Acta Optica Sinica, Volume. 45, Issue 15, 1510001(2025)

Adaptive Digital Camouflage Pattern Generation Technology Based on Deep Learning

Tiehua Zhang, Bing Han, Meng Lian*, and Tun Cao
Author Affiliations
  • School of Optoelectronic Engineering and Instrumentation Science, Dalian University of Technology, Dalian 116024, Liaoning , China
  • show less
    Figures & Tables(12)
    Flowchart of adaptive digital camouflage generation
    Data augmentation effect diagram. (a) Original image; (b) image with color transformation; (c) image with cropping and stitching; (d) image with occlusion transformation
    Target and mask. (a) Camouflage target; (b) target detection result; (c) camouflage target mask
    Impact of mask expansion coefficient on camouflage image quality evaluation metrics
    Impact of data augmentation on detection performance. (a) Collected image; (b) detection results with data augmentation; (c) detection results without data augmentation
    Object detection results. (a) Dry grass; (b) lush grass; (c) wooden texture; (d) urban ground
    Comparison of digital camouflage generated by different methods. (a) Target to be camouflaged; (b) watershed method; (c) greedy iterative method; (d) raster method; (e) Markov chain method; (f) pix2pix; (g) CycleGan; (h) proposed method
    Camouflage results in different environments. (a)(e) Dry grass and its camouflage result; (b)(f) lush grass and its camouflage result; (c)(g) wooden texture and its camouflage result; (d)(h) urban ground and its camouflage result
    Camouflage results in diverse environments. (a) Represent the captured images of different background environments; (b) the corresponding camouflage results
    Camouflage results for different target objects and backgrounds. (a) Different target objects and background environments; (b) corresponding camouflage results
    • Table 1. Statistics of parameter count and computational complexity

      View table

      Table 1. Statistics of parameter count and computational complexity

      ParameterYOLOv8LAMATotal
      Parameters /1063.4150.9854.39
      Computational complexity /(109 FLOPs)3.07198.12201.19
    • Table 2. Evaluation metrics generated by different methods (xMean±xStd)

      View table

      Table 2. Evaluation metrics generated by different methods (xMean±xStd)

      ParameterWatershed methodGreedy iterative methodRaster methodMarkov chain methodpix2pixCycleGanProposed method
      SSIM0.86510.86720.86760.86950.88700.87490.8994
      PSNR32.521632.529832.533332.527932.648532.542832.6894
      MSE36.385236.316636.286636.332435.337036.206935.0058
    Tools

    Get Citation

    Copy Citation Text

    Tiehua Zhang, Bing Han, Meng Lian, Tun Cao. Adaptive Digital Camouflage Pattern Generation Technology Based on Deep Learning[J]. Acta Optica Sinica, 2025, 45(15): 1510001

    Download Citation

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

    Category: Image Processing

    Received: Apr. 2, 2025

    Accepted: Apr. 27, 2025

    Published Online: Aug. 8, 2025

    The Author Email: Meng Lian (mlian@dlut.edu.cn)

    DOI:10.3788/AOS250832

    CSTR:32393.14.AOS250832

    Topics