CN117826448B - Self-correcting high-order aberration lens based on transition region optimization and processing method - Google Patents
Self-correcting high-order aberration lens based on transition region optimization and processing method Download PDFInfo
- Publication number
- CN117826448B CN117826448B CN202410100043.0A CN202410100043A CN117826448B CN 117826448 B CN117826448 B CN 117826448B CN 202410100043 A CN202410100043 A CN 202410100043A CN 117826448 B CN117826448 B CN 117826448B
- Authority
- CN
- China
- Prior art keywords
- parameter
- data
- transition region
- neural network
- super
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000007704 transition Effects 0.000 title claims abstract description 89
- 230000004075 alteration Effects 0.000 title claims abstract description 49
- 238000005457 optimization Methods 0.000 title claims abstract description 20
- 238000003672 processing method Methods 0.000 title abstract description 5
- 238000000034 method Methods 0.000 claims abstract description 45
- 230000008859 change Effects 0.000 claims abstract description 44
- 238000003062 neural network model Methods 0.000 claims abstract description 42
- 230000003287 optical effect Effects 0.000 claims abstract description 37
- 230000008569 process Effects 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 30
- 239000011521 glass Substances 0.000 claims abstract description 14
- 230000006870 function Effects 0.000 claims description 58
- 238000009499 grossing Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 14
- 238000012360 testing method Methods 0.000 claims description 14
- 230000000694 effects Effects 0.000 claims description 10
- 238000012795 verification Methods 0.000 claims description 8
- 230000003213 activating effect Effects 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000013101 initial test Methods 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 4
- 238000004088 simulation Methods 0.000 claims description 4
- 230000008451 emotion Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000010200 validation analysis Methods 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 abstract description 7
- 239000000463 material Substances 0.000 abstract description 3
- 238000011960 computer-aided design Methods 0.000 abstract description 2
- 238000012937 correction Methods 0.000 description 15
- 238000013461 design Methods 0.000 description 8
- 238000013528 artificial neural network Methods 0.000 description 5
- 210000001508 eye Anatomy 0.000 description 5
- 238000002360 preparation method Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 3
- 206010010071 Coma Diseases 0.000 description 2
- 235000004035 Cryptotaenia japonica Nutrition 0.000 description 2
- 206010052143 Ocular discomfort Diseases 0.000 description 2
- 102000007641 Trefoil Factors Human genes 0.000 description 2
- 235000015724 Trifolium pratense Nutrition 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 208000003464 asthenopia Diseases 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000001815 facial effect Effects 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 210000001747 pupil Anatomy 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 206010019233 Headaches Diseases 0.000 description 1
- 201000009310 astigmatism Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000005252 bulbus oculi Anatomy 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000004313 glare Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 231100000869 headache Toxicity 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G02—OPTICS
- G02C—SPECTACLES; SUNGLASSES OR GOGGLES INSOFAR AS THEY HAVE THE SAME FEATURES AS SPECTACLES; CONTACT LENSES
- G02C7/00—Optical parts
- G02C7/02—Lenses; Lens systems ; Methods of designing lenses
- G02C7/024—Methods of designing ophthalmic lenses
- G02C7/027—Methods of designing ophthalmic lenses considering wearer's parameters
-
- G—PHYSICS
- G02—OPTICS
- G02C—SPECTACLES; SUNGLASSES OR GOGGLES INSOFAR AS THEY HAVE THE SAME FEATURES AS SPECTACLES; CONTACT LENSES
- G02C7/00—Optical parts
- G02C7/02—Lenses; Lens systems ; Methods of designing lenses
- G02C7/024—Methods of designing ophthalmic lenses
-
- G—PHYSICS
- G02—OPTICS
- G02C—SPECTACLES; SUNGLASSES OR GOGGLES INSOFAR AS THEY HAVE THE SAME FEATURES AS SPECTACLES; CONTACT LENSES
- G02C7/00—Optical parts
- G02C7/02—Lenses; Lens systems ; Methods of designing lenses
- G02C7/024—Methods of designing ophthalmic lenses
- G02C7/028—Special mathematical design techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Ophthalmology & Optometry (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Optics & Photonics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Eyeglasses (AREA)
Abstract
The invention belongs to the technical field of computer aided design, and particularly relates to a self-correcting high-order aberration lens based on transition region optimization and a processing method thereof, wherein the method comprises the following steps: collecting parameter data of an optical zone of the lens and ergonomic data of the lens dispenser; designing a neural network model, inputting the neural network model into parameter data and human engineering data, and outputting the neural network model into a curvature change scheme of a transition region; training the neural network model, and performing super-parameter adjustment based on feedback data of a user in the training process; outputting a curvature change scheme of the transition region by using the trained neural network model; the lens is processed based on the parameter data of the optical zone, the curvature profile of the transition zone, and the dimensions of the edge zone. According to the curvature change scheme of the transition area, the lens material or curvature is adjusted, so that the transition area can be smoothly transited to different areas, the generation of optical distortion is reduced, and the visual performance and individuation adaptability of the glasses are improved.
