• Laser & Optoelectronics Progress
  • Vol. 62, Issue 6, 0615007 (2025)
Guangle Wang*, Yatong Zhou, and Zhao Wang
Author Affiliations
  • School of Electronic and Information Engineering, Hebei University of Technology, Tianjin 300401, China
  • show less
    DOI: 10.3788/LOP241782 Cite this Article Set citation alerts
    Guangle Wang, Yatong Zhou, Zhao Wang. Lightweight Model for Irregular Wear Detection in Power Operations[J]. Laser & Optoelectronics Progress, 2025, 62(6): 0615007 Copy Citation Text show less
    YOLO-WWS structure diagram
    Fig. 1. YOLO-WWS structure diagram
    YOLOv8n detection head
    Fig. 2. YOLOv8n detection head
    Structure of SCTADH
    Fig. 3. Structure of SCTADH
    bottleneck and DStar structures. (a) bottleneck structure;(b) DStar structure
    Fig. 4. bottleneck and DStar structures. (a) bottleneck structure;(b) DStar structure
    MLFCA attention module
    Fig. 5. MLFCA attention module
    Model parameter quantity distribution before and after improvement. (a) YOLOv8n parameter quantity distribution; (b) YOLO-WWSP parameter quantity distribution
    Fig. 6. Model parameter quantity distribution before and after improvement. (a) YOLOv8n parameter quantity distribution; (b) YOLO-WWSP parameter quantity distribution
    Comparison of YOLOv8n (left) and YOLO-WWSP (right) test results
    Fig. 7. Comparison of YOLOv8n (left) and YOLO-WWSP (right) test results
    Training parameterValue
    Learning rate0.01
    Batch size32
    Epochs200
    Image size640×640
    Momentum0.937
    Weight decay0.0005
    Table 1. Experimental model parameters
    ModelSCTADHDStarMLFCALAMPmAP@0.5 /%mAP@0.5∶0.95 /%ParamsFLOPs /109Model size /MB
    191.174.331519048.75.95
    291.376.122452358.74.48
    391.575.328491527.85.38
    491.075.921246598.14.26
    592.576.221259958.14.28
    691.875.89534025.32.05
    Table 2. Results of the ablation experiments
    ModelmAP@0.5 /%mAP@0.5∶0.95 /%
    491.075.9
    +CBAM92.475.9
    +ELA92.176.1
    +EMA92.275.3
    +LSKA91.775.9
    +MLFCA92.576.2
    Table 3. Comparison of the five attention mechanisms
    ModelmAP@0.5 /%mAP@0.5∶0.95 /%ParamsFLOPs /109Model size /MB
    RTDETR-L87.072.328607660101.056.33
    YOLOv3n88.970.31216878419.023.25
    YOLOv5n89.773.426492007.75.02
    YOLOv6n91.475.2449539213.08.28
    YOLOv8n91.174.331519048.75.95
    YOLOv10n91.172.922992646.75.50
    NanoDet-m88.467.99108001.33.78
    NanoDet-m-1.5×89.069.420106402.78.05
    NanoDet-plus-m-1.5×89.770.324044963.429.99
    NanoDet-g84.662.437531844.914.99
    PicoDet-xs83.957.26895621.32.67
    PicoDet-s84.760.511761841.94.53
    PicoDet-m87.263.534612165.113.25
    YOLO-WWSP91.875.89534025.32.05
    Table 4. Comparison of the experimental results