• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    Methods and apparatus for designing optical systems are provided. In doing so, machine learning is performed (21) on a variety of known optical systems to train a computing device. After this training, the computing device may generate (23) a design for an optical system based on parameters describing desired properties of an optical system (22).
    公开号:CH714206A2
    申请号:CH01081/18
    申请日:2018-09-13
    公开日:2019-03-29
    发明作者:Toufic Jabbour Dr；Christian Wojek Dr；Christoph Menke Dr；Schwab Markus
    申请人:Zeiss Carl Ag；Zeiss Carl Smt Gmbh；
    IPC主号:
    专利说明:

    Description The present application relates to methods and devices for designing optical systems.
    The design of optical systems such as camera lenses or microscope lenses is a complex process that involves several stages. In general, the technical requirements for the optical system are first analyzed. On the basis of the technical requirements, a first draft of an optical system is determined, which has free parameters. These parameters are then optimized.
    For these parameter optimizations, there are powerful software programs that perform these parameter optimizations of an initial design very quickly and efficiently. However, these available software programs have two disadvantages:
    First, they can only optimize the initial design, but cannot change it structurally without human intervention.
    Second, the success of the optimization depends heavily on the initial design.
    This means that if the initial design does not have a suitable structure, the optimization of the system by the software programs cannot produce satisfactory results. For example, the software programs cannot add or remove bullets to improve the design.
    Some optical system design software programs offer a collection of well-corrected systems from which a user can choose an initial design that is well corrected (for example, for aberrations). However, this is only of limited help, since requirements for new optical systems often do not match the systems offered by the software programs.
    [0008] Therefore, there are various approaches for obtaining an initial design, also known as a start design, for an optical system. One approach is described in David Shafer, A simple method for designing lenses Proc.SPIE 023, 1980, International Lens Design Conference, 234 (September 16, 1980). This approach combines only two special types of surfaces, aplanates and concentric surfaces, to create different systems.
    This method was then expanded, as in I. Antripova Livshits, Simple method forcomputer-aided lens design with the elements of artificial intelligence, Proc.SPIE Voi. 1780, pages 210-213, published 04/1993 or I. Livshits et al., Information technologies in CAD System for lens design Proc.SPIE 7506, International Conference on Optical Instruments and Technology, November 25, 2009. Here the number of different surfaces is increased to five: aplanat, concentric with respect to a pupil, concentric with respect to the image, flat and close to the image. The authors of this approach divide the combination of two types of surfaces into four groups: basic, correction, wide angle and light powerful. An initial design for an optical system can then be described by a sequence of these groups, such as shown in the references above.
    This type of functional design offers a good starting point for a final design, and starting from such a design, optimization can then be carried out with the software packages mentioned above.
    [0011] However, this approach only helps to find an initial design to a limited extent. The classification of different elements mentioned helps to understand the function of each element, but does not give any indication as to how an optical system should ultimately be put together for a specific purpose. The references state that expert knowledge, which is acquired through long experience in the field of optical system design, is used here. However, this requires such long-term knowledge and does not allow the automation of the design of optical systems.
    Another approach is described by Donald Dilworth (see, for example, Applications of Artificial Intelligence to Computer-Aided Lens design, SPIE Voi. 766 Recent Trends in Optical Systems Design; Computer Lens Design Workshop (1987)), in which wave fronts within optical systems are defined, with the wavefronts serving as interfaces so that parts of optical systems can be combined with one another. However, this only allows the combination of parts of optical systems if the corresponding wavefronts are present.
    It is therefore a task to improve the automation of the creation of initial designs of optical systems, which can then be further optimized. This object is achieved by a method according to claim 1 or 10, a computer program according to claim 12 and an apparatus according to claim 13 or 15. The subclaims define further embodiments.
    According to the invention, there is provided a method for designing optical systems, comprising:
    Providing a variety of optical systems, and
    Carrying out a machine learning process based on the large number of optical systems.
