Laser & Optoelectronics Progress, Volume. 61, Issue 10, 1012003(2024)

Multispectral Apple Surface Defect Detection Based on Improved YOLOv7-tiny

Chunjian Hua1,2、*, Mingchun Sun1,2, Yi Jiang1,2, Jianfeng Yu1,2, and Ying Chen3
Author Affiliations
  • 1School of Mechanical Engineering, Jiangnan University, Wuxi 214122, Jiangsu , China
  • 2Jiangsu Key Laboratory of Advanced Food Manufacturing Equipment & Technology, Wuxi 214122, Jiangsu , China
  • 3School of Internet of Things Engineering, Jiangnan University, Wuxi 214122, Jiangsu , China
  • show less
    Figures & Tables(12)
    Spectral image comparison. (a) RGB; (b) NIR (850 nm); (c) NIR (940 nm)
    Images of each category. (a) (b) Stem; (c) (d) calyx; (e) (f) defect; (g) (h) bruise; (i) (j) stab; (k) (l) rust
    Improved YOLOv7-tiny network structure
    CA module structure
    CoT module structure
    ELAN and Bi-ELAN structures. (a) ELAN; (b) Bi-ELAN
    Comparison of P-R curves of different models
    Detection results of the improved algorithm
    Comparison of detection results before and after model improvement. (a)‒(e) Original YOLOv7-tiny' detection results; (f)‒(j) improved YOLOv7-tiny' detection results
    • Table 1. Comparison of average accuracy of different models

      View table

      Table 1. Comparison of average accuracy of different models

      ModelAP /%mAP@0.5 /%
      stemcalyxdefectbruisestabrust
      YOLOv7-tiny (RGB)97.199.892.889.494.168.590.3
      YOLOv7-tiny (NIR)97.396.771.598.495.115.479.1
      YOLOv7-tiny (RGB-NIR)96.099.293.398.597.267.792.0
      Improved YOLOv7-tiny (RGB-NIR)98.199.794.598.497.471.193.2
    • Table 2. Results of ablation experiment

      View table

      Table 2. Results of ablation experiment

      Experiment No.CACoTBi-ELANFocal-EIoUmAP@0.5 /%Params /106FLOPs /109
      192.06.0313.28
      292.46.0413.32
      392.96.6113.74
      492.66.0313.28
      592.76.0313.28
      692.96.0413.32
      793.16.6113.74
      893.26.6213.78
    • Table 3. Performance comparison of different models

      View table

      Table 3. Performance comparison of different models

      ModelmAP@0.5 /%Params /106FLOPs /109Speed /(frame·s-1
      Faster-RCNN83.741.1591.356.0
      Swin-Transformer85.144.7792.045.8
      YOLOv389.561.55155.5861.3
      YOLOv5s90.57.0416.22104.2
      YOLOv7-tiny (baseline)92.06.0313.28108.7
      Improved YOLOv7-tiny93.26.6213.7889.3
    Tools

    Get Citation

    Copy Citation Text

    Chunjian Hua, Mingchun Sun, Yi Jiang, Jianfeng Yu, Ying Chen. Multispectral Apple Surface Defect Detection Based on Improved YOLOv7-tiny[J]. Laser & Optoelectronics Progress, 2024, 61(10): 1012003

    Download Citation

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

    Category: Instrumentation, Measurement and Metrology

    Received: Aug. 10, 2023

    Accepted: Oct. 9, 2023

    Published Online: Apr. 29, 2024

    The Author Email: Chunjian Hua (277795559@qq.com)

    DOI:10.3788/LOP231895

    CSTR:32186.14.LOP231895

    Topics