A lens assembly correction and compensation method based on key tolerance prediction

Abstract

This invention discloses a lens assembly correction and compensation method based on key tolerance prediction. The method comprises: obtaining the model of the lens to be corrected; determining all lens tolerance items and the MTF image corresponding to the lens model; determining the key tolerance items based on all lens tolerance items and each curve in the MTF image; inputting the MTF image into a pre-trained first prediction model to predict the key tolerance items; and inputting the key tolerance items into a second prediction model as basic parameters of a particle swarm algorithm to solve for the optimal correction compensation. By fully utilizing a deep network, this invention achieves precise compensation for lens assembly.

Description

Technical Field

[0001] The present invention relates to the field of computer technology applications, and in particular to a lens correction and compensation method based on key tolerance prediction. Background Art

[0002] In order to meet various differentiated needs, a variety of precision lenses are required. Precision lenses are often assembled from multiple lenses.

[0003] During the lens manufacturing stage, production processes are carefully controlled to maximize precision and uniformity between lenses. However, lens tolerances still exist. During lens assembly, assembly errors can interact with each other, making it difficult to meet the requirements of the optical lens design documents. Therefore, various methods can be used during the lens assembly stage to control these tolerances, allowing them to interact and offset each other, thereby achieving high-quality lenses and improving the yield rate of the lens production line.

[0004] Next, we will introduce the correction and compensation algorithm of traditional lenses:

[0005] Before manufacturing each lens product, a set of simulation data needs to be obtained first to establish a quadratic function relationship. This is used to describe the relationship between the tolerance items of the last lens and the lens MTF image when there are no errors in the assembly of other lenses.

[0006] In the subsequent actual manufacturing process, first attach all lenses except the last one as axially symmetrically as possible, then place the last lens in the initial position, obtain the MTF image, and check whether there is any problem with the MTF. If the MTF image does not meet the evaluation criteria, calculate the tolerance items and the adjustment tolerance amounts to be adjusted based on the current MTF image and the functional relationship established by the simulation, then move the last lens so that its tolerance items are the calculated values, and then regenerate the MTF image. If the MTF image still does not meet the standards, repeat the above steps. Iterate in this way until the MTF image is qualified or the number of iterations reaches the upper limit. If the above process is completed and the MTF image still does not meet the MTF evaluation criteria, the lens is unqualified and discarded.

[0007] The above correction process has several problems: First, it ignores existing errors. The errors of the previous lenses are not taken into account, but are treated as ideal. Second, the functional relationship is roughly described. The tolerance terms and MTF images are described only by quadratic functions, and the interaction between the tolerance terms is ignored. Third, the adjustment time is long. The position of each calculated result needs to be adjusted. The actual detection of the MTF curve requires multiple iterations.

[0008] As the pace of digitalization in industrial manufacturing accelerates, data from every manufacturing process is continuously collected and stored, and the amount of available data continues to increase. With the rise and application of neural networks and deep learning, how to leverage existing data to optimize production lines through machine learning and other methods has garnered widespread attention.

[0009] Current machine learning optimization in production processes falls into two main areas: 1. Optimization without changing production parameters during the manufacturing process. Examples include root cause analysis to prevent similar issues from recurring and early prediction of manufacturing results to reduce unnecessary subsequent production processes for problematic products. These methods do not directly adjust production parameters, but indirectly improve product quality. 2. Dynamic adjustment of production parameters for process optimization. The optimal production parameters are determined using data from already-running production processes. Higher quality is achieved by adjusting parameters based on product characteristics and specific optimization goals. This is typically achieved through additional optimization modules and self-optimizing control systems based on analytical process models.

[0010] In the field of process optimization, in order to use machine learning methods to optimize process parameters, the typical workflow includes the following four steps: 1. Generate a database through a small number of experiments or run simulations using the DOE method; 2. Use statistical or machine learning methods to model the physical correlation between process parameters and quality standards; 3. Use the created process model to optimize process parameters; 4. Manually or automatically adjust process parameters.