    In this way, a device can be trained on the basis of known optical systems in order to be able to subsequently provide designs of optical systems. The large number of optical systems thus serves as a large number of training examples for the machine learning process. Optical systems as training examples can be simulated or with the aid of a sampling method (see, for example, reinforcement learning: An Introduction, Richard S. Sutton
    CH 714 206 A2 and Andrew G. Barto, Second Edition, in progress MIT Press, Cambridge, MA, 2017, online http://incompleteideas.net/sutton/book/bookdraft2017june19.pdf. As of September 26, 2017). In addition, known design rules for the design of optical systems for the production of optical systems can be used as training examples or methods of imitation learning can be used that derive rules by observing an expert (see for example IEEE ROBOTICS & AUTOMATION MAGAZINE, VOL. 17 , NO. 2, JUNE 20101 Imitation and Reinforcement Learning Practical Learning Algorithms for Motor Primitives in Robotics by Jens Kober and Jan Peters).
    The provision of the large number of optical systems can thus take place in the form of a database, on the basis of which the learning method is then carried out. However, in addition or as an alternative, the provision of the plurality of optical systems can also include designing the plurality of optical systems by an optical designer, the learning method then being a method of imitative learning that, as it were, observes the optical designer during the design and thus trains the device accordingly.
    Performing machine learning may include training an artificial neural network. Neural networks are particularly preferred because they represent a well-understood, trainable system. An execution of the machine learning process can include neural networks with a multitude of layers, which are often used in practice.
    As a learning method, methods of reinforcement learning (e.g. reinforcement learning), e.g. Deep reinforcement learning methods are used to change the basic configuration of optical surfaces of an optical system in order to arrive at the optical systems that are used as training examples, for example in the database described above.
    [0019] Providing a plurality of optical systems may include classifying the plurality of optical systems to provide parameters that describe the plurality of optical systems, wherein machine learning is performed based on the parameters.
    By classifying the known designs of optical systems for machine learning are prepared. A device, in particular a device with a neural network, is then appropriately trained by machine learning, so that the device itself can then create drafts of optical systems. Further information can be found in the Wikipedia article Machine Learning Who English-language Wikipedia, as of July 31, 2017, and the article Artificial Neural Networks, also as of July 31, 2017, each with additional evidence. Existing optical systems with their properties are therefore used to train a device with machine learning algorithms, in particular a neural network. Various conventional approaches to machine learning can be used, such as learning through imitation and (deep) reinforcement learning, as mentioned in the Wikipedia article above.
    After training with this method, the trained device can then be given the desired parameters of an optical system to be designed, and the device then delivers a corresponding design that best meets the required system specifications.
    For the learning method and / or when using the trained device, a metric can be used, which is used to evaluate optical systems with regard to their properties. This metric can be based on Delano diagrams of the optical systems.
    The use of Delano diagrams can simplify the evaluation and determination of parameters of the optical system.
    The parameters may include first parameters that describe components of the plurality of optical systems. In this way, the first parameters are used to describe the optical systems, in particular individual components such as lenses thereof.
    The parameters can also include second parameters that describe the optical and mechanical properties of the large number of optical systems. The second parameters thus classify the optical systems according to their properties.
    There is also provided a method of designing an optical system, comprising: supplying parameters describing desired properties of the optical system to a device trained with the method as discussed above, and obtaining a design of an optical system from the device.
    Thus, an initial design for an optical system with desired parameters can be provided in an automated manner, which can then be further optimized using conventional optimization methods, for example the software packages mentioned at the beginning. The parameters can describe the optical system in the same way as the above-mentioned second parameters. The control of the individual optimization steps of the optimization method, in particular the addition or removal of optical active surfaces, can be controlled by a method based on (deep) reinforcing learning, which can be trained in an embodiment, for example, by methods of imitative learning.
    According to the invention, a computer program is also provided with a program code which, when it is carried out on a processor, carries out the method as described above.
    CH 714 206 A2 According to the invention, a device for designing optical systems is also provided, comprising a computing device which is set up to carry out machine learning on the basis of a large number of optical systems.
    [0030] The device can be set up to carry out the method as described above.
    Furthermore, a device is provided, comprising a computing device trained according to one of the above methods. The statements regarding the methods apply accordingly to the device.
    The invention is explained below using exemplary embodiments with reference to the accompanying drawings. Show it:
    Fig. 1 2 shows a block diagram of a device according to an exemplary embodiment, Fig. 2 2 shows a flowchart of a method according to an exemplary embodiment, 3A and 3B Diagrams to illustrate a classification of different optical systems, Figure 4A an optical system for the illustration of Delano diagrams, Figure 4B 4 shows a Delano diagram of the system of FIG. 4A, Fig. 5 a diagram illustrating the extraction of parameters from Delano diagrams, and 6A-6D Delano diagrams for different types of lenses for further illustration.
    [0033] The invention is explained in detail below using various exemplary embodiments. These exemplary embodiments are only illustrative and are not to be interpreted as restrictive. In particular, other exemplary embodiments can have different features and components than those shown.
    Features, elements or components of different exemplary embodiments can be combined with one another to form further exemplary embodiments. Variations and modifications that are described with respect to one of the exemplary embodiments can also be applied to other exemplary embodiments. In particular, variations and modifications which are discussed for exemplary embodiments of methods can also be applied to corresponding devices and vice versa.
    1 shows a device according to an embodiment. The device in FIG. 1 comprises a computing device 10, for example a computer or also a plurality of computers connected via a network. The computing device 10 can also include specific hardware components such as, for example, user-specific integrated circuits (A6) in order to implement the functions described here. In particular, the computing device 10 in the exemplary embodiment in FIG. 1 implements one or more artificial neural networks 11. However, exemplary embodiments are not limited to neural networks.
    [0036] The computing device 10 can receive input data and output data. In a training phase, the computing device 10 receives a variety of designs e.g. in the form of first parameters, which describe the components of the optical systems (e.g. shape, type, material and position of lenses or mirrors) and second parameters, which describe properties of the optical systems (e.g. focal length, aperture etc.). On the basis of this input data, the neural network 11 is trained in a manner known per se, in particular in order to create links between the first parameters and the second parameters, i.e. between the structure of the optical systems and the resulting property. In other embodiments, the input data describes design processes by optics designers, and the neural network then learns to design optical systems using learning methods of mimicking learning. Details of this learning will be explained in more detail later.
    [0037] After this training phase, the computing device 10 can be supplied with desired parameters, which describe the properties of an optical system to be designed, as input data. These parameters correspond in their function to the second parameters above. The computing device 10 then outputs as output data a design for a corresponding optical system by means of the then trained neural network 11.
    FIG. 2 shows a flowchart of a corresponding method which can be carried out in the computing device of FIG. 1.
    In step 20, optional optical systems are classified to give respective first and second parameters of the optical systems in a standardized form, i.e. in a form applicable to all optical systems. The components of optical systems can be specified according to a conventional lens classification, for example according to the table below:
    CH 714 206 A2
    notation Surname J Aperture (also called speed gravel optical system) W angle range F focal length L spectral Q picture quality S Back focal length D Entrance pupil position
    The optical system and individual elements of the optical systems can also be classified according to their purpose. For example, B designates basic elements that determine the basic optical properties of the optical system. K K denotes correction elements which are used to correct image errors such as chromatic operations of the basic elements. Designated with S Verden wide-angle elements, which are used to determine a picture angle of the optical system. C H denotes elements which are used to define the aperture of the optical system, in particular to enlarge it.
    The above letter names serve only as an example, and the classification can also be done with other symbols, numbers, characters and the like. In addition, optical systems or their elements can be classified according to the type of optical surfaces, for example aplanar surface close to the image, surface concentric around a main beam, surface concentric around an edge beam or flat surface. These classifications are used to determine first and second parameters that describe the respective optical system.
    These parameters will be explained in more detail later with reference to FIGS. 3-6. In particular, Delano diagrams can be used for classification, as will be explained in more detail later. The optical systems and / or the parameters are then preferably provided in a database.
    In step 21, the optical systems thus classified, i.e. the optical systems with their assigned parameters, used for machine learning. For example, the neural network 11 of FIG. 1 can be trained here. Other types of machine learning as discussed above can also be used. Possibilities for this machine learning are now explained in more detail.
    For machine learning as well as for designing optical systems, which follows later in steps 22 and 23, a metric is required with which the quality of an optical system can be assessed and which can be used accordingly for optimization methods of optical systems. Aberrations in an optical system are caused by reflection or refraction of light rays through optical elements of the optical system. A metric that can be used in exemplary embodiments is a metric that measures the total deflection angle of all optical elements in a representation of the optical system in a Delano diagram. This is e.g. in Automatic generation of optical initial configuration based on Delano diagram, Kai-Yuan Zhang et al., 2016 Res. Astron. Astrophys. 16007. Delano diagrams will be explained in more detail later.