Description
Technical Field
The invention belongs to the technical field of computer aided design, and particularly relates to a self-correcting high-order aberration lens based on transition region optimization and a processing method.
Background
Higher order aberrations refer to changes in the shape of a wavefront at a position farther from the pupil or center of the optical system, and are more complex, more subtle deformations of the wavefront, typically represented by polynomials of three and higher orders, unlike common lower order aberrations, such as spherical aberration, astigmatism, etc., including coma, trefoil, and spherical aberration.
Higher order aberrations may cause the image to distort, deform or irregularly shaped, thereby reducing visual quality, which is more pronounced especially in low light conditions or in higher contrast scenes; higher order aberrations can lead to visual fatigue, eye discomfort and even headache, particularly where the line of sight is frequently changed or the lighting conditions are greatly changed, the effect of higher order aberrations on vision is more pronounced; for tasks requiring high visual performance, such as driving, reading small words, etc., higher order aberrations may limit the sensitivity of the eye to details and the accurate perception of the shape of the object, thereby affecting the task's performance.
In the current eyeglass manufacturing technology, the main focus is on the selection of parameters of optical areas of lenses, which does play an important role in the adjustment of higher order aberrations, however, the conventional method ignores the importance of transition areas in the correction of higher order aberrations, so that a series of challenges and problems exist in practical use, and the importance of transition areas is mainly represented in the following aspects:
The transition region connects the optical region and the edge region, and the design goal is to smoothly adjust the wave front shape in the process, the higher-order aberration is usually represented by complex deformation of the wave front, and the existence of the transition region enables the lens to more smoothly transition to different regions, so that the generation of optical distortion is reduced; through personalized design of the transition region, higher-order aberration correction which is more accurate and meets actual needs can be provided according to different factors such as eyeball shape, wavefront aberration and the like of an individual, and the lens can better adapt to visual demands of different individuals through personalized design of the transition region, so that correction effect is improved; the reasonably designed transition region is beneficial to reducing visual discomfort such as glare, image distortion and the like possibly caused in the high-order aberration correction process, and the optimization of the transition region enables eyes to adapt to different visual scenes more easily, so that visual fatigue and discomfort are reduced.
How to realize the better preparation of the self-correcting higher-order aberration lens through the personalized design of the transition area on the basis of the optical area and the edge area becomes a technical problem to be solved at present.
Disclosure of Invention
The invention provides a self-correcting high-order aberration lens based on transition region optimization and a processing method thereof, which can effectively solve the problems in the background technology.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
the lens comprises an optical area, a transition area and an edge area, wherein the optical area is used for correcting high-order aberration through parameter setting, and the edge area is connected with a glasses frame through size adjustment;
Comprising the following steps:
collecting parameter data of an optical zone of the lens and ergonomic data of a lens dispenser;
Designing a neural network model, wherein the input of the neural network model is the parameter data and the human engineering data, and the output is a curvature change scheme of the transition region;
Training the neural network model, and performing super-parameter adjustment based on feedback data of a user in the training process;
outputting a curvature change scheme of the transition region by using the trained neural network model;
the lens is processed based on the parameter data of the optical zone, the curvature profile of the transition zone, and the dimensions of the edge zone.
Further, performing hyper-parameter adjustment based on feedback data of a user during training includes:
Processing the feedback data, and extracting user preference information and specific task requirements;
and adjusting the super-parameter candidate value through the user preference information, and adjusting the super-parameter combination through the specific task requirement.
Further, the adjusting process of the super parameter includes:
Determining a value point range for each super parameter to be adjusted;
Combining the value point ranges of the super parameters to form a grid;
For each hyper-parameter combination in the grid, performing model training by using a training set, and evaluating model performance by using a verification set;
The best performing hyper-parameter combination is selected based on the performance of the model on the validation set.
Further, the adjusting process of the super parameter includes:
Determining a value point range for each super parameter to be adjusted;
selecting a prior model for evaluating a relationship between the hyper-parameters and an objective function;
defining the objective function, wherein the objective function maps the super parameters to performance indexes of the neural network model on a verification set;
The following steps are repeatedly circulated until the preset iteration times are met or convergence conditions are reached:
randomly selecting partial initial hyper-parameter combinations for testing to obtain initial test results of the neural network model;
Updating the prior model through the initial test result to obtain an estimated value of the relation between the super parameter and the objective function;
based on the estimated values, determining the next super-parameter combination for test by using the prior model.
Further, the neural network model includes:
the input layer is used for directly splicing the parameter data and the ergonomic data to obtain spliced data, the node number of the input layer is m+n, m is the dimension of the parameter data, and n is the dimension of the ergonomic data;
the first full-connection layer is connected with m nodes corresponding to the parameter data of the input layer;
The second full-connection layer is connected with n nodes of the input layer corresponding to the human engineering data;
the dimension adjusting layer is used for adjusting the dimensions of the first full-connection layer and the second full-connection layer to be consistent and correlating the dimensions to obtain an output matrix;
Activating a function, and introducing nonlinear characteristics after adding each node of the first full-connection layer and the second full-connection layer;
the output layer outputs the rate change scheme, and the node number of the output layer is matched with the curvature change scheme dimension of the transition region;
A loss function for measuring the difference between the model output and the real transition region curvature variation scheme.
