• Laser & Optoelectronics Progress
  • Vol. 59, Issue 8, 0815009 (2022)
Yang Liu, Chunping Hou, Bangbang Ge, Zhipeng Wang*, and Cheng Peng
Author Affiliations
  • School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
  • show less
    DOI: 10.3788/LOP202259.0815009 Cite this Article Set citation alerts
    Yang Liu, Chunping Hou, Bangbang Ge, Zhipeng Wang, Cheng Peng. Image Anomaly Detection Algorithm Based on Discrete-Continuous Feature Coupling[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0815009 Copy Citation Text show less
    Flow chart of overall network
    Fig. 1. Flow chart of overall network
    Schematic diagram of similar block selection. (a) Block selection; (b) block reconstruction
    Fig. 2. Schematic diagram of similar block selection. (a) Block selection; (b) block reconstruction
    Variation diagram of network training accuracy
    Fig. 3. Variation diagram of network training accuracy
    Change diagram of overall training loss
    Fig. 4. Change diagram of overall training loss
    Detection results of proposed algorithm for some abnormal sample classes
    Fig. 5. Detection results of proposed algorithm for some abnormal sample classes
    Block selection results of proposedalgorithm on MVTec AD dataset. (a) Abnormal; (b) normal
    Fig. 6. Block selection results of proposedalgorithm on MVTec AD dataset. (a) Abnormal; (b) normal
    Relationship curves between parameter α and system performance on MVTec AD dataset
    Fig. 7. Relationship curves between parameter α and system performance on MVTec AD dataset
    Relationship curve between initial learning rate and network performance
    Fig. 8. Relationship curve between initial learning rate and network performance
    CategoryAVID23AE-L224AnoGAN25DSVDD26VAE-grad27GT28GANnomaly16ARNet29Proposed algorithm
    Mean73.071.055.072.077.067.076.283.986.9
    Bottle88.080.069.086.086.074.089.294.197.3

    Hazelnut

    Capsule

    Metal Nut

    Leather

    Pill

    Wood

    Carpet

    Tile

    Grid

    Cable

    Transistor

    Toothbrush

    Screw

    Zipper

    86.0

    85.0

    63.0

    58.0

    86.0

    83.0

    70.0

    66.0

    59.0

    64.0

    58.0

    73.0

    66.0

    84.0

    88.0

    62.0

    73.0

    44.0

    62.0

    74.0

    50.0

    77.0

    78.0

    56.0

    71.0

    98.0

    69.0

    80.0

    50.0

    58.0

    50.0

    52.0

    62.0

    68.0

    49.0

    51.0

    51.0

    53.0

    67.0

    57.0

    35.0

    59.0

    71.0

    69.0

    75.0

    73.0

    77.0

    87.0

    54.0

    81.0

    59.0

    71.0

    65.0

    70.0

    64.0

    74.0

    74.0

    86.0

    78.0

    71.0

    80.0

    89.0

    67.0

    81.0

    83.0

    56.0

    70.0

    89.0

    71.0

    67.0

    33.0

    67.0

    82.0

    82.0

    65.0

    48.0

    46.0

    54.0

    62.0

    85.0

    80.0

    94.0

    44.0

    87.0

    78.5

    73.2

    70.0

    84.2

    74.3

    83.4

    69.9

    79.4

    70.8

    75.7

    79.2

    65.3

    74.6

    74.5

    85.5

    68.1

    66.7

    86.2

    78.6

    92.3

    70.6

    73.5

    88.3

    83.2

    84.3

    100

    100

    87.6

    92.9

    74.3

    68.9

    67.3

    74.7

    99.2

    89.3

    96.4

    91.3

    86.3

    87.9

    97.5

    100

    80.5

    Table 1. AUC of different algorithms in each category
    Yang Liu, Chunping Hou, Bangbang Ge, Zhipeng Wang, Cheng Peng. Image Anomaly Detection Algorithm Based on Discrete-Continuous Feature Coupling[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0815009
    Download Citation