A method for designing an optical system light shield based on a bayesian optimization algorithm

Abstract

This invention relates to a method for designing optical system shields, specifically a method based on Bayesian optimization algorithms. This method addresses the technical problems of low design efficiency, cumbersome simulation processes, and difficulty in obtaining globally optimal structures in existing shield design methods. The invention first determines the shield boundary constraints based on optical system parameters and establishes a geometric intersection model. Then, it constructs a Monte Carlo stray light tracing engine integrating surface scattering calculation, light splitting control, and energy statistics modules. Using PST as the evaluation metric, the initial structure obtained through geometric construction is injected into the Bayesian optimizer as prior knowledge. The Bayesian optimization algorithm is then used to automatically optimize multi-dimensional parameters such as the number, position, tilt angle, and chamfer of the light-blocking rings globally. This invention achieves highly efficient and automated design of the shield structure, improves optimization efficiency, and achieves stray light suppression performance superior to traditional methods.

Description

Technical Field

[0001] This invention relates to a method for designing optical system light shields, specifically a method for designing optical system light shields based on a Bayesian optimization algorithm. Background Technology

[0002] With the rapid development of optical imaging technology, high-performance optical systems are increasingly widely used in aerospace, automotive lidar, high-end photography, and medical imaging. In these systems, stray light is one of the key factors that seriously affects image quality. Stray light refers to unexpected light signals arriving at the image plane of an optical system. It creates background noise, reduces the contrast and signal-to-noise ratio of the image plane, and in severe cases, the target signal may be completely obscured by stray light, causing the optical system to fail in imaging or detection.

[0003] A lens hood is a core structural component for suppressing stray light from outside the optical system. By placing a lens hood and its internal light-blocking ring structure at the front end of the optical system, stray light from outside the field of view can be effectively blocked or attenuated from entering the optical system, thereby improving image quality. Therefore, the structural design of the lens hood is a crucial aspect of optical system design.

[0004] Currently, the conventional design method for light shields mainly includes the following steps: First, optical design engineers, based on the field of view, aperture, and stray light suppression angle of the optical system, use geometric drawing methods to initially determine the length, entrance diameter, and the position, height, and number of light-blocking rings of the light shield, obtaining the initial structure of the light shield; then, a three-dimensional model is established, and stray light analysis is performed using commercial optical simulation software such as TracePro and LightTools to calculate PST (Point Source Transmittance) to evaluate the stray light suppression performance of the light shield; if the analysis results cannot meet the suppression index, the initial structure needs to be manually modified, and the model re-analyzed, repeating this iterative process until the design requirements are met. However, this technical solution still has the following shortcomings:

[0005] First, the design efficiency is low. The light shield structure has numerous parameters, including its length, inlet diameter, number of light-blocking rings, axial position, height, tilt angle, and chamfer angle of each ring. These parameters are interconnected, resulting in a vast and non-convex design space. Traditional geometric construction methods can only provide an initial structure; subsequent optimization heavily relies on engineers' experience for repeated trial and error and simulation analysis. This makes it difficult to achieve global optimization within a limited timeframe, leading to long design cycles and low efficiency.

[0006] Secondly, the simulation process of commercial optical simulation software is cumbersome. Although existing commercial optical simulation software has stray light analysis capabilities, the optimization process requires repeated manual geometric modeling, parameter adjustment, simulation settings, and result extraction. It cannot achieve automatic closed-loop iteration of structural parameters and simulation evaluation, thus limiting the improvement of optimization efficiency.

[0007] Third, it is difficult to obtain the globally optimal structure. Due to the high dimensionality and non-convexity of the design space of the light shield, traditional manual parameter tuning methods are prone to getting stuck in local optima and it is difficult to explore unconventional structural forms (such as nonlinear curved surface profiles, variable tilt angle light-blocking rings, etc.), resulting in the inability to obtain the globally optimal light shield structure. Summary of the Invention

[0008] The purpose of this invention is to solve the technical problems of low design efficiency, cumbersome simulation process, and difficulty in obtaining the globally optimal structure in existing light shield design methods, and to provide a light shield design method for optical systems based on Bayesian optimization algorithm.

[0009] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0010] A method for designing a light shield for an optical system based on a Bayesian optimization algorithm, characterized by the following steps:

[0011] S1. Determine the boundary constraints of the light shield based on the design specifications of the optical system; the design specifications include the stray light suppression angle α, the half field of view ω, and the aperture D of the optical system; the boundary constraints include the length L and the inlet diameter D1 of the light shield.