Further, the dimension adjustment layer performs element-by-element multiplication on the first full-connection layer and the second full-connection layer after dimension adjustment is consistent.
Further, the curvature change scheme is a group of numerical values, and represents the change condition of curvature in the transition area;
Also included is a process for adjusting the set of values, comprising:
selecting a smoothing function and determining an adjustment region of the transition region for which a smooth transition is desired based on the set of values;
Setting parameters of the smoothing function based on the adjustment region;
Applying the smoothing function with the parameter set to the curvature change scheme;
And evaluating the effect of smooth transition through simulation or actual test, and obtaining a curvature change scheme meeting the smooth requirement through iteration.
The self-correcting high-order aberration lens based on transition region optimization comprises an optical region, a transition region and an edge region, wherein the optical region is used for correcting high-order aberration through parameter setting, and the edge region is connected with a glasses frame through size adjustment;
The curvature change scheme of the transition region is output through a neural network model;
the input of the neural network model is the parameter data of the optical area and the human engineering data of the lens matching person, and the neural network model carries out super-parameter adjustment based on the feedback data of the user in the training process.
Further, the candidate values of the super parameters are adjusted according to the user preference information extracted from the feedback data, and the super parameter combination is adjusted according to the specific task requirements extracted from the feedback data.
Further, the neural network model comprises an input layer, the parameter data and the human engineering data are directly spliced to obtain spliced data, the node number of the input layer is m+n, m is the dimension of the parameter data, and n is the dimension of the human engineering data;
the first full-connection layer is connected with m nodes corresponding to the parameter data of the input layer;
The second full-connection layer is connected with n nodes of the input layer corresponding to the human engineering data;
the dimension adjusting layer is used for adjusting the dimensions of the first full-connection layer and the second full-connection layer to be consistent and correlating the dimensions to obtain an output matrix;
Activating a function, and introducing nonlinear characteristics after adding each node of the first full-connection layer and the second full-connection layer;
the output layer outputs the rate change scheme, and the node number of the output layer is matched with the curvature change scheme dimension of the transition region;
A loss function for measuring the difference between the model output and the real transition region curvature variation scheme.
According to the technical scheme, the lens material or curvature can be adjusted according to the curvature change scheme of the transition region, so that the transition region can be smoothly transited to different regions, and the generation of optical distortion is reduced; meanwhile, the size of the edge area is adjusted, so that the lenses can be well connected with the glasses frame, and the wearing comfort of the whole glasses is maintained; by reasonably adjusting various parameters of the lens, the self-correction of the higher-order aberration is realized, and the visual performance and personalized adaptability of the glasses are improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present invention, and other drawings may be obtained according to the drawings without inventive effort to those skilled in the art.
FIG. 1 is a flow diagram of a method of self-correcting higher order aberration lens processing based on transition zone optimization;
FIG. 2 is a flow chart diagram of a super parameter adjustment based on user feedback data;
FIG. 3 is a flow chart diagram of a process for adjusting a super parameter;
fig. 4 is a structural diagram of a neural network model.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.
Example 1
A method for processing a self-correcting high-order aberration lens based on transition region optimization comprises an optical region, a transition region and an edge region, wherein the optical region is used for correcting high-order aberration through parameter setting, and the edge region is connected with a spectacle frame through size adjustment; as shown in fig. 1, the preparation process includes:
Collecting parameter data of an optical zone of the lens and ergonomic data of the lens dispenser;
in this step, during the eye measurement of the lens dispenser, an autorefractor or other optical measurement device may be used to obtain measurement data, where the data needs to include parameter data related to higher order aberrations, and the part of the data at least includes:
The wavefront distortion coefficient describes the change in shape of the wavefront as the light passes through the optical system, including the effects of higher order terms; polynomial coefficients of higher order aberrations, polynomial coefficients representing more complex, subtle deformations of the wavefront, may include coma, trefoil, spherical aberration, etc.; zernike polynomial coefficients, mathematical tools for describing the shape of a wavefront, reflect aberrations of different orders and directions, including higher order aberrations.
During the preparation of the glasses, the acquisition of ergonomic data of the lens dispenser, which determines the dimensions of the edge area, ensures the comfort and adaptability of the glasses, may include:
Acquiring data of the facial shape of the lens dispenser using a facial scanner or measuring tool to better accommodate the lens; measuring the distance between the pupils of the dispenser to ensure a proper field of view; the distance from the eyes to the lenses, which is measured to ensure that the lenses meet the needs of the lens dispenser when worn; such ergonomic data helps to adjust the eyeglass frame and transition region to provide a better wearing experience.
