Active compensation method for manufacturing tolerance of glass aspherical surface in glass-plastic hybrid lens

Abstract

The application discloses an active compensation method for manufacturing tolerance of glass aspherical surface in a glass-plastic hybrid lens. The method comprises the following steps: firstly, a first glass-plastic hybrid lens with tolerance is constructed, and a real point spread function image is obtained; then, a point spread function simulation model is built; then, a mean square error between the real point spread function image and a simulation point spread function image is calculated and used as a loss function, a predicted tolerance and a predicted point spread function image are obtained; a second glass-plastic hybrid lens with tolerance is constructed according to the design parameters of the first glass-plastic hybrid lens with tolerance and the predicted tolerance, six pose variables of the aspherical glass lens are updated, and an optimal compensation pose is obtained; and the aspherical glass lens of the first glass-plastic hybrid lens with tolerance is adjusted according to the compensation pose, so that the active compensation is completed. The active compensation method has the characteristics of fast calculation speed and low cost consumption, and can compensate for the image quality degradation caused by the manufacturing tolerance by adjusting the pose of the aspherical glass lens.

Description

Technical Field

[0001] This invention relates to an active compensation method in the field of lens assembly, specifically an active compensation method for manufacturing tolerances of aspherical glass in glass-plastic hybrid lenses. Background Technology

[0002] In the design of actual optical imaging systems, in order to minimize image quality degradation caused by aberrations, different evaluation functions are set. By changing parameters such as lens shape and material, various monochromatic aberrations and chromatic aberrations are corrected as much as possible while meeting the design objectives.

[0003] In recent years, mobile imaging devices have shown a clear trend towards lightweighting and miniaturization. To reduce the overall height and size of imaging devices, optical designers have begun to introduce aspherical lenses. Due to limitations in manufacturing processes and costs, most aspherical lenses in mobile phone modules are made of plastic. However, with the increasing popularity of mobile phone photography, people's requirements for the image quality of mobile phone lenses have gradually increased, and the image quality of plastic aspherical lenses has begun to fail to meet these needs due to material limitations. Therefore, some mobile phone manufacturers have begun to try replacing the original two plastic lenses with a single glass lens. Compared with plastic, glass has a more mature coating process, which can further reduce glare and ghosting. At the same time, glass has a wider refractive index and a larger selection space for dispersion coefficient, which is more advantageous in solving problems such as edge aberrations. In addition, replacing multiple plastic lenses with a single glass lens can further reduce the size of the imaging module. However, due to manufacturing cost limitations, the manufacturing error of glass aspherical lenses is always greater than that of plastic aspherical lenses, resulting in a decrease in the image quality of actual tolerance lenses. However, it is impractical to measure the tolerance of each lens individually during mass production, so how to quickly obtain the tolerance value of the lens has become an urgent problem to be solved.

[0004] Furthermore, to compensate for image quality degradation caused by manufacturing tolerances, the only option is to adjust the position of some mirrors or detectors during assembly. The most common assembly compensation method is to adjust the image plane. In 2022, Wu et al. proposed a machine learning alignment method based on stellar images to compensate for image quality degradation that occurs during telescope system operation by adjusting the pose of the image plane. In addition, adjusting the pose of some mirrors in the imaging system can also improve image quality; however, since changing the mirror pose directly affects the final imaging result, existing compensation methods require multiple iterations to calculate the optimal compensation pose, and the time cost is insufficient to meet the needs of actual mass production. Summary of the Invention

[0005] To address the technical problems existing in the background art, such as the difficulty in measuring the tolerance of hybrid glass-plastic lenses individually during mass production, the severe image quality degradation caused by large manufacturing tolerances of the aspherical glass, and the high time cost required to calculate the optimal compensation pose during assembly compensation, this invention proposes an assembly compensation method for the manufacturing tolerances of the aspherical glass in hybrid glass-plastic lenses. This invention constructs a differentiable point spread function simulation model and combines it with actual captured point spread function images of the lens with tolerances to predict the tolerances. Then, the tolerances are incorporated into the ideal design, and the gradient of the loss function with respect to the pose parameters of the aspherical glass lens is calculated backward using the chain rule and updated accordingly. Finally, the method quickly converges to the optimal compensation position, thereby improving the image quality of the lens with tolerances.

[0006] This invention is achieved through the following technical solution:

[0007] S1: Construct a first tolerance glass-plastic hybrid lens and obtain the real point spread function images of the first tolerance glass-plastic hybrid lens at different field-of-view positions;

[0008] S2: Construct a point spread function simulation model based on the design parameters of the first tolerance glass-plastic hybrid lens;

[0009] S3: The simulated point spread function image is obtained by using the point spread function simulation model. Then, the mean square error between the real point spread function image and the simulated point spread function image is calculated and used as the loss function. After updating the tolerance in the point spread function simulation model using the loss function, the final predicted tolerance and the predicted point spread function image are obtained.

