An efficient and differentiable method for optical system degradation simulation
Through the differential operator and automatic differentiation method of the point spread function, the problem of high computational overhead of existing optical degradation simulation in small-sized optical systems is solved, and efficient and accurate optical degradation simulation is achieved, which is suitable for the joint optimization design of optical systems.
Patent Information
- Application Number
- CN202411861071.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-09-16
- Estimated Expiration
- 2044-12-17
AI Technical Summary
Existing differentiable optical degradation simulation methods have high computational overhead in small-sized, diffraction-limited optical systems, making them difficult to apply to mobile terminal modules. In addition, existing methods have errors in handling coherence and diffraction effects.
The differential operator of the point spread function is used to simulate the image quality degradation of the optical system. The entrance pupil is obtained through the paraxial optical method, and uniform sampling and coherent complex amplitude processing are performed. The automatic differentiation method is combined with gradient back propagation to optimize the parameters of the point spread function.
It achieves efficient and accurate optical degradation simulation in small-aperture, diffraction-limited optical systems, reduces video memory overhead, and is suitable for joint optical and numerical optimization tasks.
Smart Images
Figure CN119596544B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of optical simulation, and in particular relates to an efficient and differentiable optical system degradation simulation method. Background Art
[0002] In recent years, computational optics has bridged the gap between optical design and downstream image post-processing algorithms, using a combination of hardware and software to find optimal solutions for imaging tasks, rather than simply being constrained by optimal optical imaging quality or the most suitable image post-processing algorithm. However, differentiable degradation simulation of optical systems is currently limited to geometric aberrations, making it difficult to apply to mobile terminal modules that are restricted by diffraction effects. Furthermore, the memory space of graphics cards limits the computational overhead of differentiable optical degradation simulation. Therefore, new differentiable optical degradation simulation methods are urgently needed to address this problem.
[0003] Existing simulation methods for differentiable optical degradation can be roughly summarized into two types: (1) Incoherent degradation simulation based on geometric aberration. It approximates the point spread function of the optical system as a light point diagram and renders the optical degradation by incoherent superposition. However, since it ignores the coherence between light rays and the diffraction effect of small-aperture systems, this type of method is only applicable to optical systems with limited geometric aberrations, and the analysis results of small-size, diffraction-limited systems have large deviations. (2) Coherent degradation simulation based on exit pupil diffraction. It performs diffraction calculations based on the optical path difference on the exit pupil surface and can accurately model the degraded point spread function caused by the diffraction effect of the optical system. However, this method has a large memory overhead in the differentiable calculation process and is difficult to apply to the optimization task of solving multiple point spread functions. Summary of the Invention
[0004] In order to overcome the shortcomings of the existing technology and solve the problems existing in the background technology, the present invention proposes an efficient and differentiable optical system degradation simulation method. The present invention uses the differential operator of the point spread function to accurately realize the image quality degradation simulation of the optical system, and realizes gradient backpropagation with low computing power overhead. It is widely applicable to the joint light-number optimization tasks of general optical systems.
[0005] The technical solution adopted in the present invention is as follows:
[0006] An efficient and differentiable optical system degradation simulation method includes the following steps:
[0007] S1. In computer optical simulation software, obtain the optical surface, optical medium, and object plane of the optical system, and use the paraxial optical method to perform conjugate processing on the optical system based on the optical surface and optical medium of the optical system to obtain the entrance pupil of the optical system.
[0008] S2. Obtain an object point of the optical system according to the object plane, perform uniform sampling within the entrance pupil to obtain sampling points, emit light from the object point to the sampling point and perform surface-by-surface tracing processing according to the optical surface to obtain several intersection points, and then perform centroid processing and grid point processing in sequence on the several intersection points to obtain several sampling grid points.
[0009] S3. Performing coherent complex amplitude processing on a plurality of sampling grid points to obtain coherent complex amplitude functions of the plurality of sampling grid points, and then performing conjugate product processing on the coherent complex amplitude functions to obtain point spread functions and parameters in the point spread functions.
[0010] S4. Perform forward propagation processing, gradient processing, and loss function processing on the point spread function in sequence, and then use the automatic differentiation method to perform reverse gradient propagation on the point spread function to obtain the optimized gradient of the parameters in the point spread function.
[0011] S5. Degenerate the optical system according to the optimized gradient of the parameters in the point spread function to obtain a degraded optical system.
[0012] The step S2 is specifically as follows:
[0013] S21. Acquire an object point of the optical system according to the object plane, perform uniform sampling in the entrance pupil to obtain a plurality of sampling points, and emit light from the object point to each sampling point in the entrance pupil to obtain a plurality of sampling rays.
[0014] S22. Obtain the light position, light direction, and light wave number according to the sampled light, and use the light position, light direction, and light wave number as storage data.
[0015] S23. Obtain the type of the next surface of the optical system according to the optical surface, perform ray tracing on the next surface, obtain several intersection points of the sampling light on the next surface, and obtain the optical path of each sampling light.
[0016] S24. Repeat step S23 until all sampling light rays reach the last optical surface along the propagation direction, that is, reach the image plane, and obtain the intersection point of each sampling light ray on the image plane, that is, obtain several intersection points on the image plane.
[0017] S25. Accumulate all optical paths of each sampling light to obtain the total optical path of each sampling light, that is, obtain several optical paths of light.
[0018] S26. Taking the centroids of the plurality of intersection points on the image plane to obtain the centroids of the plurality of intersection points on the image plane.
[0019] S27 , establishing a grid point process on the image plane with the centroid on the image plane as the center and the preset pixel size of the image plane as the interval to obtain a plurality of sampling grid points, and using the sampling grid points as storage data.
[0020] The coherent complex amplitude function is set according to the following formula:
[0021]
[0022] Among them, U(x, y) is the complex amplitude function, e j Represents the phase factor of the complex number, i is the index, opd i represents the cumulative optical path of the i-th ray from the object point to the image plane, k represents the wave number, Δr i (x, y) represents the optical path difference from the sampling point (x, y) in the point spread function to the wavefront of the sampling light i, Represents the propagation direction vector of a single sampling ray and the image plane normal vector The inner product of It represents the complex amplitude of all sampled rays summed at the sampling point (x, y) in the point spread function.
[0023] The point spread function is set according to the following formula:
[0024]
[0025] Where PSF(x, y) is the point spread function, U(x, y) is the complex amplitude function, is the conjugate function of U(x, y)
[0026] The step S4 is specifically as follows:
[0027] S41. Perform forward propagation processing on the point spread function and construct a differential operator based on the stored data.
[0028] S42. Performing gradient processing on the point spread function using an analytical differential relationship method according to the differential operator to obtain the gradient of the parameters in the point spread function.
[0029] S43. Perform loss function processing on the point spread function to obtain a loss function of the point spread function.
[0030] S44. Perform reverse gradient propagation on the loss function of the point spread function using an automatic differentiation method to obtain an optimized gradient of the parameters in the point spread function.
[0031] The stored data in step S41 includes the light position, light direction, light path, light wave number and sampling grid points.
[0032] The gradient of the parameters in the point spread function in step S42 includes the gradient of the light position, the gradient of the light direction and the gradient of the light path.
[0033] The gradient of the ray position is set according to the following formula:
[0034]
[0035] Among them, G o Represents the gradient of the light position o, G PSF represents the gradient of the point spread function, PSF represents the point spread function, U represents the complex amplitude function, Indicates derivative, o indicates the position of the light, Re() and Im() indicate the real and imaginary parts, is the phase of the complex amplitude function U.
[0036] The gradient of the light direction is set according to the following formula:
[0037]
[0038] Among them, d represents the direction of light, G d Represents the gradient of the light direction d, opd i represents the optical path length of the i-th ray, Δr i (x, y) represents the optical path difference from the i-th ray to the grid point (x, y), e j represents the complex phase factor, k represents the wave number of the light, Indicates the normal direction of the image plane.
[0039] The gradient of the light path length is set according to the following formula:
[0040]
[0041] Among them, opd represents the optical path of the light, G opd Represents the gradient of the light path length.
[0042] The beneficial effects of the present invention are:
[0043] This invention is applied to joint computational optics design tasks. The accuracy of coherent point spread function simulation based on plane wave characterization is superior to that of rendered incoherent point spread functions using geometric aberration estimation, making it suitable for small-aperture, diffraction-limited optical systems. The differential operator of the point spread function uses an analytical formula to process the gradient during the return process, improving the high memory overhead of coherent degradation simulation and enabling efficient differentiable optical degradation simulation. This invention enables optical-digital joint design tasks to accurately simulate optical degradation while meeting the requirements of high memory overhead. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 Schematic diagram of the overall process of the method of the present invention;
[0045] Figure 2Schematic diagram of ray tracing in Example 2 of the present invention;
[0046] Figure 3 This is a specific schematic diagram on the object surface in Example 2 of the present invention;
[0047] Figure 4 This is a specific schematic diagram of the image plane in Example 2 of the present invention;
[0048] Figure 5 This is a result diagram of the point spread function in Example 2 of the present invention;
[0049] Figure 6 This is the point spread function differential operator cost graph in embodiment 2 of the present invention. DETAILED DESCRIPTION
[0050] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0051] like Figure 1 As shown, embodiment 1 of the present invention includes the following steps:
[0052] S1. In computer optical simulation software, obtain the optical surface, optical medium, and object plane of the optical system, and use the paraxial optical method to perform conjugate processing on the optical system based on the optical surface and optical medium of the optical system to obtain the entrance pupil of the optical system.
[0053] An optical surface is a surface on an optical element (such as a lens, a reflector, a prism, etc.) that is used to interact with light waves. An optical surface is defined based on the surface type and surface parameters of an optical system.
[0054] Optical media are materials through which light passes, including gases (including air), liquids, and solids; they are defined according to the material names of optical systems.
[0055] The image plane is the last plane in the optical system where light rays are traced one by one, and it is also the plane where all light rays in the optical system converge to form a real image.
[0056] The conjugate processing is specifically to take the conjugate of the aperture in the object space, that is, to obtain the entrance pupil.
[0057] S2. Obtain an object point of the optical system according to the object plane, perform uniform sampling within the entrance pupil to obtain sampling points, emit light from the object point to the sampling point and perform surface-by-surface tracing processing according to the optical surface to obtain several intersection points, and then perform centroid processing and grid point processing in sequence on the several intersection points to obtain several sampling grid points.
[0058] An object point is a point light source on the object surface.
[0059] The surface tracing process is to trace the light rays in the optical system in sequence according to the surface sequence until they reach the image plane.
[0060] Step S2 is specifically as follows:
[0061] S21. Acquire an object point of the optical system according to the object plane, perform uniform sampling in a rectangular grid within the entrance pupil to obtain a plurality of sampling points, and emit light from the object point to each sampling point within the entrance pupil to obtain a plurality of sampling rays.
[0062] S22. Obtain the light position, light direction, and light wave number according to the sampled light, and use the light position, light direction, and light wave number as storage data.
[0063] S23. Obtain the type of the next surface of the optical system according to the optical surface, perform ray tracing on the next surface, obtain several intersection points of the sampling light on the next surface, and obtain the optical path of each sampling light.
[0064] S24. Repeat step S23 until all sampling light rays reach the last optical surface along the propagation direction, that is, reach the image plane, and obtain the intersection point of each sampling light ray on the image plane, that is, obtain several intersection points on the image plane.
[0065] S25. Accumulate all optical paths of each sampling light to obtain the total optical path of each sampling light, that is, obtain several optical paths of light.
[0066] S26. Taking the centroids of the plurality of intersection points on the image plane to obtain the centroids of the plurality of intersection points on the image plane.
[0067] S27 , establishing a grid point process on the image plane with the centroid on the image plane as the center and the preset pixel size of the image plane as the interval to obtain a plurality of sampling grid points, and using the sampling grid points as storage data.
[0068] The sampling grid is the coordinates of the pixel grid points to be solved on the image plane.
[0069] S3. Performing coherent complex amplitude processing on the plurality of sampling grid points to obtain coherent complex amplitude functions of the plurality of sampling grid points, and then performing conjugate product processing on the coherent complex amplitude functions of the plurality of sampling grid points to obtain a point spread function and parameters in the point spread function.
[0070] The coherent complex amplitude function is set according to the following formula:
[0071]
[0072] Among them, U(x, y) is the complex amplitude function, e j Represents the phase factor of the complex number, i is the index, opd i represents the cumulative optical path of the i-th ray from the object point to the image plane, k represents the wave number, Δr i(x, y) represents the optical path difference from the sampling point (x, y) in the point spread function to the wavefront of the sampling light i, Represents the propagation direction vector of a single sampling ray and the image plane normal vector The inner product of It represents the complex amplitude of all sampled rays summed at the sampling point (x, y) in the point spread function.
[0073] The point spread function is set as follows:
[0074]
[0075] Where PSF(x, y) is the point spread function, U(x, y) is the complex amplitude function, is the conjugate function of U(x, y)
[0076] S4. Perform forward propagation processing, gradient processing, and loss function processing on the point spread function in sequence, and then use the automatic differentiation method to perform reverse gradient propagation on the point spread function to obtain the optimized gradient of the parameters in the point spread function.
[0077] Step S4 is specifically as follows:
[0078] S41. Perform forward propagation processing on the point spread function and construct a differential operator based on the stored data.
[0079] The stored data in step S41 includes the light position, light direction, light path, light wave number and sampling grid points.
[0080] S42, performing gradient processing on the point spread function using an analytical differential relationship method according to the differential operator to obtain the gradient of the parameters in the point spread function;
[0081] The gradient of the parameters in the point spread function in step S42 includes the gradient of the light position, the gradient of the light direction, and the gradient of the light path.
[0082] The gradient at the ray position is set by the following formula:
[0083]
[0084] Among them, G o represents the gradient of the light position o, G PSF represents the gradient of the point spread function, PSF represents the point spread function, U represents the complex amplitude function, Indicates derivative, o indicates the position of the light, Re() and Im() indicate the real and imaginary parts, is the phase of the complex amplitude function U.
[0085] The gradient of the light direction is set according to the following formula:
[0086]
[0087] Among them, d represents the direction of light, G d Represents the gradient of the light direction d, opd i represents the optical path length of the i-th ray, Δr i (x, y) represents the optical path difference from the i-th ray to the grid point (x, y), e j represents the complex phase factor, k represents the wave number of the light, Indicates the normal direction of the image plane.
[0088] The gradient of the ray path length is set according to the following formula:
[0089]
[0090] Among them, opd represents the optical path of the light, G opd Represents the gradient of the light path length.
[0091] S43, performing loss function processing on the point spread function to obtain a loss function of the point spread function;
[0092] S44. Perform reverse gradient propagation on the loss function of the point spread function using an automatic differentiation method to obtain an optimized gradient of the parameters in the point spread function.
[0093] S5. Degenerate the optical system according to the optimized gradient of the parameters in the point spread function to obtain a degraded optical system.
[0094] Degeneration processing is to optimize and update the parameters in the point spread function according to the optimized gradient of the parameters in the point spread function in order to reduce the dispersion degree of the point spread function and obtain a better optical system.
[0095] Example 2
[0096] D1. In computer optical simulation software, define the optical surface according to the surface type and surface parameters. The optical surfaces in this embodiment are all plane or spherical, including curvature, thickness, and aperture parameters; and obtain the object surface in the computer optical simulation software.
[0097] D2. Define the optical medium between the surfaces according to the material name. In this embodiment, the materials S-LAH53, S-LAH52, and S-LAH51 are from the ORAHA material library, SF56A is from the SCHOTT material library, and FD11 is from the HOYA material library.
[0098] D3. According to the paraxial optical calculation, the conjugate of the system aperture in the object space, that is, the entrance pupil, is as follows: Figure 3 As shown, it is located 4.5116 mm on the optical axis and has a diameter of 6.3468 mm.
[0099] This embodiment uses a double Gauss lens as shown in Table 1 as an example to define the optical system. Its operating wavelength is 587.6 nm, the system aperture is F / 1.2 in image space, and the aperture is 2.0123 mm.
[0100]
[0101] D3. Obtain the object point of the optical system according to the object plane. In this embodiment, starting from the object point with a 16.7° field of view, 129×129 sampling rays are uniformly distributed on the entrance pupil in a rectangular grid, as shown in FIG. Figure 3 shown.
[0102] D4. Tracing rays in sequence according to the surface sequence, first obtain the intersection of the sampling ray and the surface, and then process the emission direction after passing through the surface according to the law of refraction; Figure 2 As shown in Figure 2, the tracing process of some sampling rays is drawn.
[0103] D5. Trace all sampling rays to the image plane to obtain several intersection points. Take the centroid of these intersection points and accumulate the optical path data during the propagation of the sampling rays, that is, the refractive index in the medium multiplied by the propagation distance.
[0104] D6, with the center of mass on the image plane as the center and the preset sensor pixel size of 0.6um as the interval, define the sampling grid of the point spread function of 63×63 size, such as Figure 4 shown.
[0105] D7. In the forward calculation of the point spread function of the differential operator, the stored data (ray position o, ray direction d, ray optical path opd, ray wave number k, and point spread function sampling grid) are first stored, and then the coherent complex amplitude at the sampling points of the point spread function is processed according to the complex amplitude formula represented by the plane wave:
[0106]
[0107] Among them, U(x, y) is the complex amplitude function, e j Represents the phase factor of the complex number, i is the index, opd i represents the cumulative optical path of the i-th ray from the object point to the image plane, k represents the wave number, Δr i (x, y) represents the optical path difference from the sampling point (x, y) in the point spread function to the wavefront of the sampling light i, Represents the propagation direction vector of a single sampling ray and the image plane normal vector The inner product of It represents the complex amplitude of all sampled rays summed at the sampling point (x, y) in the point spread function.
[0108] The point spread function is the complex amplitude function U(x, y) and its conjugate function The product, that is
[0109]
[0110] Where PSF(x, y) is the point spread function, U(x, y) is the complex amplitude function, is the conjugate function of U(x, y)
[0111] D8. In the backward propagation gradient process of the differential operator, the stored data in the forward process (ray position o, ray direction d, ray optical path opd, ray wave number k, and point spread function sampling grid) are first read out, and then the gradient of the ray data is processed according to the analytical differential relationship.
[0112] The gradients of the parameters in the point spread function include the gradient of the light position, the gradient of the light direction, and the gradient of the light path.
[0113] The gradient of the light position o is set according to the following formula:
[0114]
[0115] Among them, G o represents the gradient of the light position o, G PSF represents the gradient of the point spread function, PSF represents the point spread function, U represents the complex amplitude function, Indicates derivative, o indicates the position of the light, Re() and Im() indicate the real and imaginary parts, is the phase of the complex amplitude function U.
[0116] The gradient of the light direction d is set according to the following formula:
[0117]
[0118] Among them, d represents the direction of light, G d Represents the gradient of the light direction d, opd i represents the optical path length of the i-th ray, Δr i (x, y) represents the optical path difference from the i-th ray to the grid point (x, y), e j represents the complex phase factor, k represents the wave number of the light, Indicates the normal direction of the image plane.
[0119] The gradient of the light path length opd is set according to the following formula:
[0120]
[0121] Among them, opd represents the optical path of the light, G opd Represents the gradient of the light path length.
[0122] Figure 5 The forward calculation results of the point spread function differential operator are compared with the reference processing results output in the Zemax optical simulation software, which illustrates the accuracy of the method of the present invention in simulating optical degradation. Figure 6 This is the memory overhead curve required for processing multiple point spread functions. The point spread function differential operator saves 18.4 times the memory overhead compared to the automatic differentiation function writing method, which shows the high efficiency of the method of the present invention.
[0123] D9, the loss function Loss that processes the point spread function. In this example, we set it to the maximum value of the point spread function.
[0124] D10. Use Pytorch's automatic differentiation tool for reverse gradient propagation.
[0125] D11. Obtain the optimized gradient of the parameters in the point spread function. In this example, the optimized variable is set to the curvature of surface 1 to surface 6.
[0126] D12. Degenerate the optical system according to the optimized gradient of the parameters in the point spread function to obtain a degraded optical system.
[0127] Table 2 shows the gradient output results in this example. It is also compared with the gradient reference calculated by the finite element method. The relative error is less than one thousandth, which shows that the accuracy of the method of the present invention in differential calculation can be applied to the joint optimization design task of light and numbers.
[0128]
[0129]
[0130] The present invention proposes an efficient and differentiable optical system degradation simulation method that achieves accurate and differentiable coherent optical degradation simulation using a differential operator of a point spread function. Furthermore, the proposed differential operator of the point spread function uses an analytical formula to process gradient information, alleviating the high memory overhead of coherent degradation simulation and enabling efficient and differentiable optical degradation simulation.
Claims
1. An efficient and differentiable optical system degradation simulation method, characterized by: The following steps are involved: S1. In computer optical simulation software, obtain the optical surface, optical medium, and object plane of the optical system, and perform conjugate processing on the optical system using the paraxial optical method based on the optical surface and optical medium of the optical system to obtain the entrance pupil of the optical system; S2. Obtain an object point of the optical system according to the object plane, perform uniform sampling within the entrance pupil to obtain sampling points, emit light from the object point to the sampling point and perform surface-by-surface tracing processing according to the optical surface to obtain a number of intersection points, and then perform centroid processing and grid point processing in sequence on the intersection points to obtain a number of sampling grid points; S3. performing coherent complex amplitude processing on the plurality of sampling grid points to obtain coherent complex amplitude functions of the plurality of sampling grid points, and then performing conjugate product processing on the coherent complex amplitude functions to obtain a point spread function and parameters in the point spread function; S4, performing forward propagation processing, gradient processing, and loss function processing on the point spread function in sequence, and then performing reverse gradient propagation on the point spread function using an automatic differentiation method to obtain an optimized gradient of the parameters in the point spread function; S5. Degrading the optical system according to the optimized gradient of the parameters in the point spread function to obtain a degraded optical system; The step S4 is specifically as follows: S41, performing forward propagation processing on the point spread function and constructing a differential operator based on the stored data; S42, performing gradient processing on the point spread function using an analytical differential relationship method according to the differential operator to obtain the gradient of the parameters in the point spread function; S43, performing loss function processing on the point spread function to obtain a loss function of the point spread function; S44, performing reverse gradient propagation on the loss function of the point spread function using an automatic differentiation method to obtain an optimized gradient of the parameters in the point spread function; The gradient of the parameters in the point spread function in step S42 includes the gradient of the light position, the gradient of the light direction and the gradient of the light path.
2. The efficient and differentiable optical system degradation simulation method according to claim 1, characterized in that: The step S2 is specifically as follows: S21. Acquire an object point of the optical system according to the object plane, perform uniform sampling in the entrance pupil to obtain a plurality of sampling points, and emit light from the object point to each sampling point in the entrance pupil to obtain a plurality of sampling rays; S22, obtaining a light position, a light direction, and a light wave number according to the sampled light, and storing the light position, light direction, and light wave number as storage data; S23. Obtaining the type of the next surface of the optical system according to the optical surface, performing ray tracing on the next surface, obtaining a plurality of intersection points of the sampling light on the next surface, and obtaining the optical path of each sampling light; S24, repeating step S23 until all sampling light rays reach the last optical surface along the propagation direction, that is, reach the image plane, and obtaining the intersection point of each sampling light ray on the image plane, that is, obtaining a plurality of intersection points on the image plane; S25, summing up all the optical paths of each sampling light to obtain the total optical path of each sampling light, that is, obtaining the optical paths of several light rays; S26, taking the centroid of the plurality of intersection points on the image plane to obtain the centroid of the plurality of intersection points on the image plane; S27 , establishing a grid point process on the image plane with the centroid on the image plane as the center and the preset pixel size of the image plane as the interval to obtain a plurality of sampling grid points, and using the sampling grid points as storage data.
3. The efficient and differentiable optical system degradation simulation method according to claim 1, characterized in that: The coherent complex amplitude function is set according to the following formula: Among them, U(x,y) is the complex amplitude function, e j Represents the phase factor of the complex number, i is the index, opd i represents the cumulative optical path of the i-th ray from the object point to the image plane, k represents the wave number, Δr i (x,y) represents the optical path difference from the sampling point (x,y) in the point spread function to the wavefront of the sampling light i, Represents the propagation direction vector of a single sampling ray and the image plane normal vector The inner product of It represents the complex amplitude of all sampled rays summed at the sampling point (x,y) in the point spread function.
4. The efficient and differentiable optical system degradation simulation method according to claim 1, characterized in that: The point spread function is set according to the following formula: Among them, PSF(x,y) is the point spread function, U(x,y) is the complex amplitude function, is the conjugate function of U(x,y) 5. The efficient and differentiable optical system degradation simulation method according to claim 1, wherein: The stored data in step S41 includes the light position, light direction, light path, light wave number and sampling grid points.
6. The efficient and differentiable optical system degradation simulation method according to claim 1, characterized in that: The gradient of the ray position is set according to the following formula: Among them, G o Represents the gradient of the light position o, G PSF represents the gradient of the point spread function, PSF represents the point spread function, U represents the complex amplitude function, Indicates derivative, o indicates the position of the light, Re() and Im() indicate the real and imaginary parts, is the phase of the complex amplitude function U.
7. The efficient and differentiable optical system degradation simulation method according to claim 6, characterized in that: The gradient of the light direction is set according to the following formula: Among them, d represents the direction of light, G d Represents the gradient of the light direction d, opd i represents the optical path of the i-th ray, Δr i (x,y) represents the optical path difference from the i-th ray to the grid point (x,y), e j represents the complex phase factor, k represents the wave number of the light, Indicates the normal direction of the image plane.
8. The efficient and differentiable optical system degradation simulation method according to claim 7, characterized in that: The gradient of the light path length is set according to the following formula: Among them, opd represents the optical path of the light, G opd Represents the gradient of the light path length.