• Laser & Optoelectronics Progress
  • Vol. 60, Issue 12, 1215001 (2023)
Yuanyuan Huang1, Wenbo Xiong1, Hongwei Zhang1,2,*, and Weiwei Zhang1
Author Affiliations
  • 1School of Electronic Information, Xi'an Polytechnic University, Xi'an 710048, Shaanxi, China
  • 2State Key Laboratory of Industrial Control Technology, Zhejiang University, Hangzhou 310027, Zhejiang, China
  • show less
    DOI: 10.3788/LOP220691 Cite this Article Set citation alerts
    Yuanyuan Huang, Wenbo Xiong, Hongwei Zhang, Weiwei Zhang. Yarn-Dyed Fabric Defect Detection Based on U-Shaped Swin Transformer Auto-Encoder[J]. Laser & Optoelectronics Progress, 2023, 60(12): 1215001 Copy Citation Text show less
    Structure of Swin Transformer model
    Fig. 1. Structure of Swin Transformer model
    Structure of USTAE model
    Fig. 2. Structure of USTAE model
    Training phase of USTAE model
    Fig. 3. Training phase of USTAE model
    Defect detection phase of USTAE model
    Fig. 4. Defect detection phase of USTAE model
    Images of yarn-dyed fabric samples. (a) Defect-free samples; (b) defective samples
    Fig. 5. Images of yarn-dyed fabric samples. (a) Defect-free samples; (b) defective samples
    Influence of different noise levels on defect detection results
    Fig. 6. Influence of different noise levels on defect detection results
    Comparison of defect detection results of four models
    Fig. 7. Comparison of defect detection results of four models
    DatasetSL1SL13SP3SP5SP24CL1
    Defect-free171176168166231170
    Defective9231619274
    Table 1. Distribution of yarn-dyed fabric samples
    IndexModelSL1SL13SP3SP5SP24CL1Mean value
    Rprecision /%DCAE33.1025.1043.2657.0344.4023.3437.70
    MSDCAE45.4015.9442.3647.9932.0021.3934.18
    UDCAE61.1710.4630.4842.4824.2729.9433.13
    USTAE67.7237.7067.3662.7558.6255.0358.20
    Rrecall /%DCAE73.1415.8048.3664.0360.6990.7458.79
    MSDCAE62.6612.5749.2060.2945.6565.2349.27
    UDCAE74.227.7933.4653.5246.5356.0145.26
    USTAE66.4920.0353.1359.8259.4871.7155.11
    SF1 /%DCAE41.5017.1442.4655.8448.1536.7940.31
    MSDCAE51.6111.7741.6448.1934.4231.8436.58
    UDCAE64.535.3729.1041.5126.9036.5633.99
    USTAE62.1322.3952.1957.7954.6059.5751.44
    RIoU /%DCAE29.4212.7730.8339.8934.1823.0328.35
    MSDCAE39.428.5629.9433.3424.3919.5325.86
    UDCAE50.693.1421.4328.5518.1223.5924.30
    USTAE46.8316.5238.7342.1840.3643.3037.99
    Table 2. Comparison of four evaluation indexes of four models
    Skip connectionRprecision /%Rrecall /%SF1 /%RIoU /%
    052.4150.1646.8134.53
    155.4451.5049.3136.69
    253.9150.4848.0235.74
    358.2055.1151.4437.99
    Table 3. Ablation experiment of the number of skip connection
    Yuanyuan Huang, Wenbo Xiong, Hongwei Zhang, Weiwei Zhang. Yarn-Dyed Fabric Defect Detection Based on U-Shaped Swin Transformer Auto-Encoder[J]. Laser & Optoelectronics Progress, 2023, 60(12): 1215001
    Download Citation