[0010] S4: Based on the design parameters and predicted tolerance of the first tolerance glass-plastic hybrid lens, construct the second tolerance glass-plastic hybrid lens. Then, calculate the mean square error of the predicted point spread function image and the point spread function image of the tolerance-free glass-plastic hybrid lens and use it as the loss function. Then, use the loss function to update the pose of the second tolerance glass-plastic hybrid lens to obtain the optimal compensation pose.

[0011] S5: Adjust the pose of the glass aspherical lens in the first tolerance glass-plastic hybrid lens according to the optimal compensation pose, thereby completing the active compensation of the tolerance.

[0012] When the design parameters of the first tolerance glass-plastic hybrid lens include tolerance, the design parameters of the glass-plastic hybrid lens without tolerance are the same as the design parameters of the first tolerance glass-plastic hybrid lens except for tolerance.

[0013] When the design parameters of the first tolerance glass-plastic hybrid lens do not include tolerance, the design parameters of the tolerance-free glass-plastic hybrid lens are the same as those of the first tolerance glass-plastic hybrid lens.

[0014] Specifically, S2 is:

[0015] S21: The design parameters of the first tolerance glass-plastic hybrid lens are used as initialization parameters and input into the point spread function simulation model;

[0016] S22: Start calculating the intersection points of light rays with the surface and the direction cosines after refraction on the point spread function simulation model, one by one, until the last surface, to obtain the intersection points of light rays with the surface and the direction cosines after refraction in all surfaces;

[0017] S23: Based on the intersection points of light rays with the surface in all surface types and the direction cosines after refraction, coherent superposition of light rays is performed on the image plane to obtain point spread functions for different bands and different field positions;

[0018] S24: Linearly add the point spread functions of different bands to obtain a three-channel simulated point spread function image.

[0019] In step S3, after updating the tolerance in the point spread function simulation model using the loss function, the final predicted tolerance and the predicted point spread function image are obtained, specifically as follows:

[0020] First, the gradient of the loss function with respect to the tolerance terms of each surface in the point spread function simulation model is calculated using the chain rule. Then, an optimizer is selected based on the gradient and learning rate, and the tolerance is updated using the optimizer to keep the tolerance within the preset error, thus obtaining the final predicted tolerance and the predicted point spread function image.

[0021] In step S4, the pose of the second tolerance glass-plastic hybrid lens is updated using a loss function to obtain the optimal compensated pose, specifically as follows:

[0022] First, the chain rule is used to calculate the gradient of the loss function with respect to the six pose variables of the aspherical glass lens in the second tolerance glass-plastic hybrid lens. Then, the pose is updated based on the gradient to obtain the optimal compensation pose.

[0023] In step S1, the method for obtaining the true point spread function images of the first tolerance glass-plastic hybrid lens at different field-of-view positions includes:

[0024] LED lights were installed on a black background. The first tolerance glass-plastic hybrid lens was then fixed on the black background and focused and photographed. The real point spread function images of the first tolerance glass-plastic hybrid lens at different field-of-view positions were then obtained.

[0025] In S3, the formula for calculating the loss function is:

[0026]

[0027] Among them, LMSE Let n be the mean squared error loss function, c be the total number of fields of view, and PSF be the field of view number. c This is the true point spread function image of the c-th field of view of the first tolerance glass-plastic hybrid lens. This is the simulated point spread function image of the c-th field of view obtained through the point spread function simulation model.

[0028] In S3, the predicted tolerances include the tilt of each surface about the x, y and z axes, the eccentricity in the x and y directions, the thickness in the z direction, the radius of curvature tolerance, the refractive index tolerance of the material, and the Abbe number tolerance.

[0029] In S4, the pose includes six pose variables corresponding to each aspherical glass lens, which are the translation amount along the x, y and z axes and the rotation amount around the x, y and z axes for each aspherical glass lens.

[0030] Specifically, S5 is:

[0031] A six-axis robotic arm is used to adjust the position of the aspherical glass lens to the optimal compensation position; then the aspherical glass lens is solidified to complete the assembly of the overall lens module, achieving active tolerance compensation.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] (1) The active compensation method proposed in this invention can predict the manufacturing tolerance by using the point spread function of different fields of view obtained by shooting, without having to measure the manufacturing tolerance of each lens of the glass-plastic hybrid lens with tolerance one by one.

[0034] (2) The active compensation method of the present invention can compensate for the image quality degradation caused by manufacturing tolerance by adjusting the position and orientation of the aspherical glass lens, thereby improving the imaging quality.

[0035] (3) The present invention has fast calculation speed and low cost in the active compensation process. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating an assembly compensation method for manufacturing tolerances of aspherical glass in a glass-plastic hybrid lens, as proposed in this invention.

