A method for generating starting points for design of large field-of-view off-axis dual-mirror system based on neural network
Through the neural network-based design method, a starting point for the design of a large field of view off-axis two-inverse system is generated, which solves the problems of large field of view design and off-axis incomplete problems in the existing technology, and achieves a combination of high optimization potential and successful design.
Patent Information
- Application Number
- CN202211262909.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-14
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2042-10-14
AI Technical Summary
The existing off-axis multi-inverse system design starting point generation method is difficult to generate a large field of view off-axis two-inverse system, and there is an off-axis incomplete phenomenon, which has low optimization potential.
The starting point generation method of large field of view off-axis two-inverting system based on neural network is adopted. By setting the upper limit of design difficulty, a coaxial two-inverting spherical system is designed, off-axis operation is performed and the spherical surface is adjusted to be a conical surface to judge the rationality of the design starting point. Identify the neural network dataset and train the neural network to ensure that the generated design starting point is completely off-axis.
The generated design starting point has higher optimization potential, avoiding off-axis incompleteness and ensuring that the design starting point of the large field of view off-axis two-inverse system can be successfully optimized into the final system.
Smart Images

Figure CN115524850B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of optical system design, and in particular relates to a method for generating a design starting point of a large-field off-axis two-mirror system based on a neural network. Background Art
[0002] Off-axis multi-reflector systems include off-axis two-reflector systems, off-axis three-reflector systems, and off-axis four-reflector systems. They are important components of telescopes and imaging spectrometers and are widely used in the field of observation. Large-field off-axis multi-reflector systems have a larger field of view and are the development trend of off-axis multi-reflector systems. The design of an off-axis multi-reflector system often requires two steps, namely generating a design starting point and optimizing the design starting point into the final system. The design starting point is an optical structure that meets most design requirements (field of view, entrance pupil diameter, F number, etc.) and can be successfully optimized into the final system. The design starting point of a good off-axis multi-reflector system should have two characteristics. The first is complete off-axis, that is, each reflector in the design starting point of the off-axis multi-reflector system does not block any light passing through the off-axis multi-reflector system, and an appropriate optical margin is left between the reflector and the light, so as to facilitate optical designers to optimize the design starting point into the final system and facilitate processing; the second is optimization potential, because for the off-axis multi-reflector system, even if the design starting point is completely off-axis, if the off-axis method is unreasonable, or some parameters in the off-axis multi-reflector system are improperly set, the design starting point cannot be optimized into the final system.
[0003] At present, there are two methods available for generating the design starting point of off-axis multi-reflector systems. One is the more classic patent library matching design method, and the other is the neural network design method that has emerged in recent years. The patent library matching design method is to find an off-axis multi-reflector system in the optical patent library that has the same number of reflectors as the design target and other indicators that are relatively close as the design starting point. However, the available off-axis multi-reflector systems in the existing patent library are very limited, and it is even more difficult to find an off-axis two-reflector system with a large field of view. For this reason, optical designers have to further lower the screening criteria, especially the screening criteria for the field of view, but the off-axis method and parameter settings of such a design starting point are often not suitable for off-axis multi-reflector systems with a large field of view, so the optimization potential is low and it is easy to fail. Although the existing neural network design method can customize the design starting point of the corresponding off-axis multi-reflector system according to the design requirements, there is a certain proportion of off-axis incompleteness in the design results, that is, some reflectors block part of the light. Summary of the invention
[0004] In view of the problems existing in the prior art, the present invention provides a method for generating a design starting point for a large-field-of-view off-axis dual-mirror system based on a neural network, aiming to generate a design starting point that is completely off-axis and has optimization potential for a large-field-of-view off-axis dual-mirror system.
[0005] The technical solution of the present invention is:
[0006] A method for generating a design starting point for a large-field-of-view off-axis dual-mirror system based on a neural network, the method comprising the following steps:
[0007] Step 1: Set the upper limit of the design difficulty of the large field of view off-axis dual-mirror system;
[0008] Step 2: Design the coaxial two-anti-spherical system G0;
[0009] Step 3: Perform off-axis operation on the coaxial two-contraspherical system G0 to construct an off-axis two-contraspherical system G0′;
[0010] Step 4: Adjust the spherical surface in the off-axis spherical system G0′ to a conic surface to obtain the design starting point G of the large-field off-axis dual-mirror system;
[0011] Step 5: Determine whether the design starting point G of the large-field-of-view off-axis dual-mirror system is reasonable. If it is reasonable, execute step 6. If it is unreasonable, return to step 1 and re-set the upper limit of the design difficulty requirement of the large-field-of-view off-axis dual-mirror system.
[0012] Step 6: Obtain the lower limit of the design difficulty by reducing the field of view, relative aperture and entrance pupil diameter in the upper limit of the design difficulty, and then obtain the specification range of the large field of view off-axis dual mirror system;
[0013] Step 7: Determine a data set for training the neural network according to the specification range of the large field of view off-axis dual-mirror system, including an input set and an output set; the input set includes N sets of design requirements I = {DR1, DR2, ..., DR N}, where DR i (1≤i≤N) is the i-th group of design requirements; any DR i (ED i ,F i ,FOV i ) contains the entrance pupil diameter ED in the i-th group of design requirements i , the F number F in the i-th group of design requirements i and the field of view FOV in the i-th group of design requirements i ; N = N ED ×N F ×N FOV , where N ED 、N F and N FOV are the entrance pupil diameter, F number and number of samples of the field of view in the design requirements of the input set; the output set contains N groups of system parameters O = {G1, G2, ..., G N}, where G i (1≤i≤N) corresponds to the design starting point of the large field of view off-axis dual-mirror system of the i-th group of design requirements;
[0014] Step 8: Determine the type of neural network, set training evaluation indicators, and use the data set determined in step 7 to train the determined type of neural network;
[0015] Step 9: Input the design requirements of any large-field-of-view off-axis dual-mirror system within the specification range of the large-field-of-view off-axis dual-mirror system into the trained neural network to obtain the design starting point of the corresponding large-field-of-view off-axis dual-mirror system.
[0016] Furthermore, according to the method for generating the design starting point of the large-field-of-view off-axis two-mirror system based on a neural network, the coaxial two-mirror spherical system G0 is a Schwarzschild-type coaxial two-mirror spherical system.
[0017] Further, according to the method for generating the starting point of the design of the large-field off-axis two-mirror system based on a neural network, the coaxial two-mirror spherical system G0 is G0(R1, R2, d1, d2), and the relationship between the parameters is:
[0018]
[0019] Among them, R1 and R2 are the curvature radii of the primary mirror and the secondary mirror in the coaxial two-mirror spherical system respectively; f is the focal length of the coaxial two-mirror spherical system, and the value of f is the product of the entrance pupil diameter and the F number; β is the magnification of the secondary mirror in the coaxial two-mirror spherical system; d1 and d2 are the distance between the primary and secondary mirrors and the distance from the secondary mirror to the image point in the coaxial two-mirror spherical system respectively.
[0020] Further, according to the method for generating the design starting point of the large-field-of-view off-axis two-mirror system based on a neural network, the design starting point G obtained in step 4 is G(R1′, R2′, d1′, d2′, α1, α2, k1, k2), wherein R1′ and R2′ respectively represent the curvature radius of the center point of the primary mirror and the curvature radius of the center point of the secondary mirror in the design starting point of the large-field-of-view off-axis two-mirror system; d1′ and d2′ respectively represent the primary-secondary mirror spacing and the distance from the secondary mirror to the image point in the design starting point of the large-field-of-view off-axis two-mirror system; α1 and α2 respectively represent the inclination angle of the primary mirror and the inclination angle of the secondary mirror in the design starting point of the large-field-of-view off-axis two-mirror system, both representing the inclination angle of the local surface y-axis relative to the global y-axis; k1 and k2 respectively represent the surface shape coefficient of the primary mirror and the surface shape coefficient of the secondary mirror in the design starting point of the large-field-of-view off-axis two-mirror system.
[0021] Further, according to the method for generating the design starting point of the large field of view off-axis two-mirror system based on a neural network, the method for judging whether the design starting point G of the large field of view off-axis two-mirror system is reasonable is: first, the conic surface in G is expanded into a free surface, and the design starting point of the large field of view off-axis two-mirror system is optimized, and then it is judged whether the optimization result is close to the diffraction limit. If so, it is judged that the design starting point G of the large field of view off-axis two-mirror system is reasonable; if not, it is judged that the design starting point G of the large field of view off-axis two-mirror system is unreasonable.
[0022] Further, according to the method for generating the starting point of the design of the large-field off-axis dual-mirror system based on a neural network, the training evaluation index, represented by E, is calculated as follows:
[0023] E=MSE×J1×J2 (2)
[0024] Where MSE is the root mean square error; J1 and J2 represent the judgment index related to the occlusion and the judgment index related to the optical margin, respectively. J1 and J2 are defined as:
[0025]
[0026] Wherein, N is the number of samples; Jo1 and Jo2 are the occlusion determination numbers; Jod1 and Jod2 are the optical remainder determination numbers;
[0027] The occlusion discrimination numbers Jo1 and Jo2 are defined as:
[0028]
[0029] Among them, y h and l They represent the y coordinates of the upper edge point of the primary mirror and the lower edge point of the secondary mirror respectively; y1 is the y coordinate of the foot point of the perpendicular path from the upper edge point of the primary mirror to the light ray at the lower edge of the field of view (0, -FOV / 2) emitted to the image, and FOV is the field of view in the design requirements; y2 is the y coordinate of the foot point of the perpendicular path from the lower edge point of the secondary mirror to the light ray at the upper edge of the field of view (0, -FOV / 2) incident to the primary mirror;
[0030] The optical margin discrimination numbers Jod1 and Jod2 are defined as:
[0031]
[0032] In the formula, od is the minimum standard of optical margin; od1 and od2 are both optical margins, and the specific expressions are:
[0033]
[0034] In the formula, x s and sare the x- and y-coordinates of the imaging point in the field of view (0, -FOV / 2).
[0035] Furthermore, according to the method for generating the starting point of the design of the large-field off-axis dual-mirror system based on a neural network, the specific parameter calculation method of the obstruction discrimination number and the optical margin discrimination number is as follows:
[0036] First, the expression of the primary mirror in global coordinates is introduced:
[0037]
[0038] In the above formula, R1′, k1 and α1 are known parameters from the output set of the neural network;
[0039] Since the incident angle of the edge light in the central field of view is 0°, the incident height is -ED / 2, and ED is the entrance pupil diameter, substituting y=-ED / 2 into equation (7), the intersection point (x c ,y c )’s global coordinates; to convert (x c ,y c ) is converted to the coordinates (x cl ,y cl ), introduce the following conversion equation:
[0040]
[0041] According to the height h of the edge light under the central field of view incident on the primary mirror in the local coordinate system of the primary mirror in(1) =y cl and the incident angle θ in(1) , use the light transfer matrix between the reflectors to calculate the angle θ of the edge light in the central field of view in the local coordinate system of the primary mirror out(1) , the specific formula is:
[0042]
[0043] in:
[0044]
[0045] In the above formula, n is the mirror number. For the primary mirror, n = 1; ρ n Represents the radius of curvature at the intersection of the light and the nth reflector; α n is the tilt angle of the nth reflector in the global coordinate system; ω is the incident angle of the light, ω=0 for the central light, and ω=-FOV / 2 for the light of the (0,-FOV / 2) field of view;
[0046] According to the obtained θ out(1)and (x c ,y c ) calculates the optical path of the edge light in the central field of view after it is emitted from the primary mirror. The expression is:
[0047] y=-xtan(θ out (1)+α1)+y c +x c tan(θ out (1)+α1) (11)
[0048] Introduce the expression of the secondary mirror in global coordinates:
[0049]
[0050] In the above formula, R2′, k2 and α2 are known parameters from the output set of the neural network, where; x * ,y * It can be expressed as:
[0051]
[0052] In the above formula, d1′ comes from the known parameters in the output set of the neural network;
[0053] By combining the optical path expression (11) of the edge light in the central field of view after it is emitted from the primary mirror with the expression (12) of the secondary mirror in global coordinates, the coordinates of the edge point under the secondary mirror (x l ,y l ); Similarly, according to the object-image relationship and the principle of reversibility of the optical path, the light transfer matrix (9) between the reflectors is used to reversely trace the upper edge light of the (0, -FOV / 2) field of view to calculate the coordinates of the upper edge point of the primary mirror (x h ,y h ); It is known that the passing point (x h ,y h ) The angle at which the edge light of the (0, -FOV / 2) field of view is incident on the primary mirror is ω = -FOV / 2, so the optical path expression of the edge light of the (0, -FOV / 2) field of view is:
[0054] (-FOV / 2)x-y+y h -(-FOV / 2)x h =0 (14)
[0055] According to the obtained (x l ,y l ) and (0, -FOV / 2) the coordinates of the imaging point in the field of view (x s ,y s ), the optical path expression of the edge light between the secondary mirror and the image in the field of view (0, -FOV / 2) is obtained as follows:
[0056] (y s -y l )x+(x s -x l )y+x s y l -x l y s =0 (15)
[0057] Calculate the value from the lower edge point of the secondary mirror (x l ,y l ) The coordinates of the foot point (x1, y1) of the perpendicular line of the light incident on the primary mirror at the upper edge of the field of view (0, -FOV / 2) are:
[0058]
[0059] Calculate the value from the upper edge point of the primary mirror (x h ,y h ) The coordinates of the foot point (x2, y2) of the perpendicular line drawn between the secondary mirror and the image to the lower edge of the field of view (0, -FOV / 2) are:
[0060]
[0061] Compared with the prior art, the technical solution proposed by the present invention has the following beneficial effects:
[0062] 1. By setting a reasonable upper limit of the difficulty of the design requirements, the optimization potential of the starting point of the large field of view off-axis dual-mirror system design is guaranteed. Compared with the patent library matching design method, the generated starting point of the large field of view off-axis dual-mirror system design has a higher optimization potential;
[0063] 2. By adding judgments related to occlusion and optical margin when training the neural network, it is ensured that the generated large-field off-axis two-mirror system can be completely off-axis. Compared with the existing neural network design method, the present invention can avoid the phenomenon of incomplete off-axis at the starting point of the off-axis two-mirror system design. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] In order to more clearly illustrate the specific methods in the embodiments of the present invention, the relevant drawings involved in the embodiments will be briefly described below. The following drawings are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative changes.
[0065] Figure 1 A schematic flow chart of a method for generating a design starting point for a large-field-of-view off-axis dual-mirror system based on a neural network in this embodiment;
[0066] Figure 2This is a schematic diagram of the structure of a large-field off-axis dual-mirror system according to this embodiment; DETAILED DESCRIPTION
[0067] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the relevant drawings. The preferred embodiments of the present application are given in the drawings. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thoroughly and comprehensively understood.
[0068] The present invention proposes a generation scheme based on a neural network for the problem of generating a design starting point for an off-axis dual-mirror system with a large field of view, specifically, generating a design starting point for an off-axis dual-mirror system with a large field of view and complete off-axis performance. When modeling based on a neural network, before determining the specification range of the off-axis dual-mirror system, the upper limit of the difficulty of the design requirements of the off-axis dual-mirror system is first introduced to ensure the rationality of the specification range and the optimization potential of the design starting point; then, sampling is performed at equal intervals within the specification range to generate a data set for training the neural network; finally, judgment indicators related to occlusion and optical margin are added to the training evaluation indicators to ensure that the trained neural network can generate a design starting point for an off-axis dual-mirror system with a large field of view and complete off-axis performance.
[0069] Figure 1 It is a flowchart of a method for generating a design starting point for a large-field-of-view off-axis dual-mirror system based on a neural network in this embodiment.
[0070] Figure 2 It is a schematic diagram of the starting point of the design of the large-field off-axis dual-mirror system of this embodiment.
[0071] The method for generating a design starting point of a large-field-of-view off-axis dual-mirror system based on a neural network comprises the following steps:
[0072] Step 1: Set the upper limit of the design difficulty of the large field of view off-axis dual-mirror system;
[0073] The upper limit of difficulty of the design requirement means that the neural network can effectively generate a structure with maximum field of view and maximum relative aperture within the range of system parameters (including entrance pupil diameter, F number and field of view) of the starting point of the design of a large-field-of-view off-axis dual-mirror system, where the relative aperture refers to the reciprocal of the F number.
[0074] In this embodiment, the upper limit of the difficulty of the design requirement of the large-field-of-view off-axis dual-mirror system is: the entrance pupil diameter is 25 mm, the F number is 3.6, and the field of view is 20°×20°.
[0075] Determining the upper limit of the difficulty of the design requirements is equivalent to determining the most difficult system to optimize within the specification range of the off-axis mirror system (meeting the requirements of both the maximum field of view and the maximum relative aperture), which acts as a threshold. If the design starting point corresponding to the upper limit has optimization potential (i.e., it can be successfully optimized), then other off-axis mirror systems within the specification range designed using the same method must have optimization potential.
[0076] Step 2: Design the coaxial two-anti-spherical system G0 (R1, R2, d1, d2);
[0077] Considering the design requirement of a large field of view, this embodiment selects a Schwarzschild-type coaxial two-reflecting spherical system with a secondary mirror having a larger aperture than the primary mirror for design, and this embodiment derives the relationship between the parameters of the coaxial two-reflecting spherical system according to the paraxial optical theory as follows:
[0078]
[0079] Among them, R1 and R2 are the curvature radii of the primary mirror and the secondary mirror in the coaxial two-mirror spherical system respectively; f is the focal length of the coaxial two-mirror spherical system, and the value of f is the product of the entrance pupil diameter and the F number; β is the magnification of the secondary mirror in the coaxial two-mirror spherical system; d1 and d2 are the distance between the primary and secondary mirrors and the distance from the secondary mirror to the image point in the coaxial two-mirror spherical system respectively.
[0080] In this implementation, the absolute value of the distance between the primary and secondary mirrors is set to 2f, and the absolute value of the magnification is set to 0.85.
[0081] Step 3: Perform off-axis operation on the coaxial two-contraspherical system G0 to construct an off-axis two-contraspherical system G0′;
[0082] Using optical design software such as ZEMAX and codeV, the coaxial two-reflecting spherical system G0 is converted into an off-axis spherical system G0′ by changing the tilt angles of the primary mirror and the secondary mirror. Here, the off-axis spherical system G0′ is a z-type structure with the aperture on the secondary mirror.
[0083] Step 4: Adjust the spherical surface in the off-axis spherical system G0′ to a conic surface to obtain the design starting point G(R1′, R2′, d1′, d2′, α1, α2, k1, k2) of the large-field off-axis mirrorless system;
[0084] The R1′ and R2′ respectively represent the curvature radii of the center point of the primary mirror and the secondary mirror at the design starting point of the large-field off-axis two-mirror system. The values of R1′ and R2′ are equal to the curvature radius R1 of the primary mirror and the curvature radius R2 of the secondary mirror in the coaxial two-mirror spherical system, that is, these two parameters should not change in the entire adjustment process of the coaxial spherical two-mirror system-the off-axis spherical two-mirror system-the design starting point; the d1′ and d2′ respectively represent the distance between the primary and secondary mirrors and the distance from the secondary mirror to the image point at the design starting point of the large-field off-axis two-mirror system. d1′ and d2′ The value of 2′ is equal to the distance d1 between the primary and secondary mirrors and the distance d2 from the secondary mirror to the image point in the coaxial two-mirror spherical system, that is, these two parameters should not change in the entire adjustment process from the coaxial spherical two-mirror system to the off-axis spherical two-mirror system to the design starting point; the α1 and α2 refer to the inclination angles of the primary mirror and the secondary mirror at the design starting point of the large-field off-axis two-mirror system, respectively, and both represent the inclination angles of the local surface y-axis relative to the global y-axis; the k1 and k2 represent the surface shape coefficients of the primary mirror and the secondary mirror at the design starting point of the large-field off-axis two-mirror system, respectively.
[0085] Optical design software such as ZEMAX and codeV can be used to adjust the spherical surface in the off-axis spherical system G0′ to a conic surface. The resulting system is the design starting point G (R1′, R2′, d1′, d2′, α1, α2, k1, k2) of the large-field off-axis mirrorless system.
[0086] Step 5: Determine whether the design starting point G of the large field of view off-axis dual-mirror system is reasonable. If not, return to step 1. If reasonable, proceed to step 6.
[0087] Use optical design software, such as ZEMAX and codeV, to expand the conic surface in G into a free-form surface, and optimize the design starting point of the large-field-of-view off-axis two-mirror system on the optical design software. If the optimization result is close to the diffraction limit, it proves that the upper limit of the design requirement difficulty of the large-field-of-view off-axis two-mirror system in step 1 is reasonable, that is, it has optimization potential; otherwise, return to step 1 and re-set the upper limit of the design requirement difficulty of the large-field-of-view off-axis two-mirror system.
[0088] This implementation uses the optical design software ZEMAX to expand the conic surface in G into a free-form surface, and optimizes the design starting point of the large-field-of-view off-axis two-mirror system. The optimization results show that at 90lp / mm, the OTF modulus of the large-field-of-view free-form surface off-axis two-mirror system is greater than 0.5, and most of the imaging spot is within the Airy disk, that is, close to the diffraction limit. This proves that the upper limit of the design difficulty requirement of the large-field-of-view off-axis two-mirror system in step 1 of this implementation scheme is reasonably set and has optimization potential.
[0089] Step 6: Get the lower limit of the design difficulty by reducing the field of view, relative aperture and entrance pupil diameter in the upper limit of the design difficulty, and then get the specification range;
[0090] The specification range refers to the system parameter range within which the neural network can effectively generate a design starting point, and is composed of a series of design requirements.
[0091] In this embodiment, the specification range is: the entrance pupil diameter ranges from 20 mm to 25 mm, the F number ranges from 3.6 to 4, and the field of view ranges from 15°×15° to 20°×20°.
[0092] Step 7: Determine the data set for training the neural network based on the specification range;
[0093] The data set used to train the neural network includes an input set and an output set. The input set contains N (number of samples) sets of design requirements I = {DR1, DR2, ..., DR N}, where DR1, DR2, …, DR N is the first set of design requirements, the second set of design requirements, ..., the Nth set of design requirements. Any DR i (ED i ,F i ,FOV i ) contains the entrance pupil diameter (expressed as ED i ), the F number in the i-th group of design requirements (expressed as F i ) and the field of view (denoted as FOV) in the i-th group of design requirements i ), where 1≤i≤N. N=N ED ×N F ×N FOV , where N ED 、N F and N FOV are the entrance pupil diameter, F number and number of samples of the field of view in the design requirements of the input set; the output set contains N groups of system parameters O = {G1, G2, ..., G N}, where G1, G2, …, G N is the design starting point of the large field of view off-axis two-mirror system corresponding to the first set of design requirements, the design starting point of the large field of view off-axis two-mirror system corresponding to the second set of design requirements, ..., the design starting point of the large field of view off-axis two-mirror system corresponding to the Nth set of design requirements. Because the design target of the present invention is a large field of view system, the horizontal field of view and the vertical field of view are set to have the same maximum field of view requirement, and only one field of view parameter is used in the input set.
[0094] In order to obtain N sets of design requirements in the input set, this embodiment adopts an equal interval sampling method to sample the entrance pupil diameter, F number and field of view to obtain N ED 、NF and N FOV The eight parameters R1′, R2′, d1′, d2′, α1, α2, k1 and k2 contained in the design starting point G of each large-field off-axis mirror system in the output set are obtained by the method described in steps 2 to 4.
[0095] In this embodiment, N ED =3,N F =3, N FOV =6.
[0096] Step 8: To obtain a neural network that can generate a starting point for the design of a large field of view off-axis dual-mirror system, determine the type of neural network and train the neural network using the data set determined in Step 7.
[0097] Step 8.1: Set the hidden layer function, output layer function, training algorithm and evaluation index of the training neural network;
[0098] This embodiment adopts a BP neural network. In this embodiment, the transfer function of the hidden layer of the BP neural network is set to the Tan-sigmoid function, the transfer function of the output layer of the BP neural network is set to the purelin function, and the LM (Levenberg-Marquardt) algorithm is used to train the BP neural network. In order to make the design starting points of the large-field off-axis two-mirror system generated by the trained neural network completely off-axis, it is necessary to add judgments related to occlusion and optical margin during training. Therefore, a new training evaluation index is designed in this embodiment, represented by E, and the calculation formula is:
[0099] E=MSE×J1×J2 (2)
[0100] Where MSE is the root mean square error; J1 and J2 represent the judgment index related to the occlusion and the judgment index related to the optical margin, respectively. J1 and J2 are defined as:
[0101]
[0102] Wherein, N is the number of samples; Jo1 and Jo2 are the occlusion determination numbers; Jod1 and Jod2 are the optical remainder determination numbers.
[0103] The occlusion discrimination numbers Jo1 and Jo2 are defined as:
[0104]
[0105] Among them, y h and l They represent the y coordinates of the upper edge point of the primary mirror and the lower edge point of the secondary mirror, respectively. Figure 2As shown; y1 is the y coordinate of the foot point of the perpendicular path from the upper edge point of the primary mirror to the lower edge of the field of view (0, -FOV / 2) to the image, as shown Figure 2 As shown, FOV is the field of view in the design requirements; y2 is the y coordinate of the foot point of the perpendicular path from the lower edge point of the secondary mirror to the upper edge of the field of view (0, -FOV / 2) incident on the primary mirror, as shown Figure 2 shown.
[0106] The optical margin discrimination numbers Jod1 and Jod2 are defined as:
[0107]
[0108] Where od is the minimum standard of optical margin, which is generally set to 10 mm; od1 and od2 are both optical margins, that is, the distance between the edge light and the reflector, such as Figure 2 As shown, the specific expression is:
[0109]
[0110] In the formula, x s and s are the x- and y-coordinates of the imaging point in the field of view (0, -FOV / 2), such as Figure 2 As shown, (x s ,y s ) can be calculated based on the object-image relationship.
[0111] Step 8.2: Calculate specific parameters in the occlusion discrimination number and the optical margin discrimination number;
[0112] First, the expression of the primary mirror in global coordinates is introduced:
[0113]
[0114] Where R1′, k1 and α1 are known parameters from the output set of the neural network.
[0115] Since the incident angle of the edge light in the central field of view is 0°, the incident height is -ED / 2, and ED is the entrance pupil diameter. Therefore, substituting y=-ED / 2 into equation (7), the intersection point (x c ,y c ) is the global coordinate of c ,y c ) is converted to the coordinates (x cl ,y cl ), introduce the following conversion equation:
[0116]
[0117] According to the height h of the edge light under the central field of view incident on the primary mirror in the local coordinate system of the primary mirror in(1) =y cl and the incident angle θ in(1) , use the light transfer matrix between the reflectors to calculate the angle θ of the edge light in the central field of view in the local coordinate system of the primary mirror out(1) , the specific formula is:
[0118]
[0119] in:
[0120]
[0121] Where n is the mirror number, for the primary mirror, n = 1; ρ n Represents the radius of curvature at the intersection of the light and the nth reflector; α n is the tilt angle of the nth reflector in the global coordinate system; ω is the incident angle of the light, ω=0 for the central light, and ω=-FOV / 2 for the light in the field of view (0,-FOV / 2).
[0122] According to the obtained θ out(1) and (x c ,y c ) calculates the optical path of the edge light in the central field of view after it is emitted from the primary mirror. The expression is:
[0123] y=-xtan(θ out (1)+α1)+y c +x c tan(θ out (1)+α1) (11)
[0124] Introduce the expression of the secondary mirror in global coordinates:
[0125]
[0126] Among them, R2′, k2 and α2 are known parameters from the output set of the neural network; x * ,y * It can be expressed as:
[0127]
[0128] In the above formula, d1′ comes from the known parameters in the output set of the neural network;
[0129] By combining the optical path expression (11) of the edge light in the central field of view after it is emitted from the primary mirror with the expression (12) of the secondary mirror in global coordinates, the coordinates of the edge point under the secondary mirror (x l ,y l). Similarly, according to the object-image relationship and the principle of reversibility of the optical path, the light transfer matrix (9) between the reflectors is used to reversely trace the upper edge light of the (0, -FOV / 2) field of view and calculate the coordinates of the upper edge point of the primary mirror (x h ,y h ). It is known that the passing point (x h ,y h ) The angle at which the edge light of the (0, -FOV / 2) field of view is incident on the primary mirror is ω = -FOV / 2, so the optical path expression of the edge light of the (0, -FOV / 2) field of view is:
[0130] (-FOV / 2)x-y+y h -(-FOV / 2)x h =0 (14)
[0131] According to the obtained (x l ,y l ) and (0, -FOV / 2) the coordinates of the imaging point in the field of view (x s ,y s ), the optical path expression of the edge light between the secondary mirror and the image in the field of view (0, -FOV / 2) is obtained as follows:
[0132] (y s -y l )x+(x s -x l )y+x s y l -x l y s =0 (15)
[0133] Calculate the value from the lower edge point of the secondary mirror (x l ,y l ) The coordinates of the foot point (x1, y1) of the perpendicular line of the light incident on the primary mirror at the upper edge of the field of view (0, -FOV / 2) are:
[0134]
[0135] Calculate the value from the upper edge point of the primary mirror (x h ,y h ) The coordinates of the foot point (x2, y2) of the perpendicular line drawn between the secondary mirror and the image to the lower edge of the field of view (0, -FOV / 2) are:
[0136]
[0137] At this point, all parameters in the occlusion determination number and the optical remainder determination number can be calculated.
[0138] Step 8.3: Start training the neural network. When the value of E is between (0, MSEu ) range, the training process of the neural network ends and the trained BP neural network is output.
[0139] The MSE u is the upper limit of the root mean square error.
[0140] In this implementation, the upper limit of the root mean square error MSE is u The θ is set to 0.00025 and the OD is set to 10 mm. The trained BP neural network has 2 hidden layers, each with 8 hidden nodes, the MSE is 0.000236, and the E value is 0.000236.
[0141] Step 9: Input the design requirements of the large field of view off-axis dual-mirror system (i.e., three parameters such as entrance pupil diameter, F number and field of view) into the trained neural network to generate the design starting point. The result output by the neural network is the design starting point of the large field of view off-axis dual-mirror system.
[0142] In this embodiment, 25 groups of design requirements different from the training set are randomly selected within the design requirements (entrance pupil diameter: 20mm-25mm, F number: 3.6-4, field of view: 15°×15°-20°×20°), and the corresponding design starting point is output by the BP neural network trained in this embodiment. The output results show that there is no obstruction in all 25 groups of large-field-of-view off-axis two-mirror systems and the od1 and od2 in each system are greater than 10mm. This proves that the large-field-of-view off-axis two-mirror system generated by the trained BP neural network can achieve complete off-axis. Combined with the implementation results in step 5, it can be concluded that the method for generating the design starting point of a large-field-of-view off-axis two-mirror system based on a neural network proposed in the present invention can obtain a design starting point for a large-field-of-view off-axis two-mirror system with optimization potential and complete off-axis, which meets the characteristic requirements of the design starting point of an off-axis multi-mirror system.
[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents. Therefore, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.
Claims
1. A method for generating a design starting point for a large-field-of-view off-axis dual-mirror system based on a neural network, characterized in that: The method comprises the following steps: Step 1: Set the upper limit of the design difficulty of the large field of view off-axis dual-mirror system; Step 2: Design the coaxial two-anti-spherical system G0; Step 3: Perform off-axis operation on the coaxial two-anti-spherical system G0 to construct an off-axis two-anti-spherical system G′0; Step 4: Adjust the spherical surface in the off-axis spherical system G′0 to a conic surface to obtain the design starting point G of the large-field off-axis dual-mirror system; Step 5: Determine whether the design starting point G of the large-field-of-view off-axis dual-mirror system is reasonable. If it is reasonable, execute step 6. If it is unreasonable, return to step 1 and re-set the upper limit of the design difficulty requirement of the large-field-of-view off-axis dual-mirror system. Step 6: Obtain the lower limit of the design difficulty by reducing the field of view, relative aperture and entrance pupil diameter in the upper limit of the design difficulty, and then obtain the specification range of the large field of view off-axis dual mirror system; Step 7: Determine a data set for training the neural network according to the specification range of the large field of view off-axis dual-mirror system, including an input set and an output set; the input set includes N sets of design requirements I = {DR1, DR2, ..., DR N }, where DR i (1≤i≤N) is the i-th group of design requirements; any DR i (ED i ,F i ,FOV i ) contains the entrance pupil diameter ED in the i-th group of design requirements i , the F number F in the i-th group of design requirements i and the field of view FOV in the i-th group of design requirements i ; N = N ED ×N F ×N FOV , where N ED 、N F and N FOV are the entrance pupil diameter, F number and number of samples of the field of view in the design requirements of the input set; the output set contains N groups of system parameters O = {G1, G2, ..., G N }, where G i (1≤i≤N) corresponds to the design starting point of the large field of view off-axis dual-mirror system of the i-th group of design requirements; Step 8: Determine the type of neural network, set training evaluation indicators, and use the data set determined in step 7 to train the determined type of neural network; Step 9: Input the design requirements of any large-field-of-view off-axis dual-mirror system within the specification range of the large-field-of-view off-axis dual-mirror system into the trained neural network to obtain the design starting point of the corresponding large-field-of-view off-axis dual-mirror system.
2. The method for generating a design starting point of a large-field-of-view off-axis dual-mirror system based on a neural network according to claim 1, characterized in that: The coaxial two-anti-spherical system G0 is a Schwarzschild-type coaxial two-anti-spherical system.
3. The method for generating a design starting point of a large-field-of-view off-axis dual-mirror system based on a neural network according to claim 1 or 2, characterized in that: The coaxial two-anti-spherical system G0 is G0 (R1, R2, d1, d2), and the relationship between the parameters is: Among them, R1 and R2 are the curvature radii of the primary mirror and the secondary mirror in the coaxial two-mirror spherical system respectively; f is the focal length of the coaxial two-mirror spherical system, and the value of f is the product of the entrance pupil diameter and the F number; β is the magnification of the secondary mirror in the coaxial two-mirror spherical system; d1 and d2 are the distance between the primary and secondary mirrors and the distance from the secondary mirror to the image point in the coaxial two-mirror spherical system respectively.
4. The method for generating a design starting point of a large-field-of-view off-axis dual-mirror system based on a neural network according to claim 1, characterized in that: The design starting point G obtained in step 4 is G(R′1, R′2, d′1, d′2, α1, α2, k1, k2), wherein R′1 and R′2 represent the curvature radius of the center point of the primary mirror and the curvature radius of the center point of the secondary mirror, respectively, in the design starting point of the large field of view off-axis two-mirror system; d′1 and d′2 represent the primary and secondary mirror spacing and the distance from the secondary mirror to the image point, respectively, in the design starting point of the large field of view off-axis two-mirror system; α1 and α2 represent the inclination angle of the primary mirror and the inclination angle of the secondary mirror, respectively, in the design starting point of the large field of view off-axis two-mirror system, both representing the inclination angle of the local surface y-axis relative to the global y-axis; k1 and k2 represent the surface shape coefficient of the primary mirror and the surface shape coefficient of the secondary mirror, respectively, in the design starting point of the large field of view off-axis two-mirror system.
5. The method for generating a design starting point of a large-field-of-view off-axis dual-mirror system based on a neural network according to claim 1 or 4, characterized in that: The method for judging whether the design starting point G of the large field of view off-axis two-mirror system is reasonable is: first, the conic surface in G is expanded into a free surface, and the design starting point of the large field of view off-axis two-mirror system is optimized, and then whether the optimization result is close to the diffraction limit is judged, if so, the design starting point G of the large field of view off-axis two-mirror system is judged to be reasonable, if not, the design starting point G of the large field of view off-axis two-mirror system is judged to be unreasonable.
6. The method for generating a design starting point of a large-field-of-view off-axis dual-mirror system based on a neural network according to claim 4, characterized in that: The training evaluation index is represented by E, and the calculation formula is: E=MSE×J1×J2 (2) Where MSE is the root mean square error; J1 and J2 represent the judgment index related to the occlusion and the judgment index related to the optical margin, respectively. J1 and J2 are defined as: Wherein, N is the number of samples; Jo1 and Jo2 are the occlusion determination numbers; Jod1 and Jod2 are the optical remainder determination numbers; The occlusion discrimination numbers Jo1 and Jo2 are defined as: Among them, y h and l They represent the y coordinates of the upper edge point of the primary mirror and the lower edge point of the secondary mirror respectively; y1 is the y coordinate of the foot point of the perpendicular path from the upper edge point of the primary mirror to the light ray at the lower edge of the field of view (0, -FOV / 2) emitted to the image, and FOV is the field of view in the design requirements; y2 is the y coordinate of the foot point of the perpendicular path from the lower edge point of the secondary mirror to the light ray at the upper edge of the field of view (0, -FOV / 2) incident to the primary mirror; The optical margin discrimination numbers Jod1 and Jod2 are defined as: In the formula, od is the minimum standard of optical margin; od1 and od2 are both optical margins, and the specific expressions are: In the formula, x s and s are the x- and y-coordinates of the imaging point in the field of view (0, -FOV / 2).
7. The method for generating a design starting point of a large-field-of-view off-axis dual-mirror system based on a neural network according to claim 6, characterized in that: The specific parameter calculation method of the obstruction determination number and the optical margin determination number is as follows: First, the expression of the primary mirror in global coordinates is introduced: In the above formula, R′1, k1 and α1 are known parameters from the output set of the neural network; Since the incident angle of the edge light in the central field of view is 0°, the incident height is -ED / 2, and ED is the entrance pupil diameter, substituting y=-ED / 2 into equation (7), the intersection point (x c ,y c )’s global coordinates; to convert (x c ,y c ) is converted to the coordinates (x cl ,y cl ), introduce the following conversion equation: According to the height h of the edge light under the central field of view incident on the primary mirror in the local coordinate system of the primary mirror in(1) =y cl and the incident angle θ in(1) , use the light transfer matrix between the reflectors to calculate the angle θ of the edge light in the central field of view in the local coordinate system of the primary mirror out(1) , the specific formula is: in: In the above formula, n is the mirror number. For the primary mirror, n = 1; ρ n Represents the radius of curvature at the intersection of the light and the nth reflector; α n is the tilt angle of the nth reflector in the global coordinate system; ω is the incident angle of the light, ω=0 for the central light, and ω=-FOV / 2 for the light of the (0,-FOV / 2) field of view; According to the obtained θ out(1) and (x c ,y c ) calculates the optical path of the edge light in the central field of view after it is emitted from the primary mirror. The expression is: y=-x and(θ out(1) +α1)+y c +x c and(θ out(1) +α1) (11) Introduce the expression of the secondary mirror in global coordinates: In the above formula, R′2, k2 and α2 are known parameters from the output set of the neural network, where; x * ,y * It can be expressed as: In the above formula, d′1 comes from the known parameters in the output set of the neural network; By combining the optical path expression (11) of the edge light in the central field of view after it is emitted from the primary mirror with the expression (12) of the secondary mirror in global coordinates, the coordinates of the edge point under the secondary mirror (x l ,y l ); Similarly, according to the object-image relationship and the principle of reversibility of the optical path, the light transfer matrix (9) between the reflectors is used to reversely trace the upper edge light of the (0, -FOV / 2) field of view to calculate the coordinates of the upper edge point of the primary mirror (x h ,y h ); It is known that the passing point (x h ,y h ) The angle at which the edge light of the (0, -FOV / 2) field of view is incident on the primary mirror is ω = -FOV / 2, so the optical path expression of the edge light of the (0, -FOV / 2) field of view is: (-FOV / 2)x-y+y h -(-FOV / 2)x h =0 (14) According to the obtained (x l ,y l ) and (0, -FOV / 2) the coordinates of the imaging point in the field of view (x s ,y s ), the optical path expression of the edge light between the secondary mirror and the image in the field of view (0, -FOV / 2) is obtained as follows: (and s -and l )x+(x s -x l )y+x s and l -x l and s =0 (15) Calculate the value from the lower edge point of the secondary mirror (x l ,y l ) The coordinates of the foot point (x1, y1) of the perpendicular line of the light incident on the primary mirror at the upper edge of the field of view (0, -FOV / 2) are: Calculate the value from the upper edge point of the primary mirror (x h ,y h ) The coordinates of the foot point (x2, y2) of the perpendicular line drawn between the secondary mirror and the image to the lower edge of the field of view (0, -FOV / 2) are:
