KR-20260067586-A - DEEP LEARNING BASED LENS SHAPE REVERSE ENGINEERING METHOD

Abstract

The present invention relates to a deep learning-based lens shape reverse engineering method, comprising the steps of acquiring lens shape data by at least one processor of a computer system and reverse engineering the lens shape based on deep learning. The step of acquiring lens shape data includes the step of designing the lens shape through an optical design program, the step of parameterizing the designed lens shape, and the step of storing ray tracing results according to changes in lens shape parameters. The step of reverse engineering the lens shape based on deep learning includes the step of designing a deep learning structure, the step of learning the deep learning structure, and the step of verifying the lens shape reverse engineering results according to the deep learning structure. Accordingly, lens shape reverse engineering through deep learning is possible, thereby effectively reducing the consumption of human and material resources compared to the existing lens shape design method through trial and error.

Inventors

• 김철순
• 강민형
• 이규홍

Assignees

• 한국전자기술연구원

Dates

Publication Date

20260513

Application Date

20241106

Claims (14)

1. In a deep learning-based lens shape inverse engineering method, The method includes the step of acquiring lens shape data by at least one processor of a computer system and the step of reverse engineering the lens shape based on deep learning, The step of acquiring the above lens shape data is, Step of designing the lens shape using an optical design program; Step of parameterizing the designed lens shape; and It includes a step of storing ray tracing results according to changes in lens shape parameters, and The step of reverse engineering the lens shape based on the deep learning mentioned above is Step of designing the deep learning structure; A step of learning the above deep learning structure; and A deep learning-based lens shape reverse engineering method comprising a step of verifying the lens shape reverse engineering result according to the deep learning structure.
2. In claim 1, the step of designing the lens shape is, A step of designing an initial lens shape by combining a cone with height (z) and base radius (x) as design parameters and a hemisphere with cross-sectional radius (r) and height (y) as design parameters; A step of designing an intermediate lens shape by cutting the top surface and the left and right surfaces of a shape formed by cutting vertically from the center of the initial lens shape above horizontally and vertically, respectively; and A deep learning-based lens shape reverse engineering method characterized by including the step of designing a final lens shape by creating a rectangular space where a light-emitting part is to be located based on a horizontally cut plane of the intermediate lens shape and setting the vertical position (d) of the light-emitting part as another parameter.
3. In Article 1, A deep learning-based lens shape reverse engineering method characterized by the optical design program being the Part Designer tool of Zemax Opticstudio.
4. In Article 2, A deep learning-based lens shape inverse engineering method characterized by the step of storing ray tracing results according to changes in the lens shape parameters including acquiring ray tracing result data according to the lens shape design parameter values while changing the lens shape design parameter values x, y, z, r, and d.
5. In Article 1, A deep learning-based lens shape reverse engineering method characterized by acquiring ray tracing result data according to changes in the lens shape parameters in large quantities using the part designer tool of Zemax Opticstudio so that it can be utilized for deep learning.
6. In Article 1, A deep learning-based lens shape inverse engineering method characterized in that the above deep learning structure is a structure that predicts lens shape design parameters from light intensity according to the directional angle.
7. A deep learning-based lens shape inverse engineering method according to claim 1, characterized in that the deep learning structure is a network connecting a structure that predicts lens shape design parameters from light intensity according to
8. In Article 1, A deep learning-based lens shape inverse engineering method characterized in that the above deep learning structure is a structure that predicts lens shape design parameters from physical changes according to the lens shape.
9. A deep learning-based lens shape inverse engineering method according to claim 1, characterized in that the deep learning structure is a network connecting a structure that predicts lens shape design parameters from lens shape design parameters to a structure that predicts physical changes according to lens shape.
10. In Article 1, A deep learning-based lens shape inverse engineering method characterized in that the deep learning model according to the above deep learning structure is constructed using a Deformable Convolutional Network (DCN).
11. In Article 1, A deep learning-based lens shape inverse engineering method characterized in that the deep learning model according to the above deep learning structure is selected from a group including LSTM (Long Short Term Memory), MLP (MultiLayer Perceptron), and Transformer models.
12. In Article 1, A deep learning-based lens shape reverse engineering method characterized by training, validating, and testing a deep learning model according to the deep learning structure by dividing the entire data in an 8:1:1 ratio.
13. In Article 1, A deep learning-based lens shape reverse engineering method characterized by using Adam as the optimizer, a learning rate of 0.0001, a batch size of 32, and an epoch of 100 in a deep learning model according to the deep learning structure above.
14. A computer-readable recording medium storing a computer program for executing a method according to any one of claims 1 to 13 on a computer system.