[0037] Figure 2 A schematic diagram of a point spread function image capture system with a tolerance-controlled glass-plastic hybrid lens;

[0038] Figure 3 This is a schematic diagram for initializing the eccentricity and tilt tolerances of the surface.

[0039] Figure 4 This is a schematic diagram of ray tracing.

[0040] Figure 5 This is a comparison chart of the point spread function of the lens with tolerance predicted by this invention and the actual lens with tolerance.

[0041] Figure 6 A comparison of the point spread function of lenses with tolerances before and after active compensation.

[0042] In the diagram: LED light source 1, first tolerance glass-plastic hybrid lens 2, lateral displacement stage 3, longitudinal displacement stage 4. Detailed Implementation

[0043] The present invention will now be described in further detail with reference to the accompanying drawings.

[0044] According to the invention, the complete implementation embodiments and their implementation processes are as follows:

[0045] like Figure 1 As shown, the present invention includes the following steps:

[0046] S1: First, assemble and fix the plastic lenses in sequence. That is, apply glue to the assembled plastic lens module, and then place each aspherical glass lens in the plastic lens module to construct the first tolerance glass-plastic hybrid lens and obtain the real point spread function image of the first tolerance glass-plastic hybrid lens at different field positions.

[0047] In S1, the method for obtaining the true point spread function images of the first tolerance glass-plastic hybrid lens at different field-of-view positions includes:

[0048] LED lights are installed on a black background to simulate point light sources. The first tolerance glass-plastic hybrid lens is then fixed on the black background and focused and photographed to obtain the real point spread function images of the first tolerance glass-plastic hybrid lens at different field-of-view positions.

[0049] like Figure 2 As shown, the tolerance-based glass-plastic hybrid lens point spread function imaging system based on a displacement stage includes an LED light source 1, a first tolerance-based glass-plastic hybrid lens 2, a lateral displacement stage 3, and a longitudinal displacement stage 4. The spectral range of the LED light source 1 covers at least the visible spectrum from 400 nm to 700 nm. The first tolerance-based glass-plastic hybrid lens 2 is mounted on the displacement stages 3 and 4, where the lateral displacement stage 3 is responsible for movement perpendicular to the lens optical axis, and the longitudinal displacement stage 4 is responsible for movement parallel to the lens optical axis.

[0050] Adjust the lateral displacement stage 3 so that the optical axis of the first tolerance-adjustable glass-plastic hybrid lens 2 is aligned with the shooting scene. Then, calculate the distance between the first tolerance-adjustable glass-plastic hybrid lens 2 and the shooting scene according to the object-image relationship formula, and adjust the longitudinal displacement stage 4 based on the calculated object distance. The formula for calculating the object distance is:

[0051]

[0052] Where d is the distance between the first tolerance-adjustable hybrid glass-plastic lens 2 and the shooting scene, f is the lens focal length, H is the object height, and h is the image height. For an ideal imaging system without aberrations, the object-side point light source image still converges to a single point on the image plane. Therefore, if the object-side LED light source is considered as a point light source, theoretically, the image formed on the image plane should be an ideal point light source image of one pixel size. Thus, H corresponds to the size of the LED light source, and h corresponds to the detector pixel size at the image plane position of the first tolerance-adjustable hybrid glass-plastic lens 2. However, since actual optical systems inevitably contain aberrations and degradation, the captured point light source image will exhibit varying degrees of energy dispersion and will no longer converge to a single point. Therefore, an image of the LED light source containing multiple field-of-view positions can be captured, which is the point spread function image of the required tolerance-adjustable hybrid glass-plastic lens at different field-of-view positions.

[0053] S2: Construct a point spread function simulation model based on the design parameters of the first tolerance glass-plastic hybrid lens;

[0054] S2 specifically refers to:

[0055] S21: The design parameters (such as curvature, radius, surface thickness, and tolerance value) of the first tolerance glass-plastic hybrid lens are used as initialization parameters and input into the point spread function simulation model to construct the lens model, set the wavelength, field of view and their corresponding weights;

[0056] S22: Based on the number of pupil samples and the incident field angle set during initialization, calculate the intersection points of light rays with the surface and the direction cosines after refraction on the point spread function simulation model, one by one, until the last surface, and obtain the intersection points of light rays with the surface and the direction cosines after refraction in all surfaces.

[0057] S23: Based on the intersection points of light rays with the surface in all surface types and the direction cosines after refraction, coherent superposition of light rays is performed on the image plane. Specifically, each light ray is regarded as a wavelet, and the coherent superposition result of each wavelet at the sampling point on the image plane is calculated to obtain the point spread function value of the object point at the sampling point, thereby obtaining the point spread function of different bands and different field positions.