[0012] S2. Establish a coordinate system with the center of the light shield entrance as the origin. Based on the boundary constraints, use the geometric drawing method to solve for the radial width H, number, and axial position of the light-blocking rings to obtain the initial structure of the light shield.

[0013] S3. Construct a geometric intersection model of the light rays and the initial structure of the light shield;

[0014] S4. Based on the geometric intersection model constructed in step S3, a Monte Carlo stray light tracing engine is established. The Monte Carlo stray light tracing engine is used to perform high-precision physical simulation of the propagation process of light in the light shield and optical system.

[0015] S5. The initial structure of the light shield obtained in step S2 is injected into the Bayesian optimizer as prior knowledge; the parameters of the initial structure of the light shield are parametrically modeled to construct a high-dimensional design space. Based on the Monte Carlo stray light tracing engine, the optimal light shield structure parameters are obtained and output through the Bayesian optimization algorithm in the high-dimensional design space, thus completing the design of the optical system light shield based on the Bayesian optimization algorithm.

[0016] Further, in step S1, the formula for calculating the length L of the light shield is:

[0017] ;

[0018] In step S2, the formula for calculating the radial width H of the light-blocking ring is:

[0019] ;

[0020] In step S4, the Monte Carlo stray light tracing engine integrates a surface scattering calculation module, a light splitting control module, and an energy statistics module.

[0021] Furthermore, step S3 specifically includes:

[0022] S3.1. Abstract the light rays into parameterized three-dimensional spatial rays, and convert the outer cylindrical wall of the light shield and the array of light-blocking rings with physical thickness into implicit equations of spatial analytic geometry.

[0023] S3.2 Substitute the implicit equations of spatial analytic geometry into the quadratic surface and plane equations of the light shield, respectively, solve the candidate intersection distances through matrix operations, and perform strict filtering by combining the radial distance of the intersection points and the radial width H of the light-blocking ring to eliminate invalid intersection points that exceed the boundaries, thereby obtaining the effective intersection data between the light rays and the light shield, so as to construct the geometric intersection model between the light rays and the initial structure of the light shield.

[0024] Furthermore, step S5 specifically includes:

[0025] S5.1 Inject the initial structure of the light shield obtained in step S2 as prior knowledge into the Bayesian optimizer.

[0026] S5.2. Parametrically model the parameters of the initial structure of the light shield to construct a high-dimensional design space; the parameters of the initial structure of the light shield include the number of light-blocking rings, the axial position ratio of each light-blocking ring, the inner and outer diameters of the light-blocking rings, the tilt angle of the light-blocking rings, and the chamfer angle of the light-blocking ring edges.

[0027] S5.3 Select multiple representative off-axis angles, perform ray tracing process through the Monte Carlo stray ray tracing engine, calculate PST at each off-axis angle, and obtain PST at all off-axis angles;

[0028] S5.4 Select the largest PST in step S5.3 as the objective function value, and set geometric occlusion constraints according to the field of view of the optical system.

[0029] S5.5. The objective function is probabilistically modeled using the Bayesian optimization algorithm. The optimization is iteratively performed in the high-dimensional design space. During each iteration, the Monte Carlo stray light tracing engine established in step S4 is called to evaluate the candidate structure and update the probabilistic model until the preset number of iterations or the convergence condition is reached. The optimal light shield structure parameters are obtained and output, and the design of the optical system light shield based on the Bayesian optimization algorithm is completed.

[0030] Furthermore, step S5.3 specifically includes:

[0031] S5.3.1 Select several representative off-axis angles, and then select one of the off-axis angles in turn;

[0032] S5.3.2 Based on a selected off-axis angle, a virtual circular planar light source is constructed outside the entrance of the light shield, and the normal incident direction of the circular planar light source is adjusted according to the off-axis angle to generate a random ray array with initial ray energy weights; then, the generated random ray array is used as the input of the geometric intersection model constructed in step S3, and the geometric intersection model is called to calculate the intersection point of the ray and the light shield structure.

[0033] S5.3.3 After obtaining the intersection point of the light with the wall of the light shield or the surface of the light-blocking ring, the surface scattering calculation module in the Monte Carlo stray light tracing engine is called to calculate the total integral scattering rate TIS based on the ABg two-way scattering distribution function model, and the light energy weight is dynamically updated at the intersection point according to the incident angle.

[0034] S5.3.4 When the light intersects with the lens surface, the light splitting control module in the Monte Carlo stray light tracing engine is invoked to calculate the reflectivity and transmittance based on the Fresnel equation, execute the light splitting, dynamically update the light energy weight, and set the energy threshold to control the number of splits.

