Optical design method and system for improving parallel course learning based on curvature attenuation

Abstract

The invention provides an optical design method based on curvature attenuation improved parallel curriculum learning, and belongs to the field of optical design, and the method comprises the steps: obtaining an initial structure of an optical system in a small field of view, and setting optimization variables which comprise the curvature radius of each lens, the air interval between the lenses, the lens center thickness and the lens material; taking the geometric radius or the root mean square radius of the point sequence diagram as an evaluation function, taking the small view field as a starting point, gradually increasing a view field angle according to a view field increment until a target full view field is reached, adjusting an optimization variable by adopting a damped least square method after the view field angle is increased every time, and in the process of adjusting the optimization variable, if ray tracing fails, determining that the target full view field is reached. The curvature radius of each lens is reduced to the set proportion of the current curvature radius of the lens until the evaluation function is smaller than a set threshold value; the invention further provides an optical design system. The problem that when a large number of lenses are optimized while curriculum learning is used, along with the increase of a view field, part of the lenses cannot transmit light smoothly, and optimization cannot be continued is solved.

Description

Technical Field

[0001] This invention relates to the field of optical design technology, and in particular to an optical design method and system for improving parallel learning based on curvature attenuation. Background Technology

[0002] As various industries impose increasingly stringent requirements on optical systems, existing optical design approaches often face challenges such as a limited number of reference patents and the need for extensive optical design experience. To reduce the manpower required for optical design, a growing number of intelligent starting point generation algorithms have been proposed, such as the Chinese patent application CN119511529A, "A Method for Designing the Initial Structure of an Optical System Based on First-Order Quantities." Existing heuristic global algorithms face the problem of lacking a stable system optimization direction, making them difficult to handle complex systems with many lenses. Mass parallel optimization of lenses, however, utilizes advanced GPUs and stable local optimization algorithms to simultaneously explore a large solution space. For systems with advanced performance metrics, obtaining a large number of optimization starting points is often difficult. Therefore, by employing a learning strategy, and following an optimization path from small to large field of view—from easy to difficult—a high-performance optical system can be obtained while rapidly optimizing the starting point.

[0003] When using courses to learn and optimize a large number of systems simultaneously, many optimization failures can occur, such as... Figure 1 As shown, many structures can achieve good imaging in a small field of view. However, when optical system design specifications are extremely stringent, as the field of view gradually increases, the configuration of some systems cannot allow light to propagate smoothly, thus losing the ability to continue optimization. Furthermore, during the simultaneous optimization of a large number of lenses, this will result in a significant waste of computational resources, substantially reducing algorithm efficiency and performance. Summary of the Invention

[0004] The technical problem to be solved by this invention is: how to solve the problem that when using course learning to optimize a large number of lenses at the same time, as the field of view increases, some lenses cannot transmit light smoothly, thus making it impossible to continue optimization.

[0005] This invention solves the above-mentioned technical problems through the following technical solution: an optical design method for improving parallel learning based on curvature attenuation, the method comprising: Obtain the initial structure of the optical system under a small field of view, and set optimization variables, including the radius of curvature of each lens, the air gap between lenses, the center thickness of the lens, and the lens material; Using the geometric radius or root mean square radius of the dot plot as the evaluation function, starting from a small field of view, the field of view angle is gradually increased according to the field of view increment until the target full field of view is reached. After each increase in the field of view angle, the damped least squares method is used to adjust and optimize the variables. If ray tracing fails during the adjustment and optimization process, the curvature radius of each lens is reduced to a set proportion of its current curvature radius until the evaluation function is less than the set threshold.

[0006] This invention starts with a small field of view and gradually increases the field of view angle according to the field of view increment. After each increase in the field of view angle, if ray tracing fails during the process of adjusting and optimizing variables, based on the optical physical principle that flat curvature reduces abrupt changes in light refraction at the lens interface, the radius of curvature of each lens is reduced to a set proportion of its current radius of curvature. This temporarily "flattens" the lens, reduces the local refractive gradient on the lens surface, and allows light to pass through more smoothly without interruption or divergence due to abrupt changes in curvature. This avoids ray tracing failure and solves the problem that when optimizing a large number of lenses simultaneously using a course, some lenses cannot transmit light smoothly as the field of view increases, making further optimization impossible.

[0007] Preferably, ray tracing failure means that edge rays or principal rays in a specific field of view cannot reach the image plane.

[0008] Preferably, the field of view of a small field of view is less than or equal to 5 degrees, and the field of view increment is Δi degrees, where Δi is less than 1 degree. The field of view increment should not be too large to avoid optimization failure due to drastic changes in aberrations.

[0009] Preferably, the set ratio is 80%.

[0010] Preferably, the lens material includes the material's refractive index and Abbe number.

[0011] Preferably, the threshold is set to be within n times the diameter of the Airy disk, where n is a number greater than 0.

[0012] This embodiment also provides an optical design system for improving parallel learning based on curvature attenuation, the system comprising: The initial structure acquisition unit is used to acquire the initial structure of the optical system under a small field of view and set optimization variables, including the radius of curvature of each lens, the air gap between lenses, the center thickness of the lens, and the lens material. The optimization execution unit is used to evaluate the geometric radius or root mean square radius of the dot plot, starting from a small field of view and gradually increasing the field of view angle according to the field of view increment until the target full field of view is reached. After each increase in the field of view angle, the optimization variables are adjusted using the damped least squares method. If ray tracing fails during the adjustment of the optimization variables, the curvature radius of each lens is reduced to a set proportion of its current curvature radius until the evaluation function is less than the set threshold.

[0013] Preferably, ray tracing failure means that edge rays or principal rays in a specific field of view cannot reach the image plane.

[0014] Preferably, the field of view of the small field of view is less than or equal to 5 degrees, the field of view increment is Δi degrees, and Δi is less than 1 degree.

[0015] Preferably, the set ratio is 80%.

[0016] The advantages provided by this invention are: 1. When ray tracing fails, this invention simply reduces the radius of curvature of each lens to a set ratio of its current radius of curvature, without adjusting the air gap between lenses, the center thickness of the lens, or the lens material. Since the material properties (such as refractive index) and gap distance remain unchanged, the optical characteristics of the system (such as focal length and aberration distribution) will not change significantly, and the image quality can be restored quickly by convergence.

[0017] 2. This invention employs an optical design strategy of "from simple to complex, progressive optimization." Starting with a defined first-order small field-of-view structure, it gradually expands the field of view (step size) and iteratively applies the least squares method to optimize variables, using a dot plot as the core criterion while strictly monitoring the reliability of ray tracing. If ray tracing fails, the configuration is immediately repaired through curvature reduction. Ultimately, while maintaining focal length stability, effective control of image quality across the entire field of view is achieved. This method is particularly suitable for optical systems where aberrations vary significantly with the field of view, such as wide-angle eyepieces and large relative aperture objectives. Attached Figure Description

[0018] Figure 1 This addresses a scenario where optimization fails when using existing technology to learn from courses and simultaneously optimize a large number of shots. Figure 2 A flowchart of the optical design method for parallel course learning based on curvature attenuation provided in Embodiment 1 of the present invention; Figure 3 The diagram shows the light propagation path effect of the optical design method based on curvature attenuation-based improved parallel course learning provided in Embodiment 1 of the present invention. Figures 4(a) to 4(i) are comparative schematic diagrams of the parallel course learning optimization experiment with 9 randomly generated initial optical structures, where blue indicates that curvature attenuation is not enabled, and orange indicates that the optical design method of improving parallel course learning based on curvature attenuation of the present invention is enabled. Figure 5 A comparative diagram showing the results of a large-scale parallel optimization experiment targeting the "10+2" lens structure; Figure 6 This is a comparative diagram showing the results of a large-scale parallel optimization experiment conducted on the "6+3" lens structure. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] Example 1 like Figure 2 As shown, this embodiment provides an optical design method for improving parallel curriculum learning based on curvature attenuation, including the following steps: Step 1: Obtain the initial structure of the optical system under a small field of view. The field of view angle of the small field of view is less than or equal to 5 degrees; in this embodiment, the field of view angle is 0 degrees. Set optimization variables, including the radius of curvature of each lens, the air gap between lenses, the center thickness of the lens, and the lens material. The lens material includes the refractive index and Abbe number. Adjusting the radius of curvature of the lenses controls the light deflection, while the air gap between lenses and the center thickness of the lens affect the optical path length and the image plane position.

[0021] In practical applications, the Chinese patent application CN119511529A, entitled "Method for Designing the Initial Structure of an Optical System Based on First-Order Quantities," can be used to obtain the initial structure of an optical system in a small field of view (e.g., 0-degree field of view). The core of this method lies in utilizing primary aberration theory and the first-order characteristics of Gaussian optics (such as focal length and pupil position) to determine the system's basic parameters through analytical calculations. This ensures that the system possesses good basic imaging capabilities within an extremely small field of view. After the first-order quantity design is completed, the system should already have the target focal length.

[0022] Step 2: Using the geometric radius (Geo Radius) or root mean square (RMS) radius of the spot diagram as the evaluation function, the optimization objective is to control the spot diagram size within a certain threshold, indicating that the energy concentration meets the requirements. Starting with a small field of view, the field of view angle is gradually increased according to the field of view increment. The field of view increment is set as needed, and it should not be too large to avoid optimization failure due to drastic changes in aberrations. In this embodiment, the field of view increment is Δi degrees, where Δi is less than 1 degree, until the target full field of view is reached. After each increase in the field of view angle, the optimization variables are adjusted using the damped least squares method in optical design software such as Zemax and CodeV. During the adjustment of optimization variables, if ray tracing failure occurs, the curvature radius of each lens is reduced to a set proportion of its current curvature radius to mitigate light deflection and improve optical path continuity. The set proportion is set as needed; in this embodiment, the set proportion is 80%, until the evaluation function is less than the set threshold. Ray tracing failure means that edge rays or principal rays of a specific field of view cannot reach the image plane. The threshold is set to n times the diameter of the Airy disk, where n is a number greater than 0.

[0023] For ease of understanding, this embodiment uses a small field of view of 0 degrees, a field of view increment of Δi degrees, and a target full field of view of 60 degrees as an example to illustrate the optical design method of the present invention, where Δi is less than 1 degree: Starting from a field of view angle of 0 degrees, the field of view angle is increased according to the field of view increment to obtain the current field of view angle of Δi degrees. At this time, the optical design software uses the damped least squares method to simultaneously adjust the curvature radius of each lens, the air gap between lenses, the center thickness of the lens, and the lens material. It judges whether the geometric radius or root mean square radius of the dot plot is less than a set threshold. During the process of simultaneously adjusting the curvature radius of each lens, the air gap between adjacent lenses, the center thickness of the lens, and the lens material at the current field of view angle, if ray tracing failure occurs, such as edge rays or principal rays of a specific field of view failing to reach the image plane, even if the current dot plot size is within an acceptable range, the current optimization must be deemed invalid. At this time, only the curvature radius of each lens is reduced to a set proportion of its current curvature radius, without adjusting the air gap between lenses, the center thickness of the lens, and the lens material, until ray tracing failure no longer occurs. For example, if the curvature radius of each lens is reduced to 80% of its current curvature radius, ray tracing will not fail. In the next adjustment, the current curvature radius of each lens has become 80% of the original value. At this time, the damped least squares method is used again to adjust the current curvature radius, the air gap between lenses, the center thickness of the lens, and the lens material until the geometric radius or root mean square radius of the dot plot is less than the set threshold.

[0024] Increment the current field of view by the field of view increment to obtain the next field of view angle as 2. For Δi degrees, following the above process, simultaneously adjust the radius of curvature of each lens, the air gap between lenses, the center thickness of the lens, and the lens material. If ray tracing fails during adjustment, simply reduce the radius of curvature of each lens to a set proportion of its current radius of curvature, without adjusting the air gap between lenses, the center thickness of the lens, or the lens material. Then, reapply the damped least squares method to simultaneously adjust the current radius of curvature, the air gap between lenses, the center thickness of the lens, and the lens material until the geometric radius or root mean square radius of the dot plot is less than a set threshold. Repeat this process until the target full field of view is achieved.

[0025] Ray tracing failure typically indicates configurational problems such as: lens edge or center thickness being too thin, leading to physical impossibility or optical path obstruction; excessively extreme curvature causing total internal reflection or a sharp deterioration of aberrations; and unreasonable spacing between elements causing interruption of light propagation. In such cases, configurational repair strategies should be adopted. This invention proposes a curvature attenuation strategy, based on a key observation: when the curvature of the lens surface is relatively flat, light passes through the lens more easily without tracing errors. This phenomenon stems from the principle of optical physics—flat curvature reduces abrupt refraction changes at the lens interface, lowering the risk of light path interruption and thus ensuring stable light propagation.

[0026] When the system detects a configuration problem (such as light failing to pass through a specific lens) during the field of view expansion, it automatically reduces the curvature of each lens in the system to a predetermined proportion of its current radius of curvature, for example, 80%. This operation temporarily "flattens" the lenses but does not change other key parameters (air gap between adjacent lenses, lens center thickness, and lens material). This strategy is similar to setting up a "safety net" on an optimized path to help the system escape the erroneous state.

[0027] After the curvature is reduced, the light propagation path is restored, avoiding tracking failure. This is because the planarization process reduces the local refractive gradient on the lens surface, allowing light to pass through more smoothly without interruption or divergence caused by abrupt changes in curvature. More importantly, since material properties (such as refractive index) and gap distance remain unchanged, the system's optical characteristics (such as focal length and aberration distribution) do not undergo significant shifts. This allows the optimization process to quickly converge back to the original path and restore image quality in subsequent stages. For example, after curvature attenuation, the system can regain a stable state within several generations of optimization, avoiding disruption to the entire optimization process. Figure 3 The effect of curvature attenuation is visually demonstrated: the lens surface becomes flatter, the light propagation path is smoother and more continuous, and the probability of errors is reduced.

[0028] Therefore, during the course, the algorithm can monitor the optimization status of each system. When a system loses its ability to continue optimizing due to an unreasonable configuration as the field of view increases, the curvature of each lens in the system is reduced to 80% of its original value, temporarily "flattening" the lenses. The reduction in curvature allows the light propagation path to escape the tracking failure state, enabling optimization to continue. In addition, since the properties of the glass material and the air gap remain unchanged, the optical characteristics of the system will not change significantly, thus allowing for rapid convergence and restoration of image quality.

[0029] Example 2 This embodiment provides an optical design system based on curvature attenuation to improve parallel curriculum learning, including: The initial structure acquisition unit is used to acquire the initial structure of the optical system under a small field of view, where the field of view angle is less than or equal to 5 degrees; in this embodiment, the small field of view is 0 degrees. Optimization variables are set, including the radius of curvature of each lens, the air gap between lenses, the center thickness of the lens, and the lens material; the lens material includes the refractive index and Abbe number.

[0030] The optimization execution unit uses the geometric radius or root mean square radius of the dot plot as the evaluation function. Starting from a small field of view, it gradually increases the field of view angle according to the field of view increment. The field of view increment is set as needed; in this embodiment, the field of view increment is Δi degrees, where Δi is less than 1 degree. This continues until the target full field of view is reached. After each increase in the field of view angle, the optimization variables are adjusted using damped least squares. During the adjustment of the optimization variables, if ray tracing fails, the curvature radius of each lens is reduced to a set proportion of its current curvature radius. The set proportion is set as needed; in this embodiment, the set proportion is 80%, until the evaluation function is less than a set threshold. Ray tracing failure refers to the inability of edge rays or principal rays of a specific field of view to reach the image plane. The set threshold is n times the Airy disk diameter, where n is a number greater than 0.

[0031] Experimental Analysis To verify the effectiveness of the curvature attenuation strategy, this invention selected nine randomly generated initial optical structures for parallel curriculum learning optimization experiments. During the experiments, the system's field of view was gradually expanded according to the curriculum learning strategy (from small to large field of view), and the optimization status was monitored in real time. The results are shown in Figure 4. Figures 4(a) to 4(i) show the parallel curriculum learning optimization experiments conducted with the nine randomly generated initial optical structures. Blue indicates that curvature attenuation was not enabled: approximately 60% of the systems crashed due to ray tracing failure in the mid-to-late stages of field of view expansion, causing the evaluation function value (Loss) to spike and stagnate, resulting in the interruption of the optimization process.

[0032] Orange indicates that curvature attenuation of this invention is enabled: when the system detects a tracing failure, curvature attenuation is automatically triggered (curvature is reduced to 80% of its original value). After the lens is flattened, the light propagation path quickly returns to stability, and the system successfully converges to the low-loss region in subsequent optimizations. Ultimately, over 90% of the system achieves excellent image quality, verifying the strategy's ability to salvage complex configurations.

[0033] To further quantify the benefits of the strategy, this invention conducted a large-scale parallel optimization experiment with 20,000 independent systems for two typical lens structures ("10+2" and "6+3"). The results show that the curvature attenuation strategy significantly improves the optimization success rate and final performance. Figure 5 In the "10+2" structure: without curvature decay, only 35% of the systems successfully converged (dark blue area), and most systems suffered high losses due to tracing failures (light blue area). After enabling curvature decay (orange area), the success rate increased to 75%, the proportion of high-loss systems decreased sharply, and the overall loss distribution shifted significantly downward. Figure 6 The "6+3" structure is more sensitive to curvature changes, with a failure rate as high as 70% (Deep Blue) without a strategy. Curvature decay doubles the success rate (to 68%), and the convergent system has a lower and more concentrated loss value, confirming the universality of the strategy for different structures.

[0034] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An optical design method for improving parallel learning based on curvature attenuation, characterized by: The methods include: Obtain the initial structure of the optical system under a small field of view, and set optimization variables, including the radius of curvature of each lens, the air gap between lenses, the center thickness of the lens, and the lens material; Using the geometric radius or root mean square radius of the dot plot as the evaluation function, starting from a small field of view, the field of view angle is gradually increased according to the field of view increment until the target full field of view is reached. After each increase in the field of view angle, the damped least squares method is used to adjust and optimize the variables. If ray tracing fails during the adjustment and optimization process, the curvature radius of each lens is reduced to a set proportion of its current curvature radius until the evaluation function is less than the set threshold.

2. The optical design method for improving parallel curriculum learning based on curvature attenuation according to claim 1, characterized in that: Ray tracing failure means that edge rays or principal rays in a particular field of view cannot reach the image plane.

3. The optical design method for improving parallel curriculum learning based on curvature attenuation according to claim 1, characterized in that: A small field of view has a field angle of less than or equal to 5 degrees, and the field increment is Δi degrees, where Δi is less than 1 degree.

4. The optical design method for improving parallel curriculum learning based on curvature attenuation according to claim 1, characterized in that: The set ratio is 80%.

5. The optical design method for improving parallel curriculum learning based on curvature attenuation according to claim 1, characterized in that: Lens materials include the material's refractive index and Abbe number.

6. The optical design method for improving parallel curriculum learning based on curvature attenuation according to claim 1, characterized in that: The threshold is set to n times the diameter of the Airy disk, where n is a number greater than 0.

7. An optical design system for improving parallel learning based on curvature attenuation, characterized in that: The system includes: The initial structure acquisition unit is used to acquire the initial structure of the optical system under a small field of view and set optimization variables, including the radius of curvature of each lens, the air gap between lenses, the center thickness of the lens, and the lens material. The optimization execution unit is used to evaluate the geometric radius or root mean square radius of the dot plot, starting from a small field of view and gradually increasing the field of view angle according to the field of view increment until the target full field of view is reached. After each increase in the field of view angle, the optimization variables are adjusted using the damped least squares method. If ray tracing fails during the adjustment of the optimization variables, the curvature radius of each lens is reduced to a set proportion of its current curvature radius until the evaluation function is less than the set threshold.

8. The optical design system for improving parallel curriculum learning based on curvature attenuation according to claim 7, characterized in that: Ray tracing failure means that edge rays or principal rays in a particular field of view cannot reach the image plane.

9. The optical design system for improving parallel learning based on curvature attenuation according to claim 7, characterized in that: A small field of view has a field angle of less than or equal to 5 degrees, and the field increment is Δi degrees, where Δi is less than 1 degree.

10. The optical design system for improving parallel curriculum learning based on curvature attenuation according to claim 7, characterized in that: The set ratio is 80%.

Images