• Opto-Electronic Engineering
  • Vol. 51, Issue 10, 240196 (2024)
Yan Wang, Honghui Wang, Shudong Liu*, Yan Zhang, and Zeyu Hao
Author Affiliations
  • School of Computer and Information Engineering,Tianjin Chengjian University,Tianjin 300000,China
  • show less
    DOI: 10.12086/oee.2024.240196 Cite this Article
    Yan Wang, Honghui Wang, Shudong Liu, Yan Zhang, Zeyu Hao. Small target detection in sonar images with multilevel feature screening and task dynamic alignment[J]. Opto-Electronic Engineering, 2024, 51(10): 240196 Copy Citation Text show less
    YOLOv8s network model
    Fig. 1. YOLOv8s network model
    EMS-FPN module
    Fig. 2. EMS-FPN module
    ELA attention mechanisms
    Fig. 3. ELA attention mechanisms
    ELA-SFF module
    Fig. 4. ELA-SFF module
    TDADH module
    Fig. 5. TDADH module
    Task decomposition module
    Fig. 6. Task decomposition module
    Improved YOLOv8s target detection network model
    Fig. 7. Improved YOLOv8s target detection network model
    SCTD sonar image data enhancement. (a) Original image; (b) Random removal of pixel points; (c) Sharpening process; (d) Adjustment of brightness; (e) Adjustment of hue; (f) Image flipping
    Fig. 8. SCTD sonar image data enhancement. (a) Original image; (b) Random removal of pixel points; (c) Sharpening process; (d) Adjustment of brightness; (e) Adjustment of hue; (f) Image flipping
    Comparison of detection results of different methods for sonar images in URPC 2021. (a) RetinaNet[28" target="_self" style="display: inline;">28]; (b) PAA[29" target="_self" style="display: inline;">29]; (c) CenterNet[30" target="_self" style="display: inline;">30];(d) SparseRCNN[31" target="_self" style="display: inline;">31]; (e) YOLOF[32" target="_self" style="display: inline;">32]; (f) TOOD[33" target="_self" style="display: inline;">33]; (g) VarifocalNet[34" target="_self" style="display: inline;">34];(h) DSA-Net[35" target="_self" style="display: inline;">35]
    Fig. 9. Comparison of detection results of different methods for sonar images in URPC 2021. (a) RetinaNet[28]; (b) PAA[29]; (c) CenterNet[30];(d) SparseRCNN[31]; (e) YOLOF[32]; (f) TOOD[33]; (g) VarifocalNet[34];(h) DSA-Net[35]
    Comparison of detection results of different methods for sonar images in URPC 2021. (a) YOLOv5; (b) YOLOv6; (c) YOLOv7; (d) YOLOv8s; (e) This paper
    Fig. 10. Comparison of detection results of different methods for sonar images