[0035] S5.3.5 After all rays have been traced, the energy statistics module in the Monte Carlo stray tracing engine is called to calculate the remaining ray energy weight of all rays that reach the image plane, obtain the total energy received by the image plane, and combine the total number of rays emitted by the circular planar light source and the area ratio of the light source surface to the image plane to calculate the PST at that off-axis angle.

[0036] S5.3.6 Return to step S5.3.2, select the next off-axis angle, and continue until the PST for all off-axis angles is obtained.

[0037] Further, in step S5.3.3, the ABg two-way scattering distribution function model (BSDF) is expressed as:

[0038]

[0039] In the formula, ρ is the spatial distance between the scattering direction and the specular reflection direction; A, B, and g are the scattering parameters of the coating material; θ i φ i θ represents the polar angle and azimuth angle of the incident light, respectively, in degrees; s φ s These represent the polar angle and azimuth angle of the scattered light, respectively, in degrees;

[0040] The formula for calculating the spatial distance ρ between the scattering direction and the specular reflection direction is:

[0041] .

[0042] Further, in step S5.3.3, the update formula for the light energy weight is:

[0043] ;

[0044] In the formula, The light energy weights before the update; For the updated light energy weights;

[0045] The formula for calculating the total integrated scattering rate (TIS) is as follows:

[0046] ;

[0047] In the formula, dθ s The polar angle θ of the scattered light s The differential element represents the minute change in the scattered light along the polar angle; dφ s The azimuth angle φ of the scattered light s The differential element represents the minute change in the azimuth direction of the scattered light.

[0048] Further, in step S5.3.5, the calculation formula for PST is:

[0049] ;

[0050] In the formula, P received The total energy received by the image plane; N total R represents the total number of emitted rays. src R is the radius of the circular planar light source surface, in mm; det The radius of the image receiving area is in mm.

[0051] Further, in step S5.4, the expression for the objective function value is:

[0052] ;

[0053] In the formula, β1, β2, ..., β n For the selected off-axis angles, in degrees; PST(β1), PST(β2), ..., PST(β... n ) is the off-axis angle β1, β2, ..., β n The calculated PST is as follows.

[0054] Furthermore, in step S5.2, the optimized range of the number of light-blocking rings is 3 to 10;

[0055] The optimized range of the axial position ratio of each light-blocking ring is 0.02~0.98;

[0056] The optimized range of the chamfer angle of the cutting edge is 15°~35°;

[0057] In step S5.3.1, the plurality of off-axis angles are β = {20°, 30°, 40°, 50°, 60°, 70°, 80°};

[0058] In step S5.3.4, the energy threshold is 1×10⁻⁶. -10 When the energy weight of the split ray is lower than this threshold, further tracing of that ray is terminated.

[0059] In step S5.5, the preset number of iterations is 50.

[0060] The Bayesian optimization algorithm described above is based on tree structure probability estimation.

[0061] Compared with the prior art, the present invention has the following beneficial technical effects:

[0062] 1. A method for designing optical system hoods based on Bayesian optimization algorithms. By coupling the Bayesian optimization algorithm with a physical ray tracing engine, and using hood structural parameters as optimization variables and stray light evaluation indicators as the objective function, this method achieves efficient and automatic optimization of hood parameters, facilitating the search for optimal solutions globally. Compared to design methods that rely on manual experience for parameter adjustment, this method can automatically optimize under high-dimensional parameter conditions, thereby improving hood design efficiency.

[0063] 2. A design method for optical system shading based on Bayesian optimization algorithm. By constructing an automated ray tracing and evaluation process, the automatic solution and iterative feedback of point source transmittance are realized, which reduces the process of manual repetitive modeling and parameter adjustment in traditional commercial optical simulation software, thereby improving the computational efficiency of stray light simulation.

[0064] 3. An optical system light shield design method based on Bayesian optimization algorithm can also jointly optimize the structural parameters of the light-blocking ring according to the characteristics of different light-absorbing materials. It can also incorporate parameters such as the tilt angle of the light-blocking ring, the chamfer angle of the blade edge, and the nonlinear curved surface profile into the optimization variables, realize the automatic exploration and design expansion of non-traditional light shield structure forms, and improve the flexibility and innovation space of light shield structure design. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating an embodiment of an optical system light shield design method based on Bayesian optimization algorithm according to the present invention.

