Infrared and Laser Engineering, Volume. 54, Issue 3, 20240487(2025)

Research on infrared ship target detection algorithm based on local-global self-attention and spatial-channel sparse enhancement

Yupei LI and Zhonghua WANG
Author Affiliations
  • School of Information Engineering, Nanchang Hangkong University, Nanchang 330063, China
  • show less
    Figures & Tables(16)
    Structure of YOLOv8s
    Improved structure of YOLOv8s
    LGSA structure
    Multi-Head Self-Attention workflow
    Structural diagram of backbone network improvements
    SCSA structure
    Structural diagram of neck network improvement
    Comparison of improved Soft-NMS detection
    Distribution of the number of vessels by type
    (a) Comparison of training processes for each model's mAP0.5; (b) Comparison of training processes for each model's mAP0.5∶0.95
    (a) YOLOv8s baseline model; (b) Improved YOLOv8s model
    (a) Original figure; (b) Improved YOLOv8s heat map; (c) YOLOv5s heat map; (d) YOLOv7 heat map; (e) YOLOv9 heat map; (f) YOLOv10 s heat map; (g) YOLOv11 heat map
    • Table 1. Experimental environment

      View table
      View in Article

      Table 1. Experimental environment

      NameModel
      CPU12th Gen Intel(R) Core(TM)i5-12400
      GPUNVIDIA GeForce RTX 2060/12 G
      Memory16 G
      Computer modelPRIME B660M-K D4
    • Table 2. Training parameter settings

      View table
      View in Article

      Table 2. Training parameter settings

      IndexHyperparametersSet value
      1Lr (Learning rate)0.01
      2Momentum0.937
      3Weight_Decay0.0005
      4Epochs100
      5Batch16
      6Img_size640
      7Workers2
      8IoU0.7
    • Table 3. Ablation experiments

      View table
      View in Article

      Table 3. Ablation experiments

      IndexModelImprovement pointsmAP0.5mAP0.5∶0.95FLOPs/GPara/M
      LGSASCSASoft-NMS
      1YOLOv8s×××93.6%68.4%28.511.1
      2YOLOv8s××93.9%68.9%36.111.5
      3YOLOv8s×95.7%69.0%35.410.8
      4YOLOv8s95.7%72.8%35.410.8
    • Table 4. Comparative experiments

      View table
      View in Article

      Table 4. Comparative experiments

      IndexModelEpochsmAP0.5mAP0.5∶0.95FLOPs/GPara/M
      1Faster RCNN10070.9%39.6%948.228.3
      2YOLOv5s10093.7%64.9%15.87.0
      3YOLOv710094.4%65.6%105.237.2
      4YOLOXs10083.8%56.7%26.88.9
      5RetinaNet10069.3%47.0%170.138.0
      6YOLOv910093.5%67.9%265.060.8
      7YOLOv10s10092.8%66.0%24.58.0
      8YOLOv1110091.1%64.7%6.32.6
      9Improved YOLOv8s10095.7%72.8%35.410.8
    Tools

    Get Citation

    Copy Citation Text

    Yupei LI, Zhonghua WANG. Research on infrared ship target detection algorithm based on local-global self-attention and spatial-channel sparse enhancement[J]. Infrared and Laser Engineering, 2025, 54(3): 20240487

    Download Citation

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

    Category: Optical imaging, display and information processing

    Received: Oct. 28, 2024

    Accepted: --

    Published Online: Apr. 8, 2025

    The Author Email:

    DOI:10.3788/IRLA20240487

    Topics