training set and a validation set. The training set is used to learn the demand priority value model parameters, and the validation set is used to evaluate the generalization ability of the demand priority value model to avoid overfitting.

[0052] During the training process of the demand priority value model, combined with the dynamic learning rate adjustment strategy and introducing the early stopping mechanism, with minimizing the cross-entropy loss function as the optimization objective, when the performance of the validation set meets the preset requirements, the training is automatically stopped to ensure the convergence effect of the demand priority value model. The demand priority value model is implemented based on a neural network model architecture. The neural network model consists of an input layer, a hidden layer, and an output layer. The input layer receives the demand priority value data and converts it into a high-dimensional feature vector. The hidden layer uses the activation function to extract the non-linear pattern of the demand priority value data. The output layer calculates the probability distribution of the customized demand priority value through the softmax activation function, and outputs the customized demand priority value corresponding to the maximum probability as the prediction result.

[0053] The lightweight structure design device of the optical lens implements the lightweight structure design method of the optical lens, including:

[0054] A demand acquisition module, used to obtain lens customization demand data, where the lens customization demand data includes lens optical performance demand data, lens physical performance demand data, and customer demand budget.

[0055] The first processing module is used to process the lens customization requirement data by using a feature extraction method to obtain a set of customization requirement structures;

[0056] The second processing module is used to construct a lens customization requirement data tree according to the set of customization requirement structures;

[0057] The material selection module is used to obtain a set of lens customization materials according to the lens customization requirement data tree;

[0058] The lens optimization module is used to perform lightweight optimization on the lens according to the set of lens customization materials and the lens customization requirement data tree.

[0059] Compared with the prior art, the technical effects and advantages of the lightweight structure design method and device of the optical lens of the present invention are as follows:

[0060] Through the accurate collection and progressive processing of lens customization requirements, this solution constructs a lens customization requirement data tree. Combining the lens customization requirement data tree with a machine learning model, a set of applicable lens materials is obtained, thus realizing the intelligence, accuracy, and efficiency of lens material selection, and greatly improving the flexibility and stability of the lightweight structure design of the lens. By adopting the method of accumulating requirements layer by layer, while meeting the high-priority requirements, the secondary requirements are also taken into account, avoiding the coupling problem caused by one-time input, and improving the accuracy and interpretability of material selection.

[0061] Combining the lens customization requirement data tree with specific lens customization materials, a lens material usage range of the lens customization materials is obtained. On this basis, the elitist retention and mutation strategies of the genetic algorithm are improved to realize the customized dynamic optimization adjustment of the lens material candidate solutions, thereby realizing the adaptive adjustment of the lens material usage, and further realizing the optimal lightweight solution, that is, ensuring the stability of the key performances such as the optics and mechanics of the lens, and also meeting the customer's customization requirements.

[0062] In summary, through intelligent requirement analysis, optimization modeling, and adaptive optimization, this solution realizes the lightweight optimization of the lens while meeting the personalized requirements of customers, improves the flexibility, accuracy, and stability of the lightweight design of the lens, increases the material utilization rate, reduces the production cost, and has broad application prospects and significant economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a structural diagram of the lightweight structure design device of the optical lens according to Embodiment 1 of the present invention;

[0064] Figure 2 It is a flowchart of the lightweight structure design method of the optical lens according to Embodiment 3 of the present invention;

[0065] Figure 3 Structural design device diagram of the lightweight optical lens according to Embodiment 2 of the present invention;

[0066] Figure 4 Method flowchart for lightweight optimization of a lens based on a lens customization material set and a lens customization requirement data tree;

[0067] Figure 5 Schematic diagram of a lens customization requirement data tree. Detailed implementation manners

[0068] Next, the technical solutions in the embodiments of the present invention will be described in detail, clearly, and completely with reference to the accompanying drawings in the embodiments of the present invention. It should be noted in particular that the specific embodiments described below are only used to better illustrate and explain the technical solutions of the present invention, aiming to enable those skilled in the art to better understand and implement the present invention, and should not be construed as limiting the protection scope of the present invention. Without departing from the spirit and essence of the present invention, those skilled in the art can modify, adjust, or make equivalent replacements according to the content disclosed in the present invention, and these should all be regarded as the protection scope of the present invention.