[0066] Figure 2 This is a schematic diagram of the boundary constraint conditions of the light shield in step S1 of this embodiment of the invention;

[0067] Figure 3 This is a comparison chart of the stray light tracing engine simulation results established in step S4 of this embodiment and the results of the stray light simulation software TracePro.

[0068] Figure 4 This is a flowchart illustrating step S5 in an embodiment of the present invention;

[0069] Figure 5 This is a comparison chart of the PST curves of the light shield obtained by Bayesian algorithm optimization in step S5 of this embodiment of the invention and the light shield obtained by the graphical method. Detailed Implementation

[0070] To make the objectives, advantages, and features of this invention clearer, the following detailed description of a light shield design method for an optical system based on a Bayesian optimization algorithm, in conjunction with the accompanying drawings and specific embodiments, is provided. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of this invention and are not intended to limit the scope of protection of this invention.

[0071] A method for designing optical system light shields based on Bayesian optimization algorithms, such as... Figure 1 As shown, it includes the following steps:

[0072] S1. Determine the boundary constraints of the light shield.

[0073] The boundary constraints of the light shield are determined based on the design specifications of the optical system; these specifications include the stray light suppression angle α, the half-field angle ω, and the aperture D of the optical system; such as Figure 2 As shown, the boundary constraints of the light shield include the length L of the light shield and the inlet diameter D1.

[0074] Based on geometric relationships, the formula for calculating the length L of the light shield is:

[0075] ;

[0076] In the formula, D is the aperture of the optical system, in mm; α is the stray light suppression angle of the optical system, in degrees; and ω is the half-field angle of the optical system, in degrees. In this embodiment, the design specifications of the optical system are: stray light suppression angle α = 20°, half-field angle ω = 8°, aperture D = 40 mm, and the maximum outer envelope size of the light shield is set to 200 mm × 120 mm. To suppress stray light to the greatest extent, the inlet diameter of the light shield is set to D1 = 120 mm, and the light shield adopts a conical structure.

[0077] Based on the formula above, the length L of the sunshade is approximately 180mm.

[0078] S2. Obtain the initial structure of the light shield.

[0079] Based on the design specifications in step S1, a coordinate system is established with the center of the light shield entrance as the origin. Based on the boundary constraints, the radial width H, number, and axial position of the light-blocking rings are solved using a geometric drawing method to obtain the initial structure of the light shield.

[0080] The formula for calculating the radial width H of the light-blocking ring is as follows:

[0081] .

[0082] According to the above formula, when L=180mm, the height H of the light-blocking ring is 13.42mm, and the thickness of the light-blocking ring is 1mm.

[0083] In this embodiment, the traditional geometric drawing method is written into a Python code script, and after running it, the initial arrangement of 6 light-blocking rings is obtained, as shown in Table 1.

[0084] Table 1. Examples of initial structures of light-blocking rings obtained by geometric construction method.

[0085]

[0086] S3. Construct a geometric intersection model between the ray and the initial structure of the light shield; specifically:

[0087] S3.1. Abstract the light rays into parameterized three-dimensional spatial rays, and convert the outer cylindrical wall of the light shield and the array of light-blocking rings with physical thickness into implicit equations of spatial analytic geometry.

[0088] S3.2 Substitute the implicit equations of spatial analytic geometry into the quadratic surface and plane equations of the light shield, respectively, solve the candidate intersection distances through matrix operations, and perform strict filtering by combining the radial distance of the intersection points and the radial width H of the light-blocking ring to eliminate invalid intersection points that exceed the boundaries, thereby obtaining the effective intersection data between the light rays and the light shield, so as to construct the geometric intersection model between the light rays and the initial structure of the light shield.

[0089] Here, the light ray is abstractly defined as the parameterized three-dimensional spatial ray equation as follows:

[0090] ;

[0091] In the formula, P(t) is the endpoint position of the light ray after propagating a distance t, which is a coordinate in a three-dimensional space; O is the coordinate of the starting point of the light ray; Q is the unit vector of the light ray direction; and t is the propagation distance along the light ray direction.

[0092] S4. Establish a Monte Carlo stray light tracing engine and calculate PST.

[0093] Based on the geometric intersection model constructed in step S3, a Monte Carlo stray light tracing engine is established. The Monte Carlo stray light tracing engine is used to perform high-precision physical simulation of the propagation process of light in the shade and optical system. Figure 3 A comparison of the simulation results of the stray light tracing engine established for this invention with the results of the stray light simulation software TracePro shows that the simulation results of the two are almost identical.

[0094] The Monte Carlo stray tracing engine integrates a surface scattering calculation module, a ray splitting control module, and an energy statistics module. The surface scattering calculation module calculates the energy attenuation of light rays scattered from the surface of the lens based on the ABg two-way scattering distribution function model; the ray splitting control module handles the reflection and refraction splitting of light rays at the lens surface based on the Fresnel equation; and the energy statistics module counts the energy of light rays reaching the image plane and calculates the PST.

[0095] S5, Bayesian optimization of light shield structure parameters

[0096] The initial structure of the light shield obtained in step S2 is injected into the Bayesian optimizer as prior knowledge. The parameters of the initial light shield structure are parametrically modeled to construct a high-dimensional design space. Based on a Monte Carlo stray light tracing engine, the Bayesian optimization algorithm iteratively searches for optimal parameters in the high-dimensional design space, obtaining and outputting the optimal light shield structure parameters, thus completing the design of the optical system light shield based on the Bayesian optimization algorithm. Specifically, as follows... Figure 4 As shown:

[0097] S5.1, Prior Structure Initialization

[0098] The initial structure of the light shield obtained in step S2 is injected into the Bayesian optimizer as prior knowledge.

[0099] In this embodiment, the axial positions of the six light-blocking rings obtained by geometric drawing in step S2 are injected into the Bayesian optimizer as prior knowledge, providing a reasonable initial search area for the optimization process and reducing the number of calculations in the random exploration phase.

[0100] In other embodiments, other parameters in the initial structure of the light shield may also be injected into the Bayesian optimizer as prior knowledge.

[0101] S5.2 Optimization of Variable Parametric Modeling

[0102] Parametric modeling of the parameters of the initial structure of the light shield is performed to construct a high-dimensional design space. The parameters of the initial structure of the light shield include the length L of the light shield, the inlet diameter D1, the number of light-blocking rings, the axial position ratio of each light-blocking ring, the inner and outer diameters of the light-blocking rings, the tilt angle of the light-blocking rings, and the chamfer angle of the light-blocking ring blades.

[0103] The optimized range for the number of light-blocking rings is 3 to 10.

[0104] The optimized range for the axial position ratio of each light-blocking ring is 0.02 to 0.98 (relative to the total length of the light shield).

[0105] The inner and outer diameters of the light-blocking ring are automatically generated by geometric constraints. The optimized range of the light-blocking ring tilt angle is 0° to 30°. The optimized range of the light-blocking ring chamfer angle is 15° to 35°.

[0106] S5.3 Constructing a multi-off-axis angle objective function

[0107] Several representative off-axis angles are selected, and a ray tracing process is performed using a Monte Carlo stray ray tracing engine to calculate the PST at each off-axis angle, thus obtaining the PST for all off-axis angles; specifically:

[0108] S5.3.1 Select multiple representative off-axis angles, and select one of them in sequence; in this embodiment, the multiple off-axis angles are β={20°, 30°, 40°, 50°, 60°, 70°, 80°}.

[0109] In other embodiments, other suitable angles may also be selected.

[0110] S5.3.2. Based on a selected off-axis angle, a virtual circular planar light source with a radius of 90mm is constructed outside the entrance of the light shield to ensure that the light source surface of the circular planar light source can completely cover the entrance of the light shield under different off-axis angles. The normal incident direction of the circular planar light source is adjusted according to the off-axis angle to generate a random ray array with an initial ray energy weight of 1. Subsequently, the generated random ray array is used as the input of the geometric intersection model constructed in step S3, and the geometric intersection model is called to calculate the intersection point of the ray and the light shield structure to start the ray tracing process.

[0111] S5.3.3 After obtaining the intersection point of the light ray with the wall of the light shield or the surface of the light-blocking ring, call the surface scattering calculation module in the Monte Carlo stray light tracing engine, calculate the total integral scattering rate (TIS) based on the ABg two-way scattering distribution function model, and dynamically update the light energy weight at the intersection point according to the incident angle.

[0112] The ABg two-way scattering distribution function model (BSDF) is expressed as:

[0113]

[0114] In the formula, ρ is the spatial distance between the scattering direction and the specular reflection direction; A, B, and g are the scattering parameters of the coating material; θ i φ i θ represents the polar angle and azimuth angle of the incident light, respectively, in degrees; s φ s These represent the polar angle and azimuth angle of the scattered light, respectively, in degrees.

[0115] The formula for calculating the spatial distance ρ between the scattering direction and the specular reflection direction is:

[0116] .

[0117] Different surface coating materials correspond to different combinations of scattering parameters. In this embodiment, the scattering parameters of the matting coating on the inner wall of the light shield are: A=0.001, B=0.015, g=2.