[0058] S24: Based on the spectral response characteristics of the detector on the image plane, the point spread functions of different bands are linearly added to obtain a three-channel simulated point spread function image. In specific implementation, the point spread functions of each band in the range of 400nm to 700nm are superimposed according to the spectral response curve of the detector to obtain a three-channel simulated point spread function image.

[0059] The initialization of wavelengths and their weights is primarily determined by the application scenario of the imaging system. Since the glass-plastic hybrid lens in this embodiment is generally used for visible light imaging, the wavelength range is set to 400–700 nm, with a wavelength sampling interval of 20 nm, meaning 16 wavelengths are initialized. The corresponding weights are determined by the detector's spectral response curve and can be uniformly set to 1 during initialization. The initialization of the field of view and its weights is similar to that of the wavelengths and their weights. Four key fields of view are selected: 0, 0.3, 0.5, and 0.8. Except for the 0 field of view, each field of view also corresponds to four different positions, resulting in a total of 13 initialized fields of view. These are represented by one central field of view (0°, 0°), and four 0.3 fields of view: (-7.016°, 12.030°), (7.016°, -12.030°), (12.030 ... The lens has four 0.5 fields of view: (-19.299°, 11.428°), (19.299°, -11.428°), (11.428°, -19.299°), and (-11.428°, 19.299°), and four 0.8 fields of view: (0°, -32.893°), (0°, 32.893°), (-32.893°, 0°), and (32.893°, 0°). The weight of each field of view is set to 1. The initialization of each lens surface mainly involves declaring the surface shape. Since this embodiment involves rotationally symmetric extended even-order aspherical surfaces, the initialization of the surface shape can be defined by the following formula:

[0060]

[0061] Where, r 2 =x 2 +y 2 r represents the distance from a point on the surface to the z-axis, c is the curvature at the vertex of the aspherical surface, k is the conic coefficient, and A j For higher-order terms r 2j The coefficients. Furthermore, the material initialization is mainly based on dispersion formulas. This embodiment uses two dispersion formulas: the Schott formula and the Sellmeier formula.

[0062] n Schott 2 (λ)=A1+A2λ 2 +A3λ -2 +A4λ -4 +A5λ -6 +A6λ -8

[0063]

[0064] Where n is the refractive index, λ is the wavelength, and A1 to A6, B1 to B3, and C1 to C3 are the dispersion coefficients of the material.

[0065] If the design file contains tolerance values, tolerance initialization is also required. In this embodiment, the tolerances considered include the tilt of each surface about the x, y, and z axes, the eccentricity in the x and y directions, the thickness in the z direction, the radius of curvature tolerance, the material's refractive index tolerance, and the Abbe number tolerance. The tilt of each surface about the x, y, and z axes, and the eccentricity in the x and y directions, can be initialized through coordinate system transformation and applied to the subsequent ray tracing section. For example... Figure 3 As shown, the eccentricity of the i-th face is O. i =(d x ,d y The tilt tolerance R, expressed as a rotation matrix, is 0). i As shown in the following formula:

[0066]

[0067] Among them, (α) i ,β i ,γ i Let be the Euler angles of the i-th surface when it is rotated sequentially around the x-axis, y'-axis, and z”-axis. The thickness tolerance δd and radius of curvature tolerance δr in the z-direction can be considered together when initializing the surface, as shown in the following equation:

[0068] d z =d z0 +δd

[0069] roc z =roc z0 +δr

[0070] Where, d z0 For the ideal design value of surface thickness, roc z0 This is the ideal design value for the radius of curvature of the surface. When initializing the surface, substitute d... z and roc z For lenses without tolerances, δd = δr = 0, at which point the surface thickness is equal to the ideal design value of the surface thickness, and the surface curvature radius is equal to the ideal design value of the surface curvature radius. Regarding the tolerances for refractive index and Abbe constant, when actually obtaining the refractive index for a specific wavelength, this embodiment adds an offset to the calculated refractive index. The formula for calculating the refractive index with the offset is shown below:

[0071] n offset (λ)=n base (λ)+n(N d +ΔN d Vd +ΔV d ,λ)-n(N d V d ,λ)

[0072] Where, n offset n is the refractive index after adding bias. base The refractive index of an ideal material is calculated using the dispersion equation specified for that material; N d V d Let D be the refractive index of the ideal material (Dlight is light with a wavelength of 587.5618 nm), Abbe constant, and ΔN, respectively. d ΔV d These represent the refractive index tolerance and Abbe constant tolerance for discrete sampling, respectively; λ is the wavelength currently being calculated.