[0069] Embodiment 1

[0070] Please refer to Figure 1 As shown, the lightweight structural design device of the optical lens in this embodiment includes a requirement acquisition module, a first processing module, a second processing module, a material selection module, and a lens optimization module. Each module is connected by wire and / or wirelessly to achieve data transmission.

[0071] The requirement acquisition module is used to obtain lens customization requirement data, and the lens customization requirement data includes lens optical performance requirement data, lens physical performance requirement data, and customer demand budget; the lens optical performance requirement data includes lens refractive index, lens transmittance, lens chromatic aberration, lens aberration, and lens focal length, and the lens physical performance requirement data includes lens weight, lens thickness, lens strength, lens durability, and lens temperature stability; the lens refractive index, lens transmittance, lens chromatic aberration, lens aberration, lens focal length, lens weight, lens thickness, lens strength, lens durability, lens temperature stability, and customer demand budget are all customization requirements.

[0072] It should be noted that aberration is an important concept in the field of lenses. When light passes through a lens, due to the geometric shape of the lens, material properties, and the complexity of the light propagation path, the light cannot be perfectly focused on a single point, resulting in problems such as blurred imaging, distortion, and chromatic aberration. These phenomena are collectively referred to as aberration. The description of lens aberration depends on specific quantization indices, such as wavefront error (RMS, PV), modulation transfer function (MTF), and Strehl Ratio, etc. Through these quantization indices, the impact of aberration on imaging quality can be clearly measured, thereby meeting the customer's requirements for clarity and precision.

[0073] The first processing module is used to process the lens customization requirement data using a feature extraction method to obtain a set of customization requirement structures.

[0074] The method for obtaining the set of customization requirement structures includes:

[0075] Step100: Denote the number of customization requirements in the lens customization requirement data as DZ, preset the initial value of dz as 1, and the value range of dz is from 1 to DZ; construct the DZ customization requirements into a lens customization requirement set;

[0076] Step101: Obtain the dz-th customization requirement from the lens customization requirement set, input the dz-th customization requirement into the requirement priority value model, and obtain the customization requirement priority value corresponding to the dz-th customization requirement;

[0077] Step102: Construct the dz-th customization requirement structure corresponding to the dz-th customization requirement by combining the dz-th customization requirement and the customization requirement priority value;

[0078] Step103: Add the dz-th customization requirement structure to the set of customization requirement structures;

[0079] Step104: Let dz = dz + 1. If dz is less than or equal to DZ, continue to execute Step101 to Step103. If dz is greater than DZ, obtain the set of customization requirement structures and end the current process.

[0080] The training method of the requirement priority value model includes:

[0081] Pre-collect a dataset of requirement priority values, where the dataset of requirement priority values includes Q groups of requirement priority value data and the customized requirement priority values corresponding to the Q groups of requirement priority value data. Q is a positive integer greater than 0, and the requirement priority value data includes customized requirements; divide the dataset of requirement priority values into a training set and a validation set. The training set is used to learn the model parameters of the requirement priority values, and the validation set is used to evaluate the generalization ability of the requirement priority value model to avoid overfitting;

[0082] During the training process of the requirement priority value model, combine a dynamic learning rate adjustment strategy and introduce an early stopping mechanism. With minimizing the cross-entropy loss function as the optimization goal, when the performance of the validation set meets the preset requirements, automatically stop the training to ensure the convergence effect of the requirement priority value model; the requirement priority value model is implemented based on a neural network model architecture; the neural network model consists of an input layer, a hidden layer, and an output layer. The input layer receives the requirement priority value data and converts it into a high-dimensional feature vector. The hidden layer uses an activation function to extract the complex non-linear patterns of the requirement priority value data. The output layer calculates the probability distribution of the customized requirement priority values through the softmax activation function, and takes the customized requirement priority value corresponding to the maximum probability as the prediction result and outputs it.

[0083] A second processing module for constructing a lens customized requirement data tree according to the set of customized requirement structures.

[0084] The method for constructing the lens customized requirement data tree includes:

[0085] Step200: Preset the height of the lens customized requirement data tree to be , is an integer greater than 0, and preset customized requirement priority thresholds, and the customized requirement priority thresholds are , ; preset the initial value of sg to 1, and the value range of sg is from 1 to SG;

[0086] Step201: If sg is equal to 1, then obtain the customized requirement structures from the set of customized requirement structures whose customized requirement priority values are greater than or equal to , and construct the customized requirement structures whose customized requirement priority values are greater than or equal to into the sg-th layer of the lens customized requirement data tree;

[0087] If sg is greater than 1 and sg is less than or equal to , then obtain the customized requirement structures from the set of customized requirement structures whose customized requirement priority values are greater than or equal to and less than The customized requirement structure with a customized requirement priority value greater than or equal to and less than is constructed into the sg-th layer of the lens customization requirement data tree;

[0088] If sg is equal to , obtain the customized requirement structure with a customized requirement priority value less than from the set of customized requirement structures, and construct the customized requirement structure with a customized requirement priority value less than into the sg-th layer of the lens customization requirement data tree;

[0089] Step202: Let sg = sg + 1. If sg is less than or equal to SG, continue to execute Step201; if sg is greater than SG, obtain the lens customization requirement data tree and end the current process.

[0090] It should be noted that as shown in Figure 5 , the customized requirements with higher customized requirement priority values are placed in the upper layers of the lens customization requirement data tree, and the customized requirements with lower customized requirement priority values are placed in the lower layers of the lens customization requirement data tree, so as to clearly show the hierarchical structure of different customized requirements and help technicians in this field better consider the customized requirements in the subsequent development design and optimization process.

[0091] The material selection module is used to obtain the lens customization material set according to the lens customization requirement data tree.

[0092] The method for obtaining the lens customization material set according to the lens customization requirement data tree includes:

[0093] Step300: Preset the material selection layer number of the lens customization requirement data tree as CLS, the value range of CLS is from 1 to , preset the initial value of cls as 1, and the value range of cls is from 1 to CLS; construct CLS requirement structure temporary storage sets, and all the CLS requirement structure temporary storage sets are empty sets when constructed;

[0094] Step301: Obtain the customized requirement structure of the cls-th layer from the lens customization requirement data tree; record the number of customized requirement structures of the cls-th layer as , add the th customized requirement structure to the cls-th requirement structure temporary storage set; if cls is greater than 1, add the customized requirement structures in the (cls - 1)-th requirement structure temporary storage set to the cls-th requirement structure temporary storage set;

[0095] Step302: Input the temporary storage set of the cls-th requirement structure into the lens material model to obtain the set of lens applicable materials for the cls-th lens;

[0096] Step303: Let cls = cls + 1. If cls is less than or equal to CLS, continue to execute Step301 to Step302; if cls is greater than CLS, obtain the set of lens applicable materials for CLS lenses, and analyze and process the set of lens applicable materials for CLS lenses according to a preset method to obtain the set of customized lens materials.

[0097] The training method of the lens material model includes:

[0098] Pre-collect a lens material data set, where the lens material data set includes P groups of lens material data and the corresponding set of lens applicable materials for the P groups of lens material data. P is a positive integer greater than 0. The lens material data includes a temporary storage set of requirement structures; divide the lens material data set into a training set and a validation set, where the training set is used to train the lens material model, and the validation set is used to evaluate the generalization performance of the lens material model;

[0099] During the training process of the lens material model, minimize the cross-entropy loss function as the optimization goal, use the early stopping strategy to monitor the performance of the validation set, and optimize the model performance by continuously adjusting the network parameters; when the prediction accuracy on the validation set reaches the preset accuracy, it is considered that the lens material model has converged and stop training; the lens material model is trained using a deep neural network based on a multi-layer perceptron;

[0100] Convert the lens material data into a high-dimensional feature vector; the input layer of the lens material model receives the high-dimensional feature vector, extracts the non-linear relationship in the lens material data through the hidden layer, and finally the output layer of the lens material model calculates the probability distribution of the set of lens applicable materials through the softmax activation function, and outputs the set of lens applicable materials corresponding to the maximum probability as the final prediction result.

[0101] It should be noted that processing the lens customization requirement data tree in a step-by-step progressive manner not only improves flexibility, accuracy, and interpretability, but also enables a more refined material optimization solution for complex requirement environments. By hierarchically and orderly analyzing and meeting the requirements at each level, the material selection is gradually optimized to ensure that while maintaining the key performance of the lens, the trade-off and adjustment of the secondary requirements of the lens are also taken into account. Thus, it can dynamically adapt to different application scenarios, support rapid adjustment under the condition of lens requirement changes, and ensure the rationality and adaptability of material selection. In contrast, if all customization requirements are input at once, although it can provide a quick overall assessment when the coupling of lens requirements is low, its flexibility is insufficient and it is difficult to meet the requirements of gradual optimization of complex customization requirements. In addition, the one-time input method often leads to instability in lens material selection due to the mutual influence between lens requirements and it is difficult to balance all lens requirements. Therefore, when it comes to the selection and optimization of lens materials, the step-by-step progressive method can better achieve fine control and meet the requirements of high-precision applications.

[0102] The method for analyzing and processing the set of materials applicable to CLS lenses according to a preset method to obtain the lens customization material set includes:

[0103] Step400: Take the intersection of the CLS temporary storage sets of requirement structures to obtain the temporary storage intersection of requirement structures; let cls = 1;

[0104] Step401: Obtain the cls-th temporary storage set of requirement structures; if the cls-th temporary storage set of requirement structures does not contain the customized requirement structure in the temporary storage intersection of requirement structures, then use the cls-th temporary storage set of requirement structures as the cls-th requirement structure de-duplication set and directly execute Step402;

[0105] If the cls-th temporary storage set of requirement structures contains the customized requirement structure in the temporary storage intersection of requirement structures, then delete the customized requirement structure in the temporary storage intersection of requirement structures from the cls-th temporary storage set of requirement structures to obtain the cls-th requirement structure de-duplication set, and execute Step402;

[0106] Step402: Let cls = cls + 1, if cls is less than or equal to CLS, then continue to execute Step401; if cls is greater than CLS, then execute Step403;

[0107] Step403: Deduplicate and aggregate the CLS requirement structures to form a total set of requirement structures. Count the number of each customized requirement structure in the total set of requirement structures, sort the total set of requirement structures in descending order according to the number of customized requirement structures, and select the top N customized requirement structures from the sorted total set of requirement structures in descending order; and construct a lens customized material set from the top N customized requirement structures and the temporary intersection of the requirement structures.

[0108] A lens optimization module for performing lightweight optimization on the lens according to the lens customized material set and the lens customized requirement data tree.

[0109] Such as Figure 4 shown, the method for performing lightweight optimization on the lens according to the lens customized material set and the lens customized requirement data tree includes:

[0110] Step500: Record the number of lens customized materials in the lens customized material set as DZSL, input the lens customized requirement data tree and DZSL lens customized materials into the material usage model respectively, and obtain the lens material usage intervals corresponding to the DZSL lens customized materials; the lens material usage interval is , is the lower limit of the lens material usage interval of the dzsl-th lens customized material, is the upper limit of the lens material usage interval of the dzsl-th lens customized material;

[0111] Initialize the lens material population, where the lens material population includes H lens material candidate solutions, the lens material candidate solutions are lens material combinations that meet the lens customized requirement data tree, and the usage amounts of the lens materials in the lens material combinations are all within the corresponding lens material usage intervals; preset the maximum number of iterations as ZDCS, preset the initial value of zdcs as 1, the value range of zdcs is from 1 to ZDCS, and execute Step501;

[0112] Step501: Perform lens material fitness evaluation on the H lens material candidate solutions in the lens material population to obtain the lens material fitness values corresponding to the H lens material candidate solutions, and execute Step502;

[0113] Step502: Sort the H lens material fitness values in descending order, select the top G lens material fitness values from the sorted H lens material fitness values, and construct a lens material elite population from the lens material candidate solutions corresponding to the G lens material fitness values, and execute Step503;

[0114] Step503: Perform mutation operations on the elite population of lens materials according to a preset method to obtain F mutated lens material candidate solutions, add the F mutated lens material candidate solutions to the elite population of lens materials; overwrite the elite population of lens materials onto the population of lens materials, let H = G + F, and execute Step504;

[0115] Step504: Let zdcs = zdcs + 1. If zdcs is less than or equal to ZDCS, continue to execute Step501; if zdcs is greater than ZDCS, select the lens material candidate solution with the largest lens material fitness value from the population of lens materials as the optimal lens material lightweight solution.

[0116] The training method of the material usage model includes:

[0117] Pre-collect a material usage data set, where the material usage data set includes Y groups of material usage data and the corresponding lens material usage intervals for the Y groups of material usage data. Y is a positive integer greater than 0. The material usage data includes the lens customization requirement data tree and the lens customization materials; divide the material usage data set into a training set and a validation set, where the training set is used to train the material usage model, and the validation set is used to evaluate the generalization performance of the material usage model;

[0118] During the training process of the material usage model, minimize the cross-entropy loss function as the optimization objective, use the early stopping strategy to monitor the performance of the validation set, and optimize the model performance by continuously adjusting the network parameters; when the prediction accuracy on the validation set reaches the preset accuracy, it is considered that the material usage model has converged and stop the training; the material usage model is trained using a deep neural network based on a multi-layer perceptron;

[0119] Convert the material usage data into a high-dimensional feature vector; the input layer of the material usage model receives the high-dimensional feature vector, extracts the non-linear relationship in the material usage data through the hidden layer, and finally the output layer of the material usage model calculates the probability distribution of the lens material usage interval through the softmax activation function, and outputs the lens material usage interval corresponding to the maximum probability as the final prediction result.

[0120] The method for obtaining the lens material fitness value includes:

[0121] ;

[0122] ;

[0123] ;

[0124] ;

[0125] Among them, is the lens material fitness value for the h-th lens material candidate solution, is the quality function for the h-th lens material candidate solution, is the optical performance function for the h-th lens material candidate solution, is the constraint penalty term, is a constant, is the mass of the -th customized lens material in the h-th lens material candidate solution, is the -th customized lens material density, is the -th customized lens material reflectivity, is the reflectivity in the lens customization requirement data tree, is the dz-th constraint condition in the h-th lens material candidate solution, is the penalty coefficient, and are fitness coefficients, DZ is the number of customization requirements, is the cosecant function, is the secant function.

[0126] The method of mutating the elite population of lens materials according to a preset method to obtain F mutated lens material candidate solutions includes:

[0127] Preset a first fitness threshold and a second fitness threshold, where the first fitness threshold is less than the second fitness threshold;

[0128] Construct a high-fitness material set from the lens material candidate solutions in the elite population of lens materials whose lens material fitness values are greater than or equal to the second fitness threshold; construct a medium-fitness material set from the lens material candidate solutions in the elite population of lens materials whose lens material fitness values are greater than or equal to the first fitness threshold and less than the second fitness threshold; construct a low-fitness material set from the lens material candidate solutions in the elite population of lens materials whose lens material fitness values are less than the first fitness threshold;

[0129] Perform times of random pairwise crossover between the lens material candidate solutions in the high-fitness material set to obtain lens material candidate solutions, and add the lens material candidate solutions to the high-fitness material set;

[0130] Perform times of random pairwise crossover between the lens material candidate solutions in the medium-fitness material set and the lens material candidate solutions in the high-fitness material set to obtain Candidate solutions for lens materials; cross the candidate solutions for lens materials in the fitness set of materials with the candidate solutions for lens materials in the fitness set of materials times randomly and cross each other to obtain candidate solutions for lens materials; candidate solutions for lens materials and candidate solutions for lens materials are added to the fitness set of materials;

[0131] Cross the candidate solutions for lens materials in the low-fitness set of materials with the candidate solutions for lens materials in the high-fitness set of materials times randomly and cross each other to obtain candidate solutions for lens materials; perform random mutual crossing in the low-fitness set of materials to obtain candidate solutions for lens materials; candidate solutions for lens materials and candidate solutions for lens materials are added to the low-fitness set of materials;

[0132] When , stop the above random mutual crossing operation to obtain F mutant candidate solutions for lens materials.

[0133] Embodiment 2

[0134] Please refer to Figure 3 shown. This embodiment provides a lightweight structure design device for an optical lens, further including:

[0135] A threshold setting module for dynamically setting a first fitness threshold and a second fitness threshold according to the lens customization requirement data tree and G lens material fitness values.

[0136] The method for dynamically setting the first fitness threshold and the second fitness threshold according to the lens customization requirement data tree and G lens material fitness values includes:

[0137] Construct a set of lens material fitness values from G lens material fitness values;

[0138] Input the lens customization requirement data tree and the set of lens material fitness values into a threshold setting model to obtain the first fitness threshold and the second fitness threshold.

[0139] The training method of the threshold setting model includes:

[0140] Pre-collect a threshold setting data set, where the threshold setting data set includes S sets of threshold setting data, as well as the first fitness threshold and the second fitness threshold corresponding to the S sets of threshold setting data. S is a positive integer greater than 0. The threshold setting data includes a lens customization requirement data tree and a lens material fitness value set; divide the threshold setting data set into a training set and a validation set. The training set is used to learn the threshold setting model parameters, and the validation set is used to evaluate the generalization ability of the threshold setting model to avoid overfitting;

[0141] During the training process of the threshold setting model, combine a dynamic learning rate adjustment strategy and introduce an early stopping mechanism. With minimizing the cross-entropy loss function as the optimization goal, when the performance of the validation set meets the preset requirements, automatically stop the training to ensure the convergence effect of the threshold setting model; the threshold setting model is implemented based on a neural network model architecture; the neural network model consists of an input layer, a hidden layer, and an output layer. The input layer receives the threshold setting data and converts it into a high-dimensional feature vector. The hidden layer uses an activation function to extract the complex non-linear patterns of the threshold setting data. The output layer calculates the probability distributions of the first fitness threshold and the second fitness threshold through the softmax activation function, and takes the first fitness threshold and the second fitness threshold corresponding to the maximum probability as the prediction result and outputs it.

[0142] It should be noted that this method makes full use of the multi-level requirement information contained in the lens customization requirement data tree and combines it with the lens material fitness value set, enabling the threshold setting to perform accurate calculations based on multi-dimensional requirement features and the lens material fitness value distribution, avoiding the limitations of traditional static threshold setting methods, and improving the accuracy and self-adaptability of the fitness threshold.

[0143] The construction of the lens material fitness value set enables the threshold setting model to comprehensively analyze the fitness distribution of candidate materials, ensuring that the first fitness threshold and the second fitness threshold can be dynamically adjusted according to the global characteristics of the material fitness, thereby realizing the efficient screening of lens materials with different performance requirements and enhancing the flexibility and pertinence of material selection.

[0144] Based on the lens customization requirement data tree and the lens material fitness value set, dynamically setting the first fitness threshold and the second fitness threshold not only improves the accuracy and flexibility of lens material selection, but also enhances the adaptability to different customization requirements, ensuring the coordinated optimization of the optical performance, mechanical performance, and lightweight target of the lens.

[0145] Embodiment 3

[0146] Please refer to Figure 2 As shown, this embodiment provides a lightweight structure design method for an optical lens, which further includes:

[0147] Obtain lens customization requirement data, where the lens customization requirement data includes lens optical performance requirement data, lens physical performance requirement data, and customer requirement budget;

[0148] Process the lens customization requirement data using a feature extraction method to obtain a set of customization requirement structures;

[0149] Construct a lens customization requirement data tree based on the set of customization requirement structures;

[0150] Obtain a lens customization material set according to the lens customization requirement data tree;

[0151] Perform lightweight optimization on the lens according to the lens customization material set and the lens customization requirement data tree.

Claims

1. A lightweight structure design method for an optical lens, characterized in that: include: Acquire lens customization demand data, wherein the lens customization demand data includes lens optical performance demand data, lens physical performance demand data and customer demand budget; Using feature extraction methods to process lens customization demand data, obtaining a customized demand structure set; Construct a lens customization requirement data tree according to the customization requirement structure set; Obtain a lens customized material set according to the lens customized demand data tree; The lens is lightweight and optimized based on the lens customized material set and lens customized demand data tree.

2. The lightweight structure design method of an optical lens according to claim 1, characterized in that: Methods for lightweight tuning of lenses include: Step 500: record the number of lens customized materials in the lens customized material set as DZSL, input the lens customized demand data tree and DZSL lens customized materials into the material usage model respectively, and obtain the lens material usage interval corresponding to the DZSL lens customized materials; initialize the lens material population containing H lens material candidate solutions; set the maximum number of iterations to ZDCS, set zdcs=1, and the value range of zdcs is 1 to ZDCS; Step 501: Evaluate the lens material fitness values ​​corresponding to H lens material candidate solutions in the lens material population; Step 502: Sort the fitness values ​​of the H lens materials in descending order, and select the fitness values ​​of the first G lens materials to form an elite population of lens materials; Step 503: Perform mutation operation on the lens material elite population according to the preset method, obtain F mutated lens material candidate solutions and add them to the lens material elite population; cover the lens material elite population to the lens material population, and set H=G+F; Step 504: Let zdcs=zdcs+1. If zdcs is less than or equal to ZDCS, continue to execute Step 501 to Step 503; if zdcs is greater than ZDCS, select the lens material candidate solution with the largest lens material fitness value from the lens material population as the optimal solution for lightweight lens material.

3. The lightweight structure design method of an optical lens according to claim 2, characterized in that: The method for obtaining F candidate solutions of variant lens materials includes: A first fitness threshold and a second fitness threshold are preset, and the first fitness threshold is smaller than the second fitness threshold; an elite population of lens materials is divided according to the first fitness threshold and the second fitness threshold to obtain a material high fitness set, a material medium fitness set, and a material low fitness set; Random crossover is performed in the material high fitness set to obtain Candidate solutions of lens materials are added to the material high fitness set; The lens material candidate solutions in the material medium fitness set are randomly crossed with the lens material candidate solutions in the material high fitness set to obtain The candidate solutions of lens materials are added to the fitness set in the material; random crossover is performed in the fitness set in the material to obtain Candidate solutions of lens materials are added to the fitness set of materials; The lens material candidate solutions in the material low fitness set are randomly crossed with the lens material candidate solutions in the material high fitness set to obtain The candidate solutions of lens materials are added to the material low fitness set; random crossover is performed in the material low fitness set to obtain Candidate solutions of lens materials are added to the material low fitness set; when When , the random crossover operation is stopped to obtain F candidate solutions of variant lens materials.

4. The lightweight structure design method of an optical lens according to claim 3, characterized in that: The method for constructing the lens customization demand data tree includes: Step 200: The height of the preset lens customization requirement data tree is , An integer greater than 0, preset Customized demand priority thresholds, The priority threshold of customized requirements is , ; Let sg = 1, the value range of sg is 1 to SG; Step 201: If sg is equal to 1, set the customized requirement priority value in the customized requirement structure set to be greater than or equal to The customized demand structure is constructed into the sg-th layer of the lens customized demand data tree; If sg is greater than 1 and less than or equal to , then the customized requirement priority value in the customized requirement structure set is greater than or equal to and less than The customized demand structure is constructed into the sg-th layer of the lens customized demand data tree; If sg is equal to , then the customized demand priority value in the customized demand structure set is less than The customized demand structure is constructed into the sg-th layer of the lens customized demand data tree; Step 202: Let sg = sg + 1. If sg is less than or equal to SG, continue to execute Step 201; if sg is greater than SG, obtain the lens customization demand data tree and end the current process.

5. The lightweight structure design method of an optical lens according to claim 4, characterized in that: The method for obtaining the customized requirement structure set includes: Step 100: record the number of customized requirements in the lens customized requirement data as DZ, set dz=1, and the value range of dz is 1 to DZ; construct DZ customized requirements into a lens customized requirement set; Step 101: Input the dzth customized requirement in the lens customized requirement set into the requirement priority value model to obtain the corresponding customized requirement priority value; Step 102: Construct the dzth customized requirement and the customized requirement priority value into a corresponding customized requirement structure; Step 103: Add the dzth customized requirement structure to the customized requirement structure collection; Step 104: Let dz = dz + 1. If dz is less than or equal to DZ, continue to execute Step 101 to Step 103. If dz is greater than DZ, a customized demand structure set is obtained and the current process ends.

6. The lightweight structure design method of an optical lens according to claim 5, characterized in that: The method for obtaining a lens customized material set according to the lens customized demand data tree includes: Step 300: The material selection layer number of the lens customization demand data tree is CLS, and the value range of CLS is 1 to , let cls=1, the value range of cls is 1 to CLS; build a temporary storage set of CLS demand structures; Step 301: Add lens customization requirements to the data tree The customized requirement structure is added to the temporary storage collection of the clsth requirement structure. is the number of customized requirement structures at the cls-th layer; if cls is greater than 1, the customized requirement structure in the cls-1-th requirement structure temporary storage set is added to the cls-th requirement structure temporary storage set; Step 302: Input the temporary storage set of the cls-th requirement structure into the lens material model to obtain the set of materials applicable to the cls-th lens; Step 303: Let cls=cls+1. If cls is less than or equal to CLS, continue to execute Step 301 to Step 302. If cls is greater than CLS, obtain CLS lens-suitable material sets, analyze and process the CLS lens-suitable material sets according to a preset method, and obtain a lens customized material set.

7. The lightweight structure design method of an optical lens according to claim 6, characterized in that: The method of analyzing and processing the CLS lens applicable material set according to the preset method to obtain the lens customized material set includes: Step 400: Take the intersection of CLS demand structure temporary storage sets to obtain the demand structure temporary storage intersection; set cls=1; Step 401: Get the clsth demand structure temporary storage set; if the clsth demand structure temporary storage set does not contain the customized demand structure in the demand structure temporary storage intersection set, then use the clsth demand structure temporary storage set as the clsth demand structure deduplication set, and directly execute Step 402; If the cls-th demand structure temporary storage set contains the customized demand structure in the demand structure temporary storage intersection set, then delete the customized demand structure in the demand structure temporary storage intersection set from the cls-th demand structure temporary storage set to obtain the cls-th demand structure deduplication set, and execute Step 402; Step 402: Let cls = cls + 1. If cls is less than or equal to CLS, continue to execute Step 401; if cls is greater than CLS, execute Step 403; Step 403: Build a total set of demand structures from the CLS demand structures without duplication, count the number of each customized demand structure in the total set of demand structures, sort the total set of demand structures in descending order according to the number of customized demand structures, select the first N customized demand structures from the total set of demand structures after descending order; and build the first N customized demand structures and the temporarily stored intersection of the demand structures into a lens customized material set.

8. The lightweight structure design method of an optical lens according to claim 7, characterized in that: The training method of the material usage model includes: Pre-collecting a material usage data set, wherein the material usage data set includes Y groups of material usage data and lens material usage intervals corresponding to the Y groups of material usage data, where Y is a positive integer greater than 0, and the material usage data includes a lens customization demand data tree and lens customization materials; dividing the material usage data set into a training set and a validation set, wherein the training set is used to train a material usage model, and the validation set is used to evaluate the generalization performance of the material usage model; During the training process of the material usage model, the minimization of the cross entropy loss function is used as the optimization goal, and the early stopping strategy is used to monitor the performance of the validation set. The model performance is optimized by continuously adjusting the network parameters. When the prediction accuracy on the validation set reaches the preset accuracy, the training is stopped. The material usage model is trained using a deep neural network based on a multi-layer perceptron. The material usage data is converted into a high-dimensional feature vector; the input layer of the material usage model receives the high-dimensional feature vector, and the nonlinear relationship in the material usage data is extracted through the hidden layer. Finally, the output layer of the material usage model calculates the probability distribution of the lens material usage interval through the softmax activation function, and outputs the lens material usage interval corresponding to the maximum probability as the final prediction result; The lens material model is obtained by the same training method as the material dosage model.

9. The lightweight structure design method of an optical lens according to claim 8, characterized in that: The training method of the demand priority value model includes: Pre-collect a demand priority value data set, wherein the demand priority value data set includes Q groups of demand priority value data and customized demand priority values ​​corresponding to the Q groups of demand priority value data, where Q is a positive integer greater than 0, and the demand priority value data includes customized demands; divide the demand priority value data set into a training set and a validation set, wherein the training set is used to learn demand priority value model parameters, and the validation set is used to evaluate the generalization ability of the demand priority value model to avoid overfitting; During the training process of the demand priority value model, a dynamic learning rate adjustment strategy is combined with an early stopping mechanism to minimize the cross entropy loss function as the optimization goal. When the performance of the validation set meets the preset requirements, training is automatically stopped to ensure the convergence effect of the demand priority value model. The demand priority value model is implemented based on the neural network model architecture. The neural network model consists of an input layer, a hidden layer, and an output layer. The input layer receives the demand priority value data and converts it into a high-dimensional feature vector. The hidden layer uses an activation function to extract the nonlinear pattern of the demand priority value data. The output layer calculates the probability distribution of the customized demand priority value through the softmax activation function, and takes the customized demand priority value corresponding to the maximum probability as the prediction result and outputs it.

10. A lightweight structure design device for an optical lens, implementing the lightweight structure design method for an optical lens according to any one of claims 1 to 9, characterized in that: include: A demand collection module is used to obtain lens customization demand data, wherein the lens customization demand data includes lens optical performance demand data, lens physical performance demand data and customer demand budget; A first processing module is used to process the lens customization requirement data using a feature extraction method to obtain a customization requirement structure set; The second processing module is used to construct a lens customization requirement data tree according to the customization requirement structure set; A material selection module, used to obtain a lens customized material set according to a lens customized demand data tree; The lens tuning module is used to perform lightweight tuning on the lens based on the lens customized material set and the lens customized demand data tree.

Images