Designing a neural network model, wherein the input of the neural network model is parameter data and ergonomic data, and the output of the neural network model is a curvature change scheme of a transition region; in the step, the neural network aims at learning the mapping relation between input data and the curvature change scheme of the transition region so as to automatically design the shape of the transition region according to the individual characteristics of a lens distributor and the parameters of the lens, thereby realizing more accurate personalized design and optimizing the correction effect of higher-order aberration;
Training the neural network model, and performing super-parameter adjustment based on feedback data of a user in the training process; in this step, a specific training process may include:
Data preparation, using the collected labeled training dataset, including optical region parameter data, ergonomic data, and corresponding curvature variation schemes of the transition region; data standardization, namely, carrying out standardization or normalization on input data, ensuring the consistent numerical ranges of different characteristics, and facilitating the stability of a training process; dividing the data set into a training set, a verification set and a test set so as to train, verify and evaluate the performance of the model; initializing a model, and initializing weights and parameters of a neural network; selecting a loss function, namely selecting a proper loss function for measuring the difference between the model output and a curvature change scheme of a real transition region, wherein the specific loss function can be selected from a mean square error, an average absolute error, a Huber loss, a logarithmic loss or a custom loss function and the like; an optimizer selection selects an optimizer, such as a variation of the gradient descent method, for adjusting the model parameters to reduce the loss function.
A training cycle is performed during the implementation: training is performed through a plurality of training loop iterations, each loop may include the following steps in the implementation: forward propagation, namely, inputting data through a neural network, and calculating the output of a model; loss calculation, namely calculating the difference between the model output and the real label by using a loss function; counter-propagating, and calculating the gradient of the loss function to the model parameters through a counter-propagating algorithm; updating parameters of the model using the selected optimizer to reduce the loss function;
outputting a curvature change scheme of the transition region by using the trained neural network model;
the lens is processed based on the parameter data of the optical zone, the curvature profile of the transition zone, and the dimensions of the edge zone.
Specifically, the lens material or curvature can be adjusted according to the curvature variation scheme of the transition region, so that the transition region can be smoothly transited to a different region, and the generation of optical distortion is reduced; meanwhile, the size of the edge area is adjusted, so that the lenses can be well connected with the glasses frame, and the wearing comfort of the whole glasses is maintained. The method is a final preparation stage of the whole process, and by reasonably adjusting various parameters of the lens, the self-correction of the higher-order aberration is realized, and the visual performance and individuation adaptability of the glasses are improved.
As a preference of the above embodiment, as shown in FIG. 2, performing super parameter adjustment based on feedback data of a user during training includes:
Processing the feedback data, and extracting user preference information and specific task requirements; the super parameter candidate value is adjusted through the user preference information, and the super parameter combination is adjusted through the specific task requirement.
During training, mechanisms may be set up to actively or passively collect feedback data of the user, which may include, in particular, user experience, assessment in real-time use, and specific feedback for different aspects, such as comfort, clarity, adaptability, etc. The collected user feedback data is processed, including but not limited to text analysis, emotion analysis, keyword extraction, etc., to extract useful information therefrom, the goal of which is to convert the subjective perception of the user into quantifiable or analyzable metrics.
Through the processed feedback data, user preference information of the user in terms of comfort, definition, adaptability and the like is extracted, which may include preference for some super-parameter values or preference for specific performance, the extracted user preference information is used for adjusting candidate values of the super-parameters, for example, if the user prefers a smaller learning rate, the user may use words like "better adapted", "more natural feel" and the like, which may be related to the smaller learning rate, then options of smaller learning rate may be more included in the super-parameter candidate values.
In the implementation process, for each super parameter to be adjusted, a set of possible value ranges can be determined through participation of user preference information, for example, different candidate values can be set for the learning rate and the hidden layer node number, the possible values of the super parameters are combined to form a grid, for example, if the learning rate has three candidate values (0.001, 0.01 and 0.1), and the hidden layer node number has three candidate values (64, 128 and 256), nine combinations can be created;
by combining these possible values, a grid is formed:
(0.001,64), (0.001,128), (0.001,256)，
(0.01,64), (0.01,128), (0.01,256)，
(0.1,64), (0.1,128), (0.1,256)，
Also, by the processed feedback data, information about the user's specific task requirements is extracted, which may include specific requirements in terms of sharpness, higher order aberration correction, etc. in certain task scenarios, the extracted specific task requirements are used to adjust the overall hyper-parameter combinations, i.e. the overall hyper-parameter combinations are adjusted by the specific task requirements, each hyper-parameter combination being adjusted. For example, if user feedback requires better higher order aberration correction in a driving scenario, the superparameter combinations may be adjusted to emphasize performance in this regard, specific examples are provided below:
one set of initial superparameter combinations is:
learning rate: 0.01;
Hidden layer node number: 128.
Higher order aberration correction module weights: 0.5;
definition optimization module weight: 0.5;
The performance of the user feedback model in terms of higher-order aberration correction in the driving scene needs to be improved, the main performance index focused by the user in the driving scene is found to be higher-order aberration correction through user feedback, and the set of super-parameter combinations are adjusted according to the requirement of higher-order aberration correction as follows:
Learning rate: 0.005, reducing the learning rate to more finely adjust the model parameters;
Hidden layer node number: 128.
Higher order aberration correction module weights: 0.7, enhancing the effect of higher order aberration correction;
definition optimization module weight: 0.3, reducing the influence of sharpness optimization on the overall performance.
As a preference of the above embodiment, the adjustment process for the super parameter includes:
Determining a value point range for each super parameter to be adjusted;
Combining the value point ranges of the super parameters to form a grid;
for each hyper-parameter combination in the grid, performing model training by using a training set, and evaluating model performance by using a verification set;
The best performing hyper-parameter combination is selected based on the performance of the model on the validation set.
This process is repeated until the optimal combination of hyper-parameters is found. The grid search in the above embodiment is a simple and intuitive method, but in the case of a large number of super parameters, the search space becomes very large, possibly resulting in high calculation costs. In practical applications, as another specific embodiment, as shown in fig. 3, the adjustment process for the super parameter includes:
Determining a value point range for each super parameter to be adjusted; in this step, it is necessary to define explicitly the value range of each super parameter to be adjusted, and the selection of the super parameter affects the performance of the neural network model, such as learning rate, number of hidden layer nodes, etc., so that the search space is ensured to be wide enough by this step, so as to effectively find the optimal super parameter combination;
Selecting a priori model, wherein the priori model is used for evaluating the relation between the super parameter and the objective function; the commonly used prior model comprises a Gaussian process which can provide uncertainty estimation of different points in the hyper-parameter space, and the searching process can be better guided by selecting a proper prior model;
Defining an objective function, wherein the objective function maps the super parameters to performance indexes of the neural network model on the verification set; the objective function defines performance indexes which are expected to be maximized or minimized in the implementation process, and maps the super parameters to the performance indexes, such as accuracy or loss functions, of the neural network model on the verification set;
The following steps are repeatedly circulated until the preset iteration times are met or convergence conditions are reached:
Randomly selecting partial initial hyper-parameter combinations for testing to obtain test results of an initial neural network model; i.e. using knowledge of the current prior model to explore new hyper-parametric points, the aim is to ensure that the search does not fall into a locally optimal solution and to provide sufficient exploration space;
Updating the prior model through an initial test result to obtain an estimated value of the relation between the super parameter and the objective function; the above-described update provides a new estimate of the relationship of the hyper-parameters to the objective function, helping to adjust the search direction to better focus on the possibly better hyper-parameter area;
based on the estimated values, the prior model is utilized to determine the next super-parameter combination for test. According to the current estimation of the super-parameter relation, the step is used for guiding searching more intelligently, and the searching efficiency is improved.
In the process, full utilization of priori knowledge and dynamic update of experimental results are combined, so that the searching direction can be intelligently adjusted, and the super-parameter combination of global optimum or local optimum can be more rapidly found; compared with the traditional grid search or random search, the method is more efficient by randomly selecting test points and using a priori model, and can find a better super-parameter combination in a smaller test number, and particularly has more obvious advantages under the condition of larger search space or non-convex objective function.
As a preference of the above embodiment, as shown in fig. 4, the neural network model includes:
the input layer is used for directly splicing the parameter data and the ergonomic data to obtain spliced data, the node number of the input layer is m+n, m is the dimension of the parameter data, and n is the dimension of the ergonomic data;
The first full-connection layer is connected with m nodes of the parameter data corresponding to the input layer;
The second full-connection layer is connected with n nodes corresponding to the human engineering data of the input layer;
the dimension adjusting layer is used for adjusting the dimensions of the first full-connection layer and the second full-connection layer to be consistent and correlating the dimensions to obtain an output matrix;
activating a function, and introducing nonlinear characteristics after adding each node of the first full-connection layer and the second full-connection layer;
the output layer outputs a rate change scheme, and the node number of the output layer is matched with the curvature change scheme dimension of the transition region;
A loss function for measuring the difference between the model output and the real transition region curvature variation scheme.
Directly splicing the parameter data and the ergonomic data through the input layer to form an input vector, so that the network can consider the two types of information at the same time; in the above preferred scheme, the first full connection layer is responsible for processing parameter data, and extracting advanced features of input data by learning weights and biases; the second full connection layer is responsible for processing ergonomic data, and also extracting advanced features of input data by learning weights and biases; through the dimension adjustment layer, the model can simultaneously consider interaction information of parameter data and human engineering data. In the implementation process, an activation function is added behind each node of the first full-connection layer and the second full-connection layer, nonlinear characteristics are introduced, and nonlinear activation functions (such as ReLU, sigmoid and the like) help the network learn more complex modes and representations; the loss function is used to measure the difference between the model output and the real transition region curvature variation scheme, and common loss functions include mean square error (Mean Squared Error) and the like, and are used to measure the average difference between the model output and the real value.
The dimension adjusting layer is used for carrying out element-by-element multiplication on the first full-connection layer and the second full-connection layer after dimension adjustment is consistent. In the implementation process, the dimension adjusting layer realizes the feature fusion of the output of the first full-connection layer and the second full-connection layer by multiplying the elements by each other, the features of the first full-connection layer and the second full-connection layer are related by the elements by multiplying the elements by each other, and fused feature representation is generated, so that the neural network can better learn the complex relationship between parameter data and human engineering data; the model can better capture the correlation between the parameter data and the human engineering data, and the sensitivity of the model to the input data is improved; after the output dimensions are consistent, the complexity of subsequent calculation in the neural network can be reduced, the feature representation with consistent dimensions is easier to process in subsequent layers, and the calculation structure of the whole network is simplified.
Preferably, the curvature change scheme is a set of values, which represent the change condition of the curvature in the transition region; also included is a process for adjusting a set of values, comprising:
Selecting a smoothing function and determining an adjustment region of the transition region requiring smooth transition based on a set of values;
setting parameters of a smoothing function based on the adjustment region;
Applying the smooth function with the parameters being set and a curvature change scheme;
And evaluating the effect of smooth transition through simulation or actual test, and obtaining a curvature change scheme meeting the smooth requirement through iteration.
The optimization scheme composed of the steps can realize smooth adjustment of the curvature variation scheme so as to ensure that the ideal optical performance is obtained in the transition region. In the implementation process, selecting a proper smoothing function is a key for ensuring the smoothness of curvature change, the selection of the smoothing function can be customized according to the characteristics and design requirements of curvature change, and by using the proper smoothing function, unnecessary spikes or fluctuations in a curvature change scheme can be avoided, and common smoothing functions include a gaussian function, an S-curve (Sigmoid), a B-spline curve and the like.
By analyzing a set of values of the curvature change scheme, an adjustment area in which smooth transition is required is determined, so that the scheme can intensively process the possible changes in the optical transition area instead of excessively adjusting the whole lens, the parameters of the smoothing function can precisely control the smoothness and the characteristics of the transition area, and the parameterized settings enable the scheme to flexibly adapt to different optical requirements and different expectations of smoothness and transition. Applying the selected smoothing function to the curvature variation scheme ensures a smooth transition in the adjustment region, which can be achieved by applying the smoothing function point by point, or using interpolation methods, which ensures that the curvature variation maintains consistency in the transition region.
The adjusted curvature variation scheme is evaluated through simulation or actual testing, which includes checking smoothness in the transition region, consistency of the overall curvature, and whether the user design requirements are met, and the result of the evaluation can be used for further adjustment and optimization. If the evaluation result does not meet the requirements, the scheme allows to iteratively adjust the parameters of the smoothing function until an ideal smoothing effect is obtained, this flexibility ensuring the adjustability and adaptability of the scheme.
Example two
The self-correcting high-order aberration lens based on transition region optimization comprises an optical region, a transition region and an edge region, wherein the optical region is used for correcting high-order aberration through parameter setting, and the edge region is connected with a glasses frame through size adjustment; the curvature change scheme of the transition region is output through a neural network model; the input of the neural network model is the parameter data of the optical area and the human engineering data of the lens matching person, and the neural network model carries out super-parameter adjustment based on the feedback data of the user in the training process.
As a preference of the embodiment, the candidate values of the super parameters are adjusted according to the user preference information extracted from the feedback data, and the super parameter combination is adjusted according to the specific task requirements extracted from the feedback data.
The neural network model comprises an input layer, wherein parameter data and ergonomic data are directly spliced to obtain spliced data, the node number of the input layer is m+n, m is the dimension of the parameter data, and n is the dimension of the ergonomic data;
The first full-connection layer is connected with m nodes of the parameter data corresponding to the input layer;
The second full-connection layer is connected with n nodes corresponding to the human engineering data of the input layer;
the dimension adjusting layer is used for adjusting the dimensions of the first full-connection layer and the second full-connection layer to be consistent and correlating the dimensions to obtain an output matrix;
activating a function, and introducing nonlinear characteristics after adding each node of the first full-connection layer and the second full-connection layer;
the output layer outputs a rate change scheme, and the node number of the output layer is matched with the curvature change scheme dimension of the transition region;
A loss function for measuring the difference between the model output and the real transition region curvature variation scheme.
The technical effects achieved in this embodiment are the same as those in the above embodiment, and will not be described here again.
The foregoing has outlined and described the basic principles, features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (6)
1. The lens comprises an optical area, a transition area and an edge area, wherein the optical area is used for correcting high-order aberration through parameter setting, and the edge area is connected with a glasses frame through size adjustment;
Characterized by comprising the following steps:
collecting parameter data of an optical zone of the lens and ergonomic data of a lens dispenser;
Designing a neural network model, wherein the input of the neural network model is the parameter data and the human engineering data, and the output is a curvature change scheme of the transition region;
Training the neural network model, and performing super-parameter adjustment based on feedback data of a user in the training process;
During training, a mechanism is set up to actively or passively collect feedback data of the user, specifically including feelings and evaluations of the user in real-time use and specific feedback for different aspects including comfort, definition and adaptability; processing the collected user feedback data, including text analysis, emotion analysis, keyword extraction methods, to extract useful information therefrom;
outputting a curvature change scheme of the transition region by using the trained neural network model;
processing the lens based on the parameter data of the optical zone, the curvature variation scheme of the transition zone, and the size of the edge zone;
Performing hyper-parameter adjustment based on feedback data of a user during training, including: processing the feedback data, and extracting user preference information and specific task requirements; adjusting a super-parameter candidate value through the user preference information, and adjusting a super-parameter combination through the specific task requirement;
the neural network model includes:
the input layer is used for directly splicing the parameter data and the ergonomic data to obtain spliced data, the node number of the input layer is m+n, m is the dimension of the parameter data, and n is the dimension of the ergonomic data;
the first full-connection layer is connected with m nodes corresponding to the parameter data of the input layer;
The second full-connection layer is connected with n nodes of the input layer corresponding to the human engineering data;
the dimension adjusting layer is used for adjusting the dimensions of the first full-connection layer and the second full-connection layer to be consistent and correlating the dimensions to obtain an output matrix;
Activating a function, and introducing nonlinear characteristics after adding each node of the first full-connection layer and the second full-connection layer;
the output layer outputs the rate change scheme, and the node number of the output layer is matched with the curvature change scheme dimension of the transition region;
A loss function for measuring the difference between the model output and the real transition region curvature variation scheme.
2. The method for processing the lens based on the transition region optimization and capable of automatically correcting the high-order aberration according to claim 1, wherein the adjusting process of the super parameter comprises the following steps:
Determining a value point range for each super parameter to be adjusted;
Combining the value point ranges of the super parameters to form a grid;
For each hyper-parameter combination in the grid, performing model training by using a training set, and evaluating model performance by using a verification set;
The best performing hyper-parameter combination is selected based on the performance of the model on the validation set.
3. The method for processing the lens based on the transition region optimization and capable of automatically correcting the high-order aberration according to claim 1, wherein the adjusting process of the super parameter comprises the following steps:
Determining a value point range for each super parameter to be adjusted;
selecting a prior model for evaluating a relationship between the hyper-parameters and an objective function;
defining the objective function, wherein the objective function maps the super parameters to performance indexes of the neural network model on a verification set;
The following steps are repeatedly circulated until the preset iteration times are met or convergence conditions are reached:
randomly selecting partial initial hyper-parameter combinations for testing to obtain initial test results of the neural network model;
Updating the prior model through the initial test result to obtain an estimated value of the relation between the super parameter and the objective function;
based on the estimated values, determining the next super-parameter combination for test by using the prior model.
4. The method for processing the self-correcting higher-order aberration-correcting lens based on the transition region optimization according to claim 1, wherein the dimension adjusting layer is used for carrying out element-by-element multiplication on the first fully-connected layer and the second fully-connected layer after dimension adjustment is consistent.
5. The method for processing the self-correcting higher-order aberration-correcting lens based on transition region optimization according to claim 1, wherein the curvature change scheme is a set of values representing the change condition of curvature in the transition region;
Also included is a process for adjusting the set of values, comprising:
selecting a smoothing function and determining an adjustment region of the transition region for which a smooth transition is desired based on the set of values;
Setting parameters of the smoothing function based on the adjustment region;
Applying the smoothing function with the parameter set to the curvature change scheme;
And evaluating the effect of smooth transition through simulation or actual test, and obtaining a curvature change scheme meeting the smooth requirement through iteration.
6. The self-correcting high-order aberration lens based on transition region optimization comprises an optical region, a transition region and an edge region, wherein the optical region is used for correcting high-order aberration through parameter setting, and the edge region is connected with a glasses frame through size adjustment;
the method is characterized in that the curvature change scheme of the transition region is output through a neural network model;
the input of the neural network model is the parameter data of the optical area and the human engineering data of the lens matching person, and the neural network model carries out super-parameter adjustment based on the feedback data of the user in the training process;
During training, a mechanism is set up to actively or passively collect feedback data of the user, specifically including feelings and evaluations of the user in real-time use and specific feedback for different aspects including comfort, definition and adaptability; processing the collected user feedback data, including text analysis, emotion analysis, keyword extraction methods, to extract useful information therefrom;
The candidate value of the super parameter is adjusted according to the user preference information extracted from the feedback data, and the super parameter combination is adjusted according to the specific task requirement extracted from the feedback data;
The neural network model comprises an input layer, wherein the parameter data and the human engineering data are directly spliced to obtain spliced data, the node number of the input layer is m+n, m is the dimension of the parameter data, and n is the dimension of the human engineering data;
the first full-connection layer is connected with m nodes corresponding to the parameter data of the input layer;
The second full-connection layer is connected with n nodes of the input layer corresponding to the human engineering data;
the dimension adjusting layer is used for adjusting the dimensions of the first full-connection layer and the second full-connection layer to be consistent and correlating the dimensions to obtain an output matrix;
Activating a function, and introducing nonlinear characteristics after adding each node of the first full-connection layer and the second full-connection layer;
the output layer outputs the rate change scheme, and the node number of the output layer is matched with the curvature change scheme dimension of the transition region;
A loss function for measuring the difference between the model output and the real transition region curvature variation scheme.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410100043.0A CN117826448B (en) | 2024-01-24 | 2024-01-24 | Self-correcting high-order aberration lens based on transition region optimization and processing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410100043.0A CN117826448B (en) | 2024-01-24 | 2024-01-24 | Self-correcting high-order aberration lens based on transition region optimization and processing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117826448A CN117826448A (en) | 2024-04-05 |
| CN117826448B true CN117826448B (en) | 2024-08-27 |
Family
ID=90506420
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410100043.0A Active CN117826448B (en) | 2024-01-24 | 2024-01-24 | Self-correcting high-order aberration lens based on transition region optimization and processing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117826448B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113056699A (en) * | 2018-11-15 | 2021-06-29 | 依视路国际公司 | Method and system for determining parameters for manufacturing an optical article and corresponding optical article |
| CN116670569A (en) * | 2020-11-03 | 2023-08-29 | 罗登斯托克有限责任公司 | Method for calculating spectacle lenses based on big data method and machine learning |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110110611A (en) * | 2019-04-16 | 2019-08-09 | 深圳壹账通智能科技有限公司 | Portrait attribute model construction method, device, computer equipment and storage medium |
| CN110135263A (en) * | 2019-04-16 | 2019-08-16 | 深圳壹账通智能科技有限公司 | Portrait attribute model construction method, device, computer equipment and storage medium |
| EP3730998A1 (en) * | 2019-04-23 | 2020-10-28 | Carl Zeiss Vision International GmbH | Determination of at least one optical parameter of a spectacle lens |
| CN111134613B (en) * | 2019-11-21 | 2022-04-05 | 明灏科技（北京）有限公司 | Image recognition-based orthokeratology lens fitting method and system |
| CN214623225U (en) * | 2021-01-28 | 2021-11-05 | 上海龙旗科技股份有限公司 | Glasses with intelligent warning function |
| JP6980958B1 (en) * | 2021-06-23 | 2021-12-15 | 中国科学院西北生態環境資源研究院 | Rural area classification garbage identification method based on deep learning |
| CN115308925A (en) * | 2022-06-29 | 2022-11-08 | 深圳市挚钥文化科技有限责任公司 | Eye movement tracking method based on contact lenses and contact lenses |
| CN115082269B (en) * | 2022-07-18 | 2023-01-20 | 华北理工大学 | A teaching planning method and system based on big data |
| WO2024113241A1 (en) * | 2022-11-30 | 2024-06-06 | 中国科学院深圳先进技术研究院 | Cross-user hand joint angle estimation method and system, and computer device |
-
2024
- 2024-01-24 CN CN202410100043.0A patent/CN117826448B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113056699A (en) * | 2018-11-15 | 2021-06-29 | 依视路国际公司 | Method and system for determining parameters for manufacturing an optical article and corresponding optical article |
| CN116670569A (en) * | 2020-11-03 | 2023-08-29 | 罗登斯托克有限责任公司 | Method for calculating spectacle lenses based on big data method and machine learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117826448A (en) | 2024-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7954950B2 (en) | System and method for analyzing wavefront aberrations | |
| JP5698150B2 (en) | How to calculate optics | |
| US20030090623A1 (en) | Method for designing optical elements | |
| KR102649211B1 (en) | Device for visual equipment selection based on machine learning | |
| CN102947747B (en) | Method for optimizing a spectacle lens for the wavefront aberrations of an eye and lens | |
| JP6052296B2 (en) | Spectacle lens design method and spectacle lens design system | |
| CN118115281B (en) | Risk level assessment method for banking financial clients | |
| KR20140004722A (en) | A method for determining target optical functions | |
| CN113221794A (en) | Training data set generation method, device, equipment and storage medium | |
| CN113989217A (en) | Human eye diopter detection method based on deep learning | |
| CN117826448B (en) | Self-correcting high-order aberration lens based on transition region optimization and processing method | |
| CN112584750B (en) | Method for optimizing an optical aid by automatically determining subjective visual acuity | |
| CN118052151B (en) | Intelligent real experience simulation system and method for design of child sunglasses frame | |
| US20230296918A1 (en) | Improved calculation of ophthalmological lenses | |
| CN114781187A (en) | An optimization method for smooth-phase multifocal intraocular lens design | |
| CN115826263A (en) | Adjustment of subjective and objective refraction | |
| CN113780234A (en) | Edge-guided human eye image analysis method | |
| CN114945859B (en) | Lens evaluation method, lens design method, method for manufacturing spectacle lens, and lens evaluation program | |
| US20240302678A1 (en) | Method for calculating a spectacle lens on the basis of a big data approach and machine learning | |
| CN111680741A (en) | An automatic adjustment method of computer-aided interferometer based on deep learning | |
| Timar-Fulep et al. | Modelling the human eye and designing presbyopia correction with optical simulation software | |
| CN118605039B (en) | Method and system for adapting fluorescein staining of cornea shaping mirror | |
| US20240115126A1 (en) | A device and method for automatically evaluating visual equipment | |
| CN108066006A (en) | Vision correction effect simulation system and method | |
| CN119338552A (en) | A glasses intelligent fitting system and method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |