Laser & Optoelectronics Progress, Volume. 60, Issue 14, 1415005(2023)

Anomaly Detection Method of Polarizer Appearance Based on Synthetic Defects

Xiaopin Zhong1, Junwei Zhu1, Zhihao Lie1, and Yuanlong Deng2、*
Author Affiliations
  • 1College of Mechatronics and Control Engineering, Shenzhen University, Shenzhen 510086, Guangdong, China
  • 2Shenzhen Institute of Technology, Shenzhen 518116, Guangdong, China
  • show less
    Figures & Tables(17)
    Schematic diagram of polarizer imaging experiment system
    Structured light imaging enhancement effect (left, uniform light; right, structured light)
    Overall framework of proposed model
    Anti anomaly detection network model based on encoding and decoding structure
    Anomaly detection process based on synthetic defects
    Real defect examples. (a) Point defect; (b) foreign matter; (c) bubbles; (d) crease
    Examples of composite defects
    Hyperparametric selection diagram of model loss weight
    Relationship between number of training samples and model accuracy
    Comparison of reconstruction results of some normal samples
    Comparison of reconstruction effect of defect samples
    Abnormal score graph of 100 normal samples and 100 defective samples
    Precision-recall curve of various methods
    Interference dataset images
    • Table 1. Normal samples under different characteristics

      View table

      Table 1. Normal samples under different characteristics

      Frequency(fringe spacing)Width ratio of black and white stripesBrightnessSaturationRotationEdge distortionNoise impact
      1/2π-1/π0.2-550-2000-0.90-π0.1-5Gaussian exponent
    • Table 2. Effect comparison of different methods

      View table

      Table 2. Effect comparison of different methods

      MethodAUCAverage time of single image detection /ms
      AnoGAN0.7187320
      VQ-VAE260.88325.1
      GANomaly0.79252.2
      Skip-GANomaly0.68637.4
      Skip-GANomaly(+proposed Llat0.73439.8
      Proposed method without proposed Llat0.91619.4
      Proposed method0.97919.2
    • Table 3. AUC difference between different methods and original data under interference data

      View table

      Table 3. AUC difference between different methods and original data under interference data

      MethodAUC of original test dataAUC of interference dataDecrease /%
      AnoGAN0.7180.62712.7
      VQ-VAE0.8830.74615.5
      GANomaly0.7920.65816.9
      Skip-GANomaly0.6860.4534.4
      Skip-GANomaly(+proposed Llat0.7340.64112.7
      Proposed method without proposed Llat0.9160.81810.7
      Proposed method0.9790.9334.7
    Tools

    Get Citation

    Copy Citation Text

    Xiaopin Zhong, Junwei Zhu, Zhihao Lie, Yuanlong Deng. Anomaly Detection Method of Polarizer Appearance Based on Synthetic Defects[J]. Laser & Optoelectronics Progress, 2023, 60(14): 1415005

    Download Citation

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

    Category: Machine Vision

    Received: Jul. 20, 2022

    Accepted: Sep. 26, 2022

    Published Online: Jul. 17, 2023

    The Author Email: Deng Yuanlong (dengyl@szu.edu.cn)

    DOI:10.3788/LOP222111

    Topics