    In exemplary embodiments, machine learning is based on methods of reinforcing learning, in particular of (deep) reinforcing learning. This learning method can include teaching the neural network 11 or another system to be trained to reproduce the already known optical systems of step 20. For example, an optical system with the second parameters described above can be selected, and the goal is to reproduce this optical system through the neural network in an optimization process with the metric described above. The manipulation / modification of the optical system through the optimization process (also referred to as policy) can be carried out using one or more of the following variants.
    Variant IIK Random Sampling II In this variant, the one is selected from the available optical systems that best fulfills the second parameters of the system to be reproduced, the system to be reproduced being excluded from the selection. For the optimization, various components are then randomly removed, added or permuted in accordance with the first parameters in order to minimize the metric described above. The random distribution from which new systems are drawn can take different forms (see e.g. Reinforcement Learning: An Introduction, Richard S. Sutton and Andrew G. Barto, Second Edition, in proqress MIT Press, Cambridge, MA, 2017, online http://incompleteideas.net/sutton/book/bookdraft2017june19.pdf. As of 09/26/2017).
    Variant 2: Database rules: Here too, as in variant 1, the one that is selected from the available optical systems that best meets the second parameters of the system to be reproduced is selected. Instead of random changes, changes are made here according to database rules. In other words
    CH 714 206 A2
    Permutations of the first parameters selected from the many optical systems that are made available for training in a database.
    Variant 3: IKImitation Learning linier the neural network is trained by following design steps which are carried out by experienced optics designers of optical systems (see for example IEEE ROBOTICS & AUTOMATION MAGAZINE, VOL. 17, NO. 2, JUNE 20101 Imitation and Reinforcement Learning Practical Learning Algorithms for Motor Primitives in Robotics by Jens Kober and Jan Peters). Each design step that is carried out by the designer can, for example, be transferred to a Delano diagram and thus be followed by the network to be trained.
    It should be noted that the trained network can be used through a self-observation process to identify design rules. In particular, with a correspondingly large number of training examples, design rules can also be found which are not yet known to designers of optical systems.
    With step 21, the training phase is then completed. Then the device trained in this way can then be used according to steps 22 and 23 for designing optical systems. In step 22, desired parameters for the optical system, such as aperture, focal length, image quality and the like, are specified as explained above. In step 23, an optical system is then output, for example, using the trained neural network 11. This optical system can then, as explained at the outset, be optimized as an initial draft for an optimization process by means of the conventional software packages mentioned at the outset. As soon as the neural network is trained, the network therefore knows design rules and can use these rules in conjunction with an optimization method to output a KDesign vector with first parameters that describe an optical system that fulfills conditions that are met by second parameters be specified. The same metric, in particular the metric described above, that was also used during training can be used for this.
    The classification of step 20 of FIG. 2 is explained in more detail below with reference to FIGS. 3-6. 3A and 3B show different types of optical systems, classified according to numerical aperture (NA) and half field of view (FOV field of view). Numerical apertures and image angles are examples of the second parameters mentioned above. FIG. 3A shows refractive arrangements (essentially arrangements based on lenses), and FIG. 3B shows reflecting arrangements, i.e. Arrangements based on mirrors. In addition to the numerical aperture, the f-number is also given, which specifies the ratio of the focal length to the diameter of the effective entrance pupil and is often used in photography, for example, to identify the light intensity of a lens. As can be seen, a large number of different optical systems exist over a further area of numerical apertures and image angles. All of these optical systems exist in a large number of different variants. All of these systems can then be classified and used with first and second machine learning parameters as discussed.
    Parameters of optical systems, in particular the first parameters mentioned, can advantageously be obtained from Delano diagrams. In addition, as explained above, a metric can be created on the basis of the Delano diagrams, with which optical systems can be evaluated both during machine learning (step 21) and during the automatic creation of a draft (step 23). In a Delano diagram, essentially the distance of an edge ray (English marginal ray) from an optical axis is plotted against the distance of the main ray (chief ray) from the optical axis. This will now be explained with reference to FIGS. 4A and 4B.
    4A shows an example of an optical system with an entrance lens 40, which defines an entrance pupil, a field lens 41 and a collimator lens 42. An edge ray 43 and a main ray 45 and their distances y, ÿ to the optical axis 44 are also shown.
    4B shows a corresponding Delano diagram, in which y is plotted over.. A section 46 characterizes the beam path up to the lens 40. At the position of the lens 40, the light is refracted, which is reflected in a change in direction of the section 47 relative to the section 46. A further light refraction takes place in the lens 41, which leads to a further change in direction in accordance with the section 48. A last light refraction takes place at the lens 42, which then leads to a course corresponding to the section 49. The changes in direction during refraction provide information as to whether a positive refractive power (for example beam focusing) or a negative refractive power (for example beam expansion) represents. In the simple example of FIG. 4B, it is assumed that the lenses are thin, so that the light refraction of both lens surfaces has been combined in one point. With thick lenses, the course within the lens is also visible in the Delano diagram.
    This is illustrated in Fig. 5. Here a line 50 at different angles 52, 54 and 56 is changed to lines 51, 53 and 55 by a corresponding optical element such as a lens. In the case of line 51, the assigned angle 52 is directed counterclockwise, which corresponds to a negative refractive power, in this case because of the comparatively small angle 52, a comparatively small negative refractive power. In the case of line 53, the direction change takes place clockwise according to the angle 54, which corresponds to a small positive refractive power. In the case of line 55, the direction change takes place by a comparatively large angle 56 clockwise, which corresponds to a comparatively large positive refractive power. In this way, the individual components, for example lenses, of the optical system can be analyzed and classified on the basis of the Delano diagram in order to determine first parameters. In particular, different types of lenses or others
    CH 714 206 A2 or refractive or reflective elements can be distinguished on the basis of the above angles, which are an example for first parameters. Examples of different lens types are shown in FIGS. 6A-6D along with associated Delano diagrams.
    6A shows a Delano diagram for a comparatively thick lens 61 with two kinks, which reflect the surfaces of the lens 61. 6D, on the other hand, shows the Delano diagram for a thin lens 64, in which essentially a single kink (or two kinks lying very closely next to one another) occur. 6B and 6C show Delano diagrams for meniscus lenses 62 and 63, respectively, which are curved in different directions. In this case, a light beam incident from the left in the figure is assumed. As can be seen in the Delano diagrams, one surface of meniscus lenses has a positive refractive power, while the other surface of the lens has a negative refractive power, as can be seen from the changes in direction in the Delano diagram (see the explanation for FIG. 5) ,
    In this way, different optical systems can be classified (evaluated) using Delano diagrams. In particular, the different surfaces and elements of the system can be classified as mentioned by the respective angle through which the line changes in the Delano diagram. Using an appropriate computer program, each optical system can be transferred from a set of known optical systems to a representation as a Delano diagram, and then every surface in the system can be classified with regard to refractive power, optical aberrations, position with respect to a pupil, etc. On this basis, the type, function and position of each optical element in the system can then be characterized and the respective system can thus be characterized with corresponding first parameters as explained above. The Delano diagram thus provides a description of the optical system based on the description of the components and parameters of the optical system, which can then be used as a feature vector for machine learning.
    In this way, optical systems can be described in a unified manner, which enables the subsequent machine learning, for example the training of the neural network 11.
    权利要求:
    Claims (15)
    [1]
    claims
    1. A method of designing optical systems comprising:
    Providing a variety of optical systems, and
    Carrying out a machine learning process based on the large number of optical systems.
    [2]
    2. The method of claim 1, wherein performing the machine learning method comprises training an artificial neural network.
    [3]
    3. The method of claim 2, wherein the neural network comprises multiple layers.
    [4]
    4. The method according to any one of claims 1-3, wherein providing a plurality of optical systems comprises classifying the plurality of optical systems to provide parameters that describe the plurality of optical systems, wherein machine learning is performed based on the parameters ,
    [5]
    5. The method according to claim 4, wherein the classification is based on Delano diagrams of the plurality of optical systems.
    [6]
    6. The method according to any one of claims 4 or 5, wherein the parameters include first parameters that describe components of the plurality of optical systems.
    [7]
    7. The method according to any one of claims 4-6, wherein the parameters include second parameters that describe optical properties of the plurality of optical systems.
    [8]
    8. The method according to any one of claims 1-7, wherein the machine learning method comprises a method of reinforcing learning.
    [9]
    9. The method according to any one of claims 1-8, wherein providing the plurality of optical systems comprises designing the plurality of optical systems by an optical designer, the machine learning method comprising a method of mimicking learning based on the design.
    [10]
    10. A method for designing optical systems comprising:
    Supplying parameters, which describe desired properties of the optical system, to a device trained with the method according to one of claims 1-9, and
    Obtaining an optical system design from the device.
    [11]
    11. The method according to any of claims 1-10, wherein the machine learning method and / or obtaining a design comprises applying a metric based on Delano diagrams.
    [12]
    12. Computer program with a program code which, when it is carried out on a processor, carries out the method according to one of claims 1-11.
    [13]
    13. Device for designing optical systems, comprising a computing device which is set up to carry out machine learning on the basis of a large number of optical systems.
    CH 714 206 A2
    [14]
    14. The device according to claim 13, wherein the device is set up to carry out the method according to one of claims 1-11.
    [15]
    15. Device for designing optical systems, comprising a computing device trained according to one of claims 1-11.
    CH 714 206 A2
    类似技术:
    公开号 | 公开日 | 专利标题
    DE102017220104A1|2018-05-24|Lens system with variable focal length with multi-level image processing with extended depth of focus
    EP3172611B1|2020-01-29|Method and device for imaging an object
    DE102017220101A1|2018-05-24|Inspection system using machine vision to obtain an image with extended depth of field
    WO2018177680A2|2018-10-04|Structured illumination with optimized illumination geometry
    DE102004030961A1|2006-02-02|A method of determining a matrix of transmission cross-coefficients in an optical approximation correction of mask layouts
    DE102014210641B4|2020-12-10|Test object, use of a test object and device and method for measuring the point spread function of an optical system
    CH714206A2|2019-03-29|Methods and apparatus for designing optical systems.
    EP1193687A2|2002-04-03|Speaker adaptation for a speech recognition
    EP3593190B1|2021-05-05|3d microscopy
    DE102014223326A1|2016-05-19|Method for predicting at least one illumination parameter for evaluating a lighting setting
    DE102014114864B4|2016-06-02|Method and apparatus for determining a lateral offset of a pattern on a substrate relative to a desired position
    WO2018172271A1|2018-09-27|Wavefront correction using adaptive optical element
    DE102013219524B4|2018-02-08|Device and method for determining the imaging quality of an optical system and optical system
    DE2608176A1|1977-09-01|Uniform illumination of object for microscopy etc. - using coherent light split by lens raster to set of parallel beams
    DE102015219936A1|2016-09-01|Method for producing a curved optical mirror surface
    DE102020106857A1|2021-09-16|MICROSCOPY SYSTEM AND METHOD FOR PROCESSING MICROSCOPE IMAGES
    DE102019134217A1|2021-06-17|Method for configuring an automated microscope and means for carrying it out and microscope system
    DE10250503A1|2004-05-19|Microscope system and method for the detection and compensation of changes in a recorded image content
    DE202019102339U1|2019-05-21|Device for generating an image
    WO2017060296A2|2017-04-13|Method and device for beam analysis
    EP3712823A1|2020-09-23|Method and device for controlling a technical appliance using a program code
    DE102018222147A1|2020-06-18|Optics correction through machine learning
    EP3736709A1|2020-11-11|Classifier system and method for distributed generation of classification models
    DE102016211310B3|2017-07-20|DEVICE FOR MEASURING A BERRATION, PICTURE SYSTEMS AND METHOD FOR MEASURING A BERRATION
    DE102019214437A1|2021-03-25|Device for deflecting light for a device for virtual retinal display, method for producing the same and device for virtual retinal display
    同族专利:
    公开号 | 公开日
    US20190094532A1|2019-03-28|
    US11226481B2|2022-01-18|
    DE102017122636A1|2019-03-28|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    DE102020101763A1|2020-01-24|2021-07-29|Carl Zeiss Meditec Ag|MACHINE-LEARNING ASSISTED PIPELINE FOR SIZING AN INTRAOCULAR LENS|
    法律状态:
    优先权:
    申请号 | 申请日 | 专利标题
    DE102017122636.2A|DE102017122636A1|2017-09-28|2017-09-28|Methods and apparatus for designing optical systems|
    [返回顶部]