[0118] To quantify the energy loss caused by surface scattering, the total integrated scattering rate (TIS) is calculated based on the BSDF model described above. The formula for calculating the total integrated scattering rate (TIS) is as follows:

[0119] ;

[0120] In the formula, dθ s The polar angle θ of the scattered light s The differential element represents the minute change in the scattered light along the polar angle; dφ s The azimuth angle φ of the scattered light sThe differential element represents the minute change in the azimuth direction of the scattered light. The integration region is the range of the exiting hemisphere.

[0121] During ray tracing, the ray energy weights are updated according to the following steps:

[0122] (1) When the light ray intersects the wall of the light shield or the surface of the light-blocking ring, calculate the polar angle θ of the incident light at the intersection point. i ;

[0123] (2) Extract the corresponding scattering energy ratio based on the pre-established table of the polar angle of incident light and the total integrated scattering rate (TIS).

[0124] (3) Using the total integrated scattering rate (TIS) as the energy attenuation factor, the weight of the light energy carried by the current light is updated. The formula for updating the light energy weight is:

[0125] ;

[0126] In the formula, The light energy weights before the update; This is the updated light energy weight.

[0127] S5.3.4 When a ray intersects the lens surface, the ray splitting control module in the Monte Carlo stray tracing engine is invoked. Based on the Fresnel equation, reflectivity and transmittance are calculated, ray splitting is performed, ray energy weights are dynamically updated, and energy thresholds are set to control the number of splits.

[0128] In this embodiment, the optical system has one lens, and the refractive index of the lens material is 1.5. When light is incident on the lens surface, the reflectivity and transmittance are calculated according to Fresnel's equations, and light splitting is performed: one part of the light propagates in the direction of specular reflection, while the other part of the light propagates into the lens according to the law of refraction, and its light energy weight is updated according to the transmittance.

[0129] In practical optical systems, light rays may undergo multiple reflections and refractions within a lens. Tracing all these split rays would cause the number of rays to increase rapidly with propagation, significantly increasing computational complexity. To avoid this exponential increase in ray count due to multiple reflections within the lens, an energy threshold of 1 × 10⁻⁶ is set. -10 When the energy weight of the split ray falls below this threshold, further tracing of that ray is terminated.

[0130] S5.3.5 After all rays have been traced, the energy statistics module in the Monte Carlo stray tracing engine is called to calculate the remaining ray energy weights of all rays reaching the image plane, obtain the total energy received by the image plane, and combine the total number of rays emitted by the circular planar light source and the area ratio of the light source surface to the image plane to calculate the PST at that off-axis angle.

[0131] In this embodiment, the image plane radius is 20mm. After all rays have been traced, the effective rays reaching the image plane are counted, and the weight of the remaining ray energy carried by each ray is accumulated to obtain the total energy P received by the image plane. received The calculation formula is as follows:

[0132] ;

[0133] In the formula, Let be the weight of the remaining light energy of the j-th effective ray after it has completed all propagation and scattering processes.

[0134] According to the radiometric definition, PST is defined as the average irradiance E on the image plane. det Irradiance E perpendicular to the entrance surface of the light source inc The ratio, that is:

[0135] ;

[0136] Based on the defined light source model parameters, and assuming the total number of emitted rays in the system is N. total After area normalization, PST can be expressed as a calculation formula based on energy statistics:

[0137] ;

[0138] In the formula, P received The total energy received by the image plane; N total R represents the total number of emitted rays. src R is the radius of the circular planar light source surface, in mm; det The radius of the image receiving area is in mm.

[0139] S5.3.6 Return to step S5.3.2, select the next off-axis angle, and continue until the PST for all off-axis angles is obtained.

[0140] Among them, multiple off-axis angles are β={20°, 30°, 40°, 50°, 60°, 70°, 80°}.

[0141] S5.4 Select the largest PST in step S5.3 as the objective function value, and set geometric occlusion constraints according to the field of view of the optical system to ensure that the structure generated during the optimization process will not block the effective imaging beam.

[0142] The expression for the objective function value is:

[0143] ;

[0144] In the formula, β1, β2, ..., β n For the selected off-axis angles, in degrees; PST(β1), PST(β2), ..., PST(β... n ) is the off-axis angle β1, β2, ..., β n The calculated PST is as follows.