[0073] The ray tracing process first requires setting the pupil sampling number, sampling mode, and incident angle. Next, it involves solving for the intersection of the incident ray and the surface, as well as the direction cosine of the outgoing ray after refraction. Ray tracing simulates the real-world light field by calculating the propagation paths of a large number of rays, and proper pupil sampling is crucial during the tracing process. Since the pupil in this embodiment is circular, the problem can be simplified to how to perform uniform sampling on a disk. Common sampling methods include square sampling, hexagonal sampling, and Fibonacci sampling. Because Fibonacci sampling has a smaller mean square error compared to the other two sampling modes, it is chosen for pupil sampling in this implementation. The position (x, y) of the sampling point obtained through the Fibonacci sampling method is shown. f ,y f This can be represented as:

[0074]

[0075] Where n is the number of pupil samples, i′∈{0.5,1.5,…,n-0.5}, j′∈{1,2,…,n}. To avoid undersampling causing problems in the simulation of the point spread function, the number of pupil samples is set to 128*128. The pupil sampling process is actually the initialization of the position of the ray to be traced, while the determination of the incident angle is the initialization of the direction cosine of the ray. Assuming that the angles between the incident ray and the x+ and y+ directions are α0 and β0 respectively, the direction cosine of the ray sampled at the pupil is:

[0076]

[0077] For an object at a finite distance, the incident direction vector is about the object point P. obj =(x obj ,y obj ,z objStarting from point P0 (x0, y0, z0) on the pupil and ending at point P0 (x0, y0, z0), the direction cosine of the light rays sampled at the pupil is:

[0078]

[0079] in, The magnitude of the direction cosine vector.

[0080] After completing the pupil sampling as required, the initial position P0 = (x0, y0, z0) and normalized direction vector of each ray at the entrance pupil were obtained. like Figure 4 As shown, after propagating a distance t in a certain medium, the position of the light ray is given by the following formula:

[0081] P i =P i-1 +tD i-1

[0082] Among them, P i Let P be the intersection point of the ray and the i-th surface. i-1 Let D be the intersection point of the ray and the (i-1)th face. i-1 Let be the direction cosine of the light ray after refraction through the (i-1)th surface. Therefore, the key to calculating the intersection points of the light ray with each surface of the optical system is to solve for the light ray propagation distance t. Since the system surface has many higher-order terms, directly obtaining an analytical solution is difficult; therefore, iterative algorithms are generally used to solve for numerical solutions. In the specific implementation, Newton's iteration method is used. In each iteration, t is updated, and then the updated t is substituted into the above equation to obtain the updated light ray position P. i Next, P i Substituting the surface type expression, the iteration stops when the calculated value is less than the maximum acceptable threshold, P. i This is the intersection point of the ray and the i-th surface. Furthermore, considering the effects of surface eccentricity and tilt, the position of the ray emitted from the previous surface needs to be transformed to the coordinate system of the next surface during the calculation of the intersection point. The formula for coordinate system transformation is as follows:

[0083]

[0084] Among them, X i Y i Z i Let X, Y, and Z be the coordinates of the ray in the i-th surface coordinate system; X, Y, and Z are the coordinates of the ray in the reference coordinate system; O i Let R be the eccentricity tolerance of the i-th face. i Let be the tilt tolerance of the i-th face. After calculating the intersection point of the ray and the face shape, the coordinates of the ray need to be transformed back to the optical axis coordinate system:

[0085]

[0086] Next, the refraction of light rays on the surface of the medium is calculated, thus obtaining the direction vector D of the light rays after refraction through the i-th surface. i According to Snell's law, the angle of refraction θ can be expressed as:

[0087]

[0088] Where n is the position of the surface P i The unit normal vector at η i-1 η i The refractive indices of the two sides of the i-th surface are D and D respectively. i-1 For the (i-1)th face type P i-1 The direction vector of the light ray at the location, cos<·,·> is the cosine operation for calculating the angle between vectors. After calculating the angle of refraction, the direction vector D of the light ray after refraction through the i-th surface is... i It can be calculated using the following formula:

[0089]

[0090] If there is a tolerance, the coordinate system transformation can be performed by multiplying the direction cosine by the rotation matrix, without needing to consider the effect of eccentricity. By alternately calculating the intersection point of the ray with the surface and the direction cosine after refraction through the surface, the sequential tracing of the ray can be achieved up to the exit pupil of the optical system.

[0091] The coherent superposition of light rays involves treating each ray as a wavelet and calculating the coherent superposition of these wavelets at the sampling point on the image plane, thus obtaining the point spread function value of the object point at that sampling point. By alternately calculating the intersection points of the light rays with the surface and the direction vector after refraction through the surface, the sequence of light rays can be traced up to the exit pupil plane, thereby obtaining the total optical path l of each ray from the entrance pupil to the exit pupil. e and the normalized direction cosine D at the exit pupil plane e Let P be the coordinate of the pupil plane. e =(x e ,y e ,z e The coordinates on the image plane are P. img =(x i ,y i ,z i Then the wavefront distribution W on the exit pupil plane can be expressed as:

[0092]