[0011] In the past, process parameter optimization was achieved through traditional methods such as Newton's method, hill climbing, or gradient descent algorithms, which only achieved local optimality. However, in recent years, heuristic learning combined with machine learning for process optimization has gained increasing attention.

[0012] Kant et al. used a neural network (ANN) and a genetic algorithm (GA) to optimize cutting parameters in the milling process to improve quality. Denkena et al. used a support vector machine to predict geometric deviations of the workpiece. To optimize the cutting parameters, they were sampled on a grid, and the optimal parameters were selected based on the SVM predictions. In the parameter optimization task of roller grinding, Zhang et al. used RSM for quality prediction and a hybrid PSO algorithm for optimization.

[0013] In the aforementioned fields, the manufacturing process is automated, and process parameters are automatically measured. Therefore, process parameter adjustments are made when the parameters and corresponding results are available. However, in the field of lens assembly, many indicators are difficult to measure, such as decentration errors. This results in a large number of process parameters being lost during the manufacturing process, leaving only the final lens optical characteristics. This lack of process parameters prevents mainstream deep learning-based parameter optimization algorithms from being directly applied to the lens assembly field.

[0014] Therefore, there is an urgent need to provide a lens assembly correction and compensation method. Summary of the Invention

[0015] The purpose of the present invention is to address the deficiencies of the existing technology and provide a lens assembly correction and compensation method based on key tolerance prediction, so as to overcome the problems of unknown assembly error and rough compensation function in the existing lens assembly correction and compensation process.

[0016] The object of the present invention is achieved through the following technical solutions:

[0017] According to a first aspect of an embodiment of the present invention, a lens assembly correction and compensation method based on critical tolerance prediction is provided, the method comprising:

[0018] Get the model of the lens to be corrected;

[0019] According to the model of the lens to be corrected, determine all tolerance items and MTF images of the lens corresponding to the model;

[0020] Determine the key tolerance items based on all tolerance items of the lens and each curve in the MTF image;

[0021] Input the MTF image into the pre-trained first prediction model to predict the key tolerance item;

[0022] The key tolerance items are input into the second prediction model as the basic parameters of the particle swarm optimization algorithm to obtain the optimal correction compensation amount.

[0023] Furthermore, the process of determining the key tolerance items based on all tolerance items of the lens and each curve in the MTF image includes:

[0024] Get the MTF image simulation data set;

[0025] Obtain the maximum value, maximum point, and value when the horizontal coordinate is zero of each curve in the MTF image as the key point of the MTF image;

[0026] Calculate the impact weight of each tolerance item, sort them, and select the top N tolerance items as key tolerance items;

[0027] The influence weight of each tolerance item is the sum of the square values of the Pearson coefficient of each key point.

[0028] Furthermore, the first prediction model adopts a convolutional neural network; the first prediction model takes the MTF image as input and the key tolerance item as output.

[0029] Furthermore, the first prediction model adopts the ResNet convolutional network.

[0030] Furthermore, the process of the first prediction model includes:

[0031] Get the MTF image simulation data set;

[0032] Enlarging the y-coordinate value of each point on each curve in each MTF image in the MTF image simulation data set by one thousand times;

[0033] Construct an M*N*P image, where M is the maximum value of the y coordinate of all curves, N is the number of x coordinate points, and P is the number of curves in the MTF image;

[0034] Map each curve in the MTF image onto the matrix. For each curve, select a new layer in turn, initialize the layer to all 0s, and then set the coordinate value corresponding to each curve point to 1 to obtain the constructed image.

[0035] The first prediction model is trained with the constructed image as input and the key tolerance term as output.

[0036] Furthermore, the second prediction model adopts a fully connected neural network; the second prediction model takes all tolerance terms as input and MTF image key points as output.

[0037] Furthermore, the key points of the MTF image are the maximum values, maximum points, and values when the horizontal coordinate is zero of each curve in the MTF image.

[0038] Furthermore, the key tolerance term is input into the second prediction model as the basic parameter of the particle swarm optimization algorithm. The process of solving the optimal correction compensation includes:

[0039] Before the particle swarm algorithm starts, the key tolerance items of all particles in the particle swarm are set, and the other tolerance items are zero;

[0040] Determine the weights of key points of different MTF images to form a weight vector;

[0041] The key tolerance term is added to the current solution of each particle and input into the second prediction model to predict the prediction vector of the key point of the MTF image. Then, the weight vector is added to obtain the fitness of the particle.

[0042] The tolerance items that need to be adjusted are organized into a solution set space. Each particle has a random vector in the solution set space. The optimal vector solution is obtained through the particle swarm algorithm update, which is the current optimal correction compensation amount for the lens.

[0043] According to a second aspect of an embodiment of the present invention, an electronic device is provided, comprising a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the above-mentioned lens assembly correction and compensation method based on critical tolerance prediction.

[0044] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the above-mentioned lens assembly correction and compensation method based on critical tolerance prediction is implemented.

[0045] The beneficial effect of the present invention is that it uses deep network technology to quickly and accurately implement lens assembly correction compensation, thereby reducing the number of iterations of the lens correction process and shortening the correction time, thereby improving the production capacity of the production line. At the same time, for a single lens, it is adjusted to the optimal optical characteristics, improving product quality, and also improving the yield rate of the production line, reducing resource waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 This is a flowchart of a lens assembly correction and compensation method based on key tolerance prediction provided by an embodiment of the present invention.

[0048] Figure 2 is a schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0050] The terms used in this invention are for the purpose of describing specific embodiments only and are not intended to limit the invention. The singular forms "a," "the," and "the" used in this invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0051] It should be understood that although the terms "first," "second," "third," etc. may be used in the present invention to describe various information, such information should not be limited to these terms. These terms are merely used to distinguish information of the same type from one another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present invention. Depending on the context, the term "if" as used herein may be interpreted as "when," "when," or "in response to determining."

[0052] The present invention will be described in detail below with reference to the accompanying drawings. Unless there is any conflict, the features of the following embodiments and implementations may be combined with each other.

[0053] To address the problems of missing parameters in the lens assembly process and rough compensation of the lens correction algorithm, the present invention provides a lens assembly correction and compensation method based on key tolerance prediction. It uses a convolutional neural network and a simulation data set for modeling to predict the key tolerances of the lens in each assembly process. Subsequently, a fully connected neural network is used to accurately predict the optical properties of the lens. Finally, it is combined with a particle swarm algorithm to search in the solution set space to obtain the optimal lens correction compensation parameters.

[0054] like Figure 1 As shown, the method of the present invention specifically comprises the following steps:

[0055] Step S1: Determine the lens model that needs to be corrected. In this example, the lens tolerance items and the corresponding MTF image simulation data set corresponding to the model are constructed using CODEV optical imaging design software. The data set size is 10W.

[0056] Step S2: Determine key tolerance items.

[0057] Using a simulated MTF image dataset, we obtained the maximum values, maximum points, and zero values of each curve in the MTF image as key points. We calculated the Pearson correlation coefficient between each key point and each tolerance item. The influence weight of each tolerance item is the sum of the squared Pearson coefficients of that tolerance item on each point, resulting in a weighted influence matrix. Finally, we ranked the influence weights of the tolerance items, selecting the top N items as key tolerance items.

[0058] Step S3: Using the MTF image simulation dataset, a convolutional neural network is used to construct a first prediction model M with the MTF image as input and the key tolerance item as output. I ; Specifically include:

[0059] Step S301, expanding coordinates: the values of each curve point in the MTF image are 0 to 1, and the y coordinate value of each point of each curve in the MTF image is expanded by a thousand times, so as to convert the image scatter curves into an image and extract image features.

[0060] Step S302: Image Construction: Construct an M*N*P image, where M is the maximum y-coordinate value of all curves, N is the number of x-coordinate points, and P is the number of curves in the MTF image. For each curve, select a new layer, initialize it to all zeros, and then set the coordinate value corresponding to each curve point to 1.

[0061] Step S303, model training: take the constructed image as input and the key tolerance item value as output, and train based on the convolution neural network structure to obtain the first prediction model M I In this application, the ResNet convolutional network is selected for training.

[0062] Step S4: Using the MTF image simulation dataset, a fully connected neural network is used to construct a second prediction model M with all tolerance terms as input and MTF image key points as output. Q .

[0063] Furthermore, the key points of the MTF image are the maximum value, the maximum point and the value when the horizontal coordinate is zero of each curve.

[0064] Step S5: Input the actual MTF curve into the first prediction model M I , predict the value V1 of the critical tolerance item.

[0065] The adjustable parameters of the last lens are set to machine zero value, and the actual MTF image is obtained. Then the actual MTF image is input into the first prediction model M I , obtain the actual critical tolerance item value V1 of the lens.

[0066] Step S6: Based on the key tolerance value V1 as the basic parameter, based on the second prediction M Q The particle swarm optimization algorithm of the model is used to find the optimal correction compensation. It includes the following sub-steps.

[0067] Step S601, before the particle swarm algorithm starts, the key tolerance item of all particles in the particle swarm is V1, and the other tolerance items are zero;

[0068] Step S602: Determine the weights of key points of different MTF images according to actual manufacturing needs and assessment requirements to form a weight vector.

[0069] Step S603: Determine the tolerance items to be adjusted and the adjustment range of each tolerance item.

[0070] Step S604: The current solution vector of each particle is added with the key tolerance value V1 and input into the second prediction M of the model. Q In the output, a vector containing the predicted values of the key points of each MTF image is output, and then the fitness value of the particle is obtained according to the manufacturing assessment requirements;

[0071] Specifically, the tolerance term that needs to be adjusted is regarded as a solution set space. Each particle has a random vector in the solution set space. When the particle swarm algorithm iterates, the current solution vector of each particle is added with the key tolerance term value V1, which is the current shot organization tolerance value of the particle, and is input into the second prediction model M. Q In the equation, the values of each MTF key point under the particle solution vector are obtained to form an MTF key point vector, which is multiplied by the previously determined weight vector to obtain the fitness value of the particle.

[0072] In step S605, after calculating the fitness value of each particle in the cluster, the particle's solution vector converges toward the particle with the highest fitness value in the cluster. The PSO algorithm stops if and only if the fitness value increases by less than a preset value or the number of iterations is reached. The solution vector of the particle with the highest fitness value in the current swarm is output, which becomes the optimal correction compensation solution for the lens.

[0073] It should be noted that when a lens model is determined, steps S1 to S4 only need to be performed once, and steps S5 to S6 can be repeated with the actual manufacturing data of the lens model to obtain the correction compensation solution for the different lens.

[0074] like Figure 2 As shown, an embodiment of the present application provides an electronic device, which includes a memory 101 for storing one or more programs and a processor 102. When the one or more programs are executed by the processor 102, any method of the first aspect described above is implemented.

[0075] The system also includes a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected to each other directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, and the processor 102 executes the software programs and modules stored in the memory 101 to perform various functional applications and data processing. The communication interface 103 can be used to communicate signaling or data with other node devices.

[0076] Among them, the memory 101 can be, but is not limited to, a random access memory 101 (Random Access Memory, RAM), a read-only memory 101 (Read Only Memory, ROM), a programmable read-only memory 101 (Programmable Read-Only Memory, PROM), an erasable programmable read-only memory 101 (Erasable Programmable Read-Only Memory, EPROM), an electrically erasable read-only memory 101 (Electric Erasable Programmable Read-Only Memory, EEPROM), etc.