Description

Deep Learning Based Lens Shape Reverse Engineering Method The present invention relates to a method for reverse engineering a lens shape based on deep learning. In this specification, the term "lens" refers to a lens that serves to direct light emitted from a light-emitting part, such as an LED, in a specific direction or to collect it in a specific direction, and it is important to design the shape of the lens to suit this role. For example, large electronic display boards are generally installed on tall buildings, and in order to convey visual information to people on the ground, the lens must be designed so that the light emitted from the display board is directed toward the ground. In this regard, conventional lens shape design methods proceed in the order of designing the lens shape using optical simulation software, obtaining simulation results using ray tracing to predict performance, and designing and manufacturing the actual lens if it matches the target set based on this. Another conventional lens shape design method proceeds by designing and manufacturing the lens shape directly based on the designer's experience without using optical simulation tools, and verifying its performance. However, the above conventional lens shape design methods were trial-and-error methods that required a lot of time and money because they involved numerous trials and errors. FIG. 1 is a flowchart schematically showing each step of a deep learning-based lens shape reverse engineering method according to one embodiment of the present invention. Figure 2 is a three-dimensional view illustrating the lens shape design step of the lens shape data acquisition process in the deep learning-based lens shape reverse engineering method of Figure 1. Figure 3 is an example diagram showing the ray tracing result of the lens shape data acquisition process in the deep learning-based lens shape inverse engineering method of Figure 1. FIGS. 4a to 4c are schematic diagrams illustrating deep learning models used in a deep learning-based lens shape reverse engineering method according to one embodiment of the present invention. Figure 5 is an example diagram showing the result of restoring directional angle data according to lens design parameters in the deep learning model of Figure 4a. Figure 6 is an example diagram showing the results of restoring lens design parameters according to the directional angle data in the deep learning model of Figure 4b. Figure 7 is an example diagram showing the lens design parameter prediction results based on the directional angle data in the deep learning model of Figure 4c and the directional angle data restoration results accordingly. Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. However, the present invention is not limited by the following embodiments. In addition, the same reference numerals are used for identical components in the drawings, and redundant descriptions thereof are omitted. FIG. 1 is a flowchart schematically illustrating the steps of a deep learning-based lens shape reverse engineering method according to an embodiment of the present invention. Referring to FIG. 1, the deep learning-based lens shape reverse engineering method of the present embodiment consists of a lens shape data acquisition process (10) and a deep learning-based lens shape reverse engineering process (11) performed by at least one processor of a computer system. In addition, the lens shape data acquisition process (10) is further composed of a lens shape design step (101), a lens shape parameterization step (102), and a ray tracing result storage step (103) based on changes in lens shape parameters. In this regard, FIG. 2 is a three-dimensional view for explaining the lens shape design step (101) through the optical design program of the process (10) of acquiring lens shape data in the deep learning-based lens shape reverse engineering method of FIG. 1. As shown in FIG. 2, an initial lens shape (201) is designed by combining a cone with a height (z) and a base radius (x) as design parameters and a hemisphere with a cross-sectional radius (r) and a height (y) as design parameters. At this time, the radius (x) of the base of the cone and the radius (r) of the cross-section of the hemisphere are the same. Next, as shown in FIG. 2, the upper and left and right sides of the shape formed by cutting vertically from the center of the initial lens shape (201) are cut horizontally and vertically, respectively, to create an intermediate lens shape (202). After that, as shown in FIG. 2, a rectangular space for the light-emitting part, such as an LED, to be located is created based on a horizontally cut plane in the final lens shape (203), and the vertical position (d) of the light-emitting part is set as another parameter to complete the lens shape design. According to one embodiment of the present invention, such a lens shape can be designed using Part