[0093] Where k = 2π / λ, λ is the wavelength. a0 is the amplitude of the spherical wave per unit distance. According to the wavefront decomposition principle, the wavefront of the exit pupil plane can be regarded as a collection of secondary wavelet sources. During the propagation from the exit pupil plane to the imaging plane, the complex amplitude of a certain ray can be expressed as:

[0094]

[0095] Among them, l r l is the vector from the sampling point on the exit pupil plane to the sampling point on the imaging plane. r =(x i -x e ,y i -y e ,z i -z e K(D) e ,l r The tilt factor of the wavelet (n) can be defined by the following formula:

[0096]

[0097] Where n is the normal vector of the exit pupil plane. For a point on the imaging plane, the received complex amplitude is the result of the coherent superposition of the complex amplitudes of all wavelet sources on the exit pupil plane on the imaging plane. The complex amplitude can be calculated as follows:

[0098]

[0099] Therefore, the received light intensity can be obtained by multiplying the complex amplitude by its complex conjugate:

[0100]

[0101] The above method can be used to calculate a point P on the imaging plane. img The intensity of light received at the image plane. Therefore, by traversing the sampling positions of the imaging plane and repeating the above calculation process, the intensity of light received at the image plane (x) can be obtained for the coherent point source O with wavelength λ. i ,y i The point spread function PSF(λ,x) at point ) i ,y i ).

[0102] The post-processing of the point spread function (PSF) involves superimposing the PSFs of each band within the 400nm–700nm range based on the detector's spectral response curve to obtain a three-channel PSF image. Since the captured PSF image is an RGB three-channel image, the simulated band-specific PSFs need to be superimposed using the spectral response curves to obtain the three-channel PSF image. The superposition process is shown in the following equation:

[0103]

[0104] Among them, (x i ,y i () represents different image plane positions, c refers to the three different channels R, G, and B, λ represents the wavelength, the wavelength sampling range is 400–700 nm, and the sampling interval is 20 nm. W c (λ) represents the spectral response coefficients of each wavelength in different channels after normalization.

[0105] S3: The simulated point spread function image is obtained by using the point spread function simulation model. Then, the mean square error between the real point spread function image and the simulated point spread function image is calculated and used as the loss function. After updating the tolerance in the point spread function simulation model using the loss function, the final predicted tolerance and the predicted point spread function image are obtained.

[0106] Based on the shooting steps in S1, the point spread function of the glass-plastic hybrid lens with tolerance can be obtained, but the specific tolerance value is still unknown. Therefore, in this embodiment, a simulation model of the point spread function is constructed through S2. The mean square error between the point spread function obtained from the simulation of the lens without tolerance and the point spread function of the glass-plastic hybrid lens with tolerance is used as the loss function, as shown in the following formula:

[0107]

[0108] Among them, L MSE Here, n is the total number of fields of view (13 fields of view are used in this example), c is the field of view number, and PSF is the mean squared error loss function. c This is the true point spread function image of the c-th field of view of the first tolerance glass-plastic hybrid lens. This is the simulated point spread function image of the c-th field of view obtained through the point spread function simulation model.

[0109] Next, the gradient of the loss function with respect to the tolerance terms of each surface in the point spread function simulation model is calculated using the chain rule. Then, an optimizer is selected based on the gradient and the set learning rate, and the tolerance is updated using the optimizer to ensure that the tolerance is within the preset error, thus obtaining the final predicted tolerance and the predicted point spread function image. The predicted tolerance includes the tilt of each surface about the x, y, and z axes, the eccentricity in the x and y directions, the thickness in the z direction, the radius of curvature tolerance, the refractive index tolerance of the material, and the Abbe number tolerance.

[0110] This embodiment uses an SGD optimizer. The learning rate for the tilt tolerances around the x, y, and z axes, the eccentricity tolerances in the x and y directions, and the thickness tolerance in the z direction is set to 0.01. The learning rate for the radius of curvature tolerance and the refractive index tolerance of the material is set to 0.0001. The learning rate for the Abbe number tolerance is set to 100, and the momentum parameter is 0.9. Furthermore, all optimizers use a stepped learning rate update method; when the update count is a multiple of 8, the learning rate becomes 0.9 times the original. After 20 iterations, the predicted tolerance values ​​are substituted into the original design. The resulting simulated point spread function image is highly similar to the actual point spread function image of the toleranced glass-plastic hybrid lens. Figure 5 As shown.

[0111] S4: Based on the design parameters and predicted tolerance of the first tolerance glass-plastic hybrid lens, construct the second tolerance glass-plastic hybrid lens. Then, calculate the mean square error of the predicted point spread function image and the point spread function image of the tolerance-free glass-plastic hybrid lens and use it as the loss function. Then, use the loss function to update the pose of the second tolerance glass-plastic hybrid lens to obtain the optimal compensation pose.