[0077] The processor 102 may be an integrated circuit chip having signal processing capabilities. The processor 102 may be a general-purpose processor 102, including a central processing unit (CPU) 102, a network processor (NP) 102, etc.; it may also be a digital signal processing (DSP) 102, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0078] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can also be implemented in other ways. The method and system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the methods and systems, methods, and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a portion of code, and the module, program segment, or a portion of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0079] In addition, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0080] On the other hand, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which implements a method as described in any one of the first aspects above when executed by the processor 102. If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory 101 (ROM, Read-Only Memory), a random access memory 101 (RAM, Random Access Memory), a magnetic disk or an optical disk.

[0081] The above embodiments are intended only to illustrate the design concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. The scope of protection of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design concepts disclosed in the present invention are within the scope of protection of the present invention.

Claims

1. A lens assembly correction and compensation method based on key tolerance prediction, characterized in that: The method comprises: Get the model of the lens to be corrected; According to the model of the lens to be corrected, determine all tolerance items and MTF images of the lens corresponding to the model; Determining key tolerance items based on all tolerance items of the lens and each curve in the MTF image; including: obtaining an MTF image simulation data set; obtaining the maximum value, maximum point, and value when the horizontal coordinate is zero of each curve in the MTF image as the key points of the MTF image; calculating the influence weight of each tolerance item, sorting them, and selecting the top N tolerance items as key tolerance items; wherein the influence weight of each tolerance item is the sum of the squared values of the Pearson coefficient of the tolerance item on each key point; Input the MTF image into the pre-trained first prediction model to predict the key tolerance item; The key tolerance items are input into the second prediction model as the basic parameters of the particle swarm optimization algorithm to obtain the optimal correction compensation amount.

2. The lens assembly correction and compensation method based on critical tolerance prediction according to claim 1, wherein: The first prediction model uses a convolutional neural network; the first prediction model takes the MTF image as input and the key tolerance item as output.

3. The lens assembly correction and compensation method based on critical tolerance prediction according to claim 2, wherein: The first prediction model uses the ResNet convolutional network.

4. The lens assembly correction and compensation method based on critical tolerance prediction according to claim 1, wherein: The process of the first prediction model includes: Get the MTF image simulation data set; Enlarging the y-coordinate value of each point on each curve in each MTF image in the MTF image simulation data set by one thousand times; Construct an M×N×P image, where M is the maximum y-coordinate of all curves, N is the number of x-coordinate points, and P is the number of curves in the MTF image; Map each curve in the MTF image onto the matrix. For each curve, select a new layer in turn, initialize the layer to all 0s, and then set the coordinate value corresponding to each curve point to 1 to obtain the constructed image. The first prediction model is trained with the constructed image as input and the key tolerance term as output.

5. The lens assembly correction and compensation method based on critical tolerance prediction according to claim 1, wherein: The second prediction model adopts a fully connected neural network; the second prediction model takes all tolerance items as input and MTF image key points as output.

6. The lens assembly correction and compensation method based on critical tolerance prediction according to claim 5, wherein: The key points of the MTF image are the maximum values, maximum points, and values when the horizontal coordinate is zero of each curve in the MTF image.

7. The lens assembly correction and compensation method based on critical tolerance prediction according to claim 1, wherein: The key tolerance term is input into the second prediction model as the basic parameter of the particle swarm algorithm. The process of solving the optimal correction compensation includes: Before the particle swarm algorithm starts, the key tolerance items of all particles in the particle swarm are set, and the other tolerance items are zero; Determine the weights of key points of different MTF images to form a weight vector; The key tolerance term is added to the current solution of each particle and input into the second prediction model to predict the prediction vector of the key point of the MTF image. Then, the weight vector is added to obtain the fitness of the particle. The tolerance items that need to be adjusted are organized into a solution set space. Each particle has a random vector in the solution set space. The optimal vector solution is obtained through the particle swarm algorithm update, which is the current optimal correction compensation amount for the lens.

8. An electronic device comprising a memory and a processor, characterized in that: The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the lens assembly correction and compensation method based on critical tolerance prediction as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the lens assembly correction and compensation method based on critical tolerance prediction as described in any one of claims 1 to 7 is implemented.

Images