[0145] S5.5, Bayesian Optimized Iterative Search

[0146] The objective function is probabilistically modeled using a Bayesian optimization algorithm based on tree structure probability estimation. Iterative optimization is performed in a high-dimensional design space. During each iteration, the Monte Carlo stray light tracing engine established in step S4 is invoked to evaluate candidate structures and update the probabilistic model until the preset number of iterations (50) or the convergence condition is met. The optimal light shield structure parameters are then obtained and output, completing the design of the optical system light shield based on the Bayesian optimization algorithm. Examples of the light-blocking ring structure parameters obtained by the Bayesian optimization algorithm are shown in Table 2.

[0147] Table 2 Example of light-blocking ring structure parameters

[0148]

[0149] Figure 5 The PST curves of the light shield obtained by the Bayesian optimization algorithm and the light shield obtained by the geometric construction method are compared. As can be seen from the figure, the light shield optimized using the method of this invention exhibits superior stray light suppression performance across the entire off-axis angle range, verifying the effectiveness of this invention.

[0150] In other embodiments, the Monte Carlo stray light tracing engine can also be coupled with other optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.); the ABg scattering model parameters can be adjusted according to the actual coating material; the value range of the optimization variables can be adaptively modified according to the specific optical system indicators.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present invention.

Claims

1. A method for designing a light shield for an optical system based on a Bayesian optimization algorithm, characterized in that, Includes the following steps: S1. Determine the boundary constraints of the light shield based on the design specifications of the optical system; the design specifications include the stray light suppression angle α, the half field of view ω, and the aperture D of the optical system; the boundary constraints include the length L and the inlet diameter D1 of the light shield. S2. Establish a coordinate system with the center of the light shield entrance as the origin. Based on the boundary constraints, use the geometric drawing method to solve for the radial width H, number, and axial position of the light-blocking rings to obtain the initial structure of the light shield. S3. Construct a geometric intersection model of the light rays and the initial structure of the light shield; S4. Based on the geometric intersection model constructed in step S3, a Monte Carlo stray light tracing engine is established. The Monte Carlo stray light tracing engine is used to perform high-precision physical simulation of the propagation process of light in the light shield and optical system. S5. Inject the initial structure of the light shield obtained in step S2 as prior knowledge into the Bayesian optimizer; perform parametric modeling of the parameters of the initial structure of the light shield to construct a high-dimensional design space; based on the Monte Carlo stray light tracing engine, iteratively optimize in the high-dimensional design space using the Bayesian optimization algorithm to obtain and output the optimal light shield structure parameters, thus completing the design of the optical system light shield based on the Bayesian optimization algorithm; specifically: S5.1 Inject the initial structure of the light shield obtained in step S2 as prior knowledge into the Bayesian optimizer. S5.

2. Parametrically model the parameters of the initial structure of the light shield to construct a high-dimensional design space; the parameters of the initial structure of the light shield include the number of light-blocking rings, the axial position ratio of each light-blocking ring, the inner and outer diameters of the light-blocking rings, the tilt angle of the light-blocking rings, and the chamfer angle of the light-blocking ring edges. S5.3 Select multiple representative off-axis angles, perform ray tracing process through the Monte Carlo stray ray tracing engine, calculate PST at each off-axis angle, and obtain PST at all off-axis angles; S5.4 Select the largest PST in step S5.3 as the objective function value, and set geometric occlusion constraints according to the field of view of the optical system. S5.

5. The objective function is probabilistically modeled using the Bayesian optimization algorithm. The optimization is iteratively performed in the high-dimensional design space. During each iteration, the Monte Carlo stray light tracing engine established in step S4 is called to evaluate the candidate structure and update the probabilistic model until the preset number of iterations or the convergence condition is reached. The optimal light shield structure parameters are obtained and output, and the design of the optical system light shield based on the Bayesian optimization algorithm is completed.

2. The optical system light shield design method based on Bayesian optimization algorithm according to claim 1, characterized in that, In step S1, the formula for calculating the length L of the light shield is: ； In step S2, the formula for calculating the radial width H of the light-blocking ring is: ； In step S4, the Monte Carlo stray light tracing engine integrates a surface scattering calculation module, a light splitting control module, and an energy statistics module.

3. The optical system light shield design method based on Bayesian optimization algorithm according to claim 2, characterized in that, Step S3 is as follows: S3.

1. Abstract the light rays into parameterized three-dimensional spatial rays, and convert the outer cylindrical wall of the light shield and the array of light-blocking rings with physical thickness into implicit equations of spatial analytic geometry. S3.2 Substitute the implicit equations of spatial analytic geometry into the quadratic surface and plane equations of the light shield, respectively, solve the candidate intersection distances through matrix operations, and perform strict filtering by combining the radial distance of the intersection points and the radial width H of the light-blocking ring to eliminate invalid intersection points that exceed the boundaries, thereby obtaining the effective intersection data between the light rays and the light shield, so as to construct the geometric intersection model between the light rays and the initial structure of the light shield.

4. The optical system light shield design method based on Bayesian optimization algorithm according to claim 3, characterized in that, Step S5.3 specifically includes: S5.3.1 Select several representative off-axis angles, and then select one of the off-axis angles in turn; S5.3.2 Based on a selected off-axis angle, a virtual circular planar light source is constructed outside the entrance of the light shield, and the normal incident direction of the circular planar light source is adjusted according to the off-axis angle to generate a random light array with initial light energy weights. Subsequently, the generated random ray array is used as input to the geometric intersection model constructed in step S3, and the geometric intersection model is called to calculate the intersection point of the ray and the light shield structure; S5.3.3 After obtaining the intersection point of the light with the wall of the light shield or the surface of the light-blocking ring, the surface scattering calculation module in the Monte Carlo stray light tracing engine is called to calculate the total integral scattering rate TIS based on the ABg two-way scattering distribution function model, and the light energy weight is dynamically updated at the intersection point according to the incident angle. S5.3.4 When the light intersects with the lens surface, the light splitting control module in the Monte Carlo stray light tracing engine is invoked to calculate the reflectivity and transmittance based on the Fresnel equation, execute the light splitting, dynamically update the light energy weight, and set the energy threshold to control the number of splits. S5.3.5 After all rays have been traced, the energy statistics module in the Monte Carlo stray tracing engine is called to calculate the remaining ray energy weight of all rays that reach the image plane, obtain the total energy received by the image plane, and combine the total number of rays emitted by the circular planar light source and the area ratio of the light source surface to the image plane to calculate the PST at that off-axis angle. S5.3.6 Return to step S5.3.2, select the next off-axis angle, and continue until the PST for all off-axis angles is obtained.

5. The optical system light shield design method based on Bayesian optimization algorithm according to claim 4, characterized in that, In step S5.3.3, the ABg two-way scattering distribution function model (BSDF) is expressed as: In the formula, ρ is the spatial distance between the scattering direction and the specular reflection direction; A, B, and g are the scattering parameters of the coating material; θ i φ i θ represents the polar angle and azimuth angle of the incident light, respectively, in degrees; s φ s These represent the polar angle and azimuth angle of the scattered light, respectively, in degrees; The formula for calculating the spatial distance ρ between the scattering direction and the specular reflection direction is: 。 6. The optical system light shield design method based on Bayesian optimization algorithm according to claim 4, characterized in that, In step S5.3.3, the update formula for the light energy weight is: ； In the formula, The light energy weights before the update; For the updated light energy weights; The formula for calculating the total integrated scattering rate (TIS) is as follows: ； In the formula, dθ s The polar angle θ of the scattered light s The differential element represents the minute change in the scattered light along the polar angle; dφ s The azimuth angle φ of the scattered light s The differential element represents the minute change in the azimuth direction of the scattered light.

7. The optical system light shield design method based on Bayesian optimization algorithm according to claim 5 or 6, characterized in that, In step S5.3.5, the formula for calculating PST is: ； In the formula, P received N represents the total energy received by the image plane. total R represents the total number of emitted rays. src R is the radius of the circular planar light source surface, in mm; det The radius of the image receiving area is in mm.

8. The optical system light shield design method based on Bayesian optimization algorithm according to claim 7, characterized in that, In step S5.4, the expression for the objective function value is: ； In the formula, β1, β2, ..., β n For the selected off-axis angles, in degrees; PST(β1), PST(β2), ..., PST(β... n ) is the off-axis angle β1, β2, ..., β n The calculated PST is as follows.

9. The optical system light shield design method based on Bayesian optimization algorithm according to claim 8, characterized in that: In step S5.2, the optimized range of the number of light-blocking rings is 3 to 10; The optimized range of the axial position ratio of each light-blocking ring is 0.02~0.98; The optimized range of the chamfer angle of the cutting edge is 15°~35°; In step S5.3.1, the plurality of off-axis angles are β = {20°, 30°, 40°, 50°, 60°, 70°, 80°}; In step S5.3.4, the energy threshold is 1×10⁻⁶. -10 When the energy weight of the split ray is lower than this threshold, further tracing of that ray is terminated. In step S5.5, the preset number of iterations is 50. The Bayesian optimization algorithm described above is based on tree structure probability estimation.

Images