Optics and Precision Engineering, Volume. 33, Issue 2, 311(2025)

MFL_YOLOv8 algorithm for surface defect detection of microfiber leather

Xiaodong SUN1, Qibing ZHU1、*, Huawei XU2, Tongzhen XING3, and Haibin ZHU3
Author Affiliations
  • 1School of Internet of Things Engineering, Jiangnan University, Wuxi2422, China
  • 2Hexin Kuraray Micro Fiber Leather (Jiaxing) Co. , Ltd, Jiaxing314003, China
  • 3Zhejiang Maimu Intelligent Technology Co. , Ltd, Jiaxing14000, China
  • show less
    Figures & Tables(16)
    Diagram of imaging equipment
    Surface defects of microfiber Leather
    Statistical chart of sample number of defects and their size distribution
    Structure of MFL_YOLOv8
    Structure of DLKA
    Structure of Dysample
    Calculation results of different losses in two cases
    Detection results of different models
    Heatmap visualization
    Feature map visualization
    On-site detection results in industry
    • Table 1. Defect detection results of different algorithms

      View table
      View in Article

      Table 1. Defect detection results of different algorithms

      ModelsPrecision/%Recall/%Params/MFLOPs/GFPS/(frame·s-1
      SSD72.3168.4624.362.343.2
      YOLOv4-tiny84.2782.986.016.5103.2
      YOLOv5n79.2177.831.84.2179.3
      YOLOv5s82.4783.157.018.587.2
      YOLOv7-tiny84.1683.966.113.182.5
      YOLOv8n87.0985.133.08.1149.3
      YOLOv8s87.7585.2211.128.578.2
      YOLOv9n81.3578.314.18.1120.8
      MFL_YOLOv892.4792.403.69.4135.2
    • Table 2. Ablation experiments results

      View table
      View in Article

      Table 2. Ablation experiments results

      DCNv3-LKAP2DysampleMPDIOUPrecision/%Recall/%Params/MFLOPs/GFPS/(frame·s-1
      87.0985.133.08.1149.3
      89.1888.323.58.5140.5
      88.5287.253.19.0135.6
      88.1386.693.08.1151.2
      87.9186.833.08.1149.6
      90.2589.913.69.4132.1
      91.6291.123.69.4135.1
      92.4792.403.69.4135.2
    • Table 3. Comparison of confusion matrices

      View table
      View in Article

      Table 3. Comparison of confusion matrices

      True

      Pred

      Background

      Stains

      (368)

      Debris

      (247)

      Spots

      (165)

      Wrinkles

      (169)

      Seams

      (84)

      Precision/%
      Background-22(-15)41(-44)6(-2)5(-8)3(-2)-
      Stains29(-28)343(+22)7(-2)0(-3)1090.26(+8.16)
      Debris18(-5)1(-5)198(+47)00091.24(+7.35)
      Spots9(-6)2(-2)1(-1)158(+5)00(-1)92.94(+5.51)
      Wrinkles13(-8)000163(+8)092.61(+4.54)
      Seams3(-1)001081(+3)95.29(+1.31)
      Recall/%-93.21(+5.98)80.16(+19.03)95.76(+3.03)96.45(+4.73)96.43(+3.57)-
    • Table 4. Experimental results of DLKA and DCNv3-LKA

      View table
      View in Article

      Table 4. Experimental results of DLKA and DCNv3-LKA

      ModelsPrecision/%Recall/%Params/MFLOPs/GFPS/(frame·s-1
      Baseline87.0985.133.08.1149.3
      DLKA87.8186.284.79.5108.3
      DCNv3-LKA89.1888.323.58.4140.5
    • Table 5. Fig.5 Experimental results of different upsampling modules

      View table
      View in Article

      Table 5. Fig.5 Experimental results of different upsampling modules

      ModelsPrecision/%Recall/%Params/MFLOPs/GFPS/(frame·s-1
      Nearest87.0985.133.0078.1149.3
      Bilinear86.2884.95+0.0058.2140.5
      CARAFE87.7685.33+0.0768.1138.2
      Dysample88.1386.69+0.0128.1151.2
    Tools

    Get Citation

    Copy Citation Text

    Xiaodong SUN, Qibing ZHU, Huawei XU, Tongzhen XING, Haibin ZHU. MFL_YOLOv8 algorithm for surface defect detection of microfiber leather[J]. Optics and Precision Engineering, 2025, 33(2): 311

    Download Citation

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

    Category:

    Received: Sep. 19, 2024

    Accepted: --

    Published Online: Apr. 30, 2025

    The Author Email: Qibing ZHU (zhuqib@163.com)

    DOI:10.37188/OPE.20253302.0311

    Topics