[0112] Specifically, the pose of the second-band tolerance glass-plastic hybrid lens is updated using a loss function to obtain the optimal compensated pose, as follows:

[0113] First, the chain rule is used to calculate the gradient of the loss function with respect to the six pose variables of the aspherical glass lens in the second tolerance glass-plastic hybrid lens. Then, the pose is updated based on this gradient to obtain the optimal compensated pose. The pose includes six pose variables corresponding to each aspherical glass lens, which are the translation and rotation of each aspherical glass lens along the x, y, and z axes.

[0114] When the design parameters of the first tolerance glass-plastic hybrid lens include tolerance, the design parameters of the glass-plastic hybrid lens without tolerance are the same as the design parameters of the first tolerance glass-plastic hybrid lens except for tolerance; when the design parameters of the first tolerance glass-plastic hybrid lens do not include tolerance, the design parameters of the glass-plastic hybrid lens without tolerance are the same as the design parameters of the first tolerance glass-plastic hybrid lens.

[0115] Due to limitations in manufacturing processes and processing costs, glass aspherical surfaces have larger manufacturing tolerances compared to plastic aspherical surfaces with the same processing cost. Lenses initially meeting imaging quality requirements but entirely composed of plastic aspherical surfaces actually experience image quality degradation upon the introduction of glass aspherical surfaces. Therefore, after obtaining the predicted tolerance values, six pose variables of the glass aspherical surface are adjusted to compensate for the image quality degradation caused by the manufacturing tolerances. The goal of compensation is to achieve the imaging effect of a tolerance-free lens. Therefore, the point spread function of each field of view of a tolerance-free hybrid glass-plastic lens is chosen as the objective function, and the mean square error between the current pose and the objective function is calculated as the loss function. According to the chain rule, the gradient of the loss function with respect to the pose variables of the glass aspherical surface to be updated can be obtained, where the translations of the pose variables to be updated along the x, y, and z axes and the rotations around the x, y, and z axes are considered. Next, SGD was selected as the optimizer, with the learning rate for translations along the x, y, and z axes set to 0.1, and the learning rate for rotations around the x, y, and z axes set to 0.001. After five iterations, the loss function value stopped decreasing, and the six pose variable values ​​obtained at this point became the final compensated pose. The process took approximately 30 seconds, demonstrating fast iteration speed and relatively low time cost. Figure 6 As shown, compared with the point spread function before compensation, the point spread function after active compensation has significantly more concentrated energy, thus improving the imaging quality.

[0116] To quantitatively verify the advancement of the method of this invention, it was applied to 10 glass-plastic hybrid lenses with random tolerances, and the MTF values ​​before and after compensation were compared. The closer the MTF value is to 1, the better the image quality of the lens. The comparison results are shown in Table 1. Table 1 shows that, after optimization by the active compensation method proposed in this invention, the MTF of all 10 tolerance-adjusted glass-plastic hybrid lenses was improved to varying degrees after compensation, i.e., the image quality was improved.

[0117] Table 1 shows the comparison of MTF before and after active compensation for 10 tolerance-adjusted hybrid glass-plastic lenses.

[0118] 1 0.5170 0.6002 2 0.6303 0.6512 3 0.6259 0.6531 4 0.5615 0.5796 5 0.5238 0.5948 6 0.4722 0.6607 7 0.3873 0.5791 8 0.5479 0.5932 9 0.6326 0.6505 10 0.4993 0.5484

[0119] S5: Adjust the pose of the glass aspherical lens in the first tolerance glass-plastic hybrid lens according to the optimal compensation pose, thereby completing the active compensation of the tolerance.

[0120] S5 specifically refers to:

[0121] A six-axis robotic arm is used to adjust the position of the aspherical glass lens to the optimal compensation position; then, ultraviolet light is used to irradiate and cure the aspherical glass lens, completing the assembly of the overall lens module and achieving active tolerance compensation.

[0122] Finally, it should be noted that the above embodiments and descriptions are only used to illustrate the technical solutions of the present invention and not to limit it. Those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the disclosure of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the protection scope of the claims of the present invention.

Claims

1. An active compensation method for manufacturing tolerances of aspherical glass in hybrid glass-plastic lenses, characterized in that, Includes the following steps: S1: Construct a first tolerance glass-plastic hybrid lens and obtain the real point spread function images of the first tolerance glass-plastic hybrid lens at different field-of-view positions; S2: Construct a point spread function simulation model based on the design parameters of the first tolerance glass-plastic hybrid lens; Specifically, S2 is: S21: The design parameters of the first tolerance glass-plastic hybrid lens are used as initialization parameters and input into the point spread function simulation model; S22: Start calculating the intersection points of light rays with the surface and the direction cosines after refraction on the point spread function simulation model, one by one, until the last surface, to obtain the intersection points of light rays with the surface and the direction cosines after refraction in all surfaces; S23: Based on the intersection points of light rays with the surface in all surface types and the direction cosines after refraction, coherent superposition of light rays is performed on the image plane to obtain point spread functions for different bands and different field positions; S24: Linearly add the point spread functions of different bands to obtain a three-channel simulated point spread function image; S3: The simulated point spread function image is obtained by using the point spread function simulation model. Then, the mean square error between the real point spread function image and the simulated point spread function image is calculated and used as the loss function. After updating the tolerance in the point spread function simulation model using the loss function, the final predicted tolerance and the predicted point spread function image are obtained. S4: Based on the design parameters and predicted tolerance of the first tolerance glass-plastic hybrid lens, construct the second tolerance glass-plastic hybrid lens. Then, calculate the mean square error of the predicted point spread function image and the point spread function image of the tolerance-free glass-plastic hybrid lens and use it as the loss function. Then, use the loss function to update the pose of the second tolerance glass-plastic hybrid lens to obtain the optimal compensation pose. S5: Adjust the pose of the glass aspherical lens in the first tolerance glass-plastic hybrid lens according to the optimal compensation pose, thereby completing the active compensation of the tolerance.

2. The active compensation method for manufacturing tolerances of aspherical glass in a glass-plastic hybrid lens according to claim 1, characterized in that, When the design parameters of the first tolerance glass-plastic hybrid lens include tolerance, the design parameters of the glass-plastic hybrid lens without tolerance are the same as the design parameters of the first tolerance glass-plastic hybrid lens except for tolerance. When the design parameters of the first tolerance glass-plastic hybrid lens do not include tolerance, the design parameters of the tolerance-free glass-plastic hybrid lens are the same as those of the first tolerance glass-plastic hybrid lens.

3. The active compensation method for manufacturing tolerances of aspherical glass in a glass-plastic hybrid lens according to claim 1, characterized in that, In step S3, after updating the tolerance in the point spread function simulation model using the loss function, the final predicted tolerance and the predicted point spread function image are obtained, specifically as follows: First, the gradient of the loss function with respect to the tolerance terms of each surface in the point spread function simulation model is calculated using the chain rule. Then, an optimizer is selected based on the gradient and learning rate, and the tolerance is updated using the optimizer to keep the tolerance within the preset error, thus obtaining the final predicted tolerance and the predicted point spread function image.

4. The active compensation method for manufacturing tolerances of aspherical glass in a glass-plastic hybrid lens according to claim 1, characterized in that, In step S4, the pose of the second tolerance glass-plastic hybrid lens is updated using a loss function to obtain the optimal compensated pose, specifically as follows: First, the chain rule is used to calculate the gradient of the loss function with respect to the six pose variables of the aspherical glass lens in the second tolerance glass-plastic hybrid lens. Then, the pose is updated based on the gradient to obtain the optimal compensation pose.

5. The active compensation method for manufacturing tolerances of aspherical glass in a glass-plastic hybrid lens according to claim 1, characterized in that, In step S1, the method for obtaining the true point spread function images of the first tolerance glass-plastic hybrid lens at different field-of-view positions includes: LED lights were installed on a black background. The first tolerance glass-plastic hybrid lens was then fixed on the black background and focused and photographed. The real point spread function images of the first tolerance glass-plastic hybrid lens at different field-of-view positions were then obtained.

6. The active compensation method for manufacturing tolerances of aspherical glass in a glass-plastic hybrid lens according to claim 1, characterized in that, In S3, the formula for calculating the loss function is: in, Let the mean squared error loss function be . The total number of fields of view. For the field of view number, For the first tolerance glass-plastic hybrid lens The true point spread function image of each field of view. The first point spread function simulation model was used to obtain the simulation result. Simulated point spread function image for each field of view.

7. The active compensation method for manufacturing tolerances of aspherical glass in a glass-plastic hybrid lens according to claim 1, characterized in that, In S3, the predicted tolerances include the tilt of each surface about the x, y and z axes, the eccentricity in the x and y directions, the thickness in the z direction, the radius of curvature tolerance, the refractive index tolerance of the material, and the Abbe number tolerance.

8. The active compensation method for manufacturing tolerances of aspherical glass in a glass-plastic hybrid lens according to claim 1, characterized in that, In S4, the pose includes six pose variables corresponding to each aspherical glass lens, which are the translation amount along the x, y and z axes and the rotation amount around the x, y and z axes for each aspherical glass lens.

9. The active compensation method for manufacturing tolerances of aspherical glass in a glass-plastic hybrid lens according to claim 1, characterized in that, Specifically, S5 is: A six-axis robotic arm is used to adjust the position of the aspherical glass lens to the optimal compensation position; then the aspherical glass lens is solidified to complete the assembly of the overall lens module, achieving active tolerance compensation.

Images