Photonics Research, Volume. 11, Issue 8, 1408(2023)

FreeformNet: fast and automatic generation of multiple-solution freeform imaging systems enabled by deep learning

Boyu Mao1, Tong Yang1,2、*, Huiming Xu1, Wenchen Chen1, Dewen Cheng1,3, and Yongtian Wang1,2
Author Affiliations
  • 1Beijing Engineering Research Center of Mixed Reality and Advanced Display, School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, China
  • 2Beijing Key Laboratory of Advanced Optical Remote Sensing Technology, School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, China
  • 3e-mail: cdwlxk@bit.edu.cn
  • show less
    Figures & Tables(13)
    Whole optical design framework based on deep learning.
    Illustration of the fundamental data set generation and feedback strategy.
    Illustration of input and output parameters of DNN. The superscripts i, o, and tar denote input, output, and target, respectively. The output and target surface and structure parameters values are used to construct the mean square error (MSE) and then construct the supervised loss function Lsuper.
    Sketch of the parameter space and subspace pair SPΦ(i) and SPψ(i). A 2D space (only two system parameters and two structure parameters are considered, respectively) is plotted here for clarity, but actual parameter spaces should be high-dimensional spaces. Four subspaces are plotted here as an example. The subspaces plotted in same color form a subspace pair SPΦ(i)-SPψ(i). φ and ψ of reference system RSYS(i) are at the center of SPΦ(i) and SPψ(i).
    Illustration of the training mode of combined supervised and unsupervised training.
    Fast generation process of multiple-solution freeform imaging systems using FreeformNet.
    (a) The selected folding geometry of freeform off-axis three-mirror imaging system and its structure constraints. (b) Sketch of the concept of structure parameters range.
    Typical predicted systems when system parameters and structure parameters were all provided.
    Average RMS spot diameter and maximum relative distortion of normal systems in case one (system numbers were arranged in ascending order according to the average RMS spot diameter).
    Typical predicted systems when all system parameters and partial structure parameters were provided.
    Average RMS spot diameter, maximum relative distortion, and system volume of normal systems in the second case (the system numbers were arranged in ascending order according to the average RMS spot diameter).
    Typical predicted systems when partial system parameters were provided.
    • Table 1. Comparison of Time Cost Using Different Design Methods

      View table
      View in Article

      Table 1. Comparison of Time Cost Using Different Design Methods

      MethodTime for System Generation with Narrow or Moderate FOVTime for System Generation with Wide FOV
      Starting PointWith Good Imaging Performance
      [19]Several minutes for system: FOV=3°×3°, EFL=60  mm, F#=2, LWIRNot reportedNot applicable or not reported
      [20]Several minutes for system: FOV=8°×8°, EFL=95  mm, F#=1.8, LWIR2.44 h for system: FOV=8°×8°, EFL=95  mm, F#=1.8, LWIR
      [23]Not reported. No individual starting point design processSeveral minutes for system: FOV=4°×4°, EFL=600  mm, F#=3,VIS
      [25]Not reported. No individual starting point design process5.9 min for system: FOV=3°×3°, EFL=60  mm, F#=1.5, LWIR
      [22]Not reported. No individual starting point design processAbout 30 min for system: FOV=8°×6°, EFL=50  mm, F#=1.8, LWIR
      [17]Not reported. A step-by-step design method based on nodal aberration theory: FOV=4°×4°, EFL=600  mm, F#=3, VISNot reported
      Finding systems from literaturesMaybe tens of minutes or several hours. A large probability that a feasible starting point cannot be found
      [2630]Not applicable for freeform system design or not applicable for multiple-solution design
      Our methodLess than 0.003 s per system for starting point generation for the above systems or systems with much wider FOV. If 1000 systems are simultaneously predicted, 1.5×105  s per system. If further optimization is conducted, it takes only about several seconds in general to obtain good imaging performance for a system with narrow or moderate FOV (similar with the cases shown in this table). For example, optimizing 20 different systems can be done in between 32 and 37 s, less than 2 s for each system
    Tools

    Get Citation

    Copy Citation Text

    Boyu Mao, Tong Yang, Huiming Xu, Wenchen Chen, Dewen Cheng, Yongtian Wang. FreeformNet: fast and automatic generation of multiple-solution freeform imaging systems enabled by deep learning[J]. Photonics Research, 2023, 11(8): 1408

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Imaging Systems, Microscopy, and Displays

    Received: Apr. 12, 2023

    Accepted: Jun. 7, 2023

    Published Online: Jul. 31, 2023

    The Author Email: Tong Yang (yangtong@bit.edu.cn)

    DOI:10.1364/PRJ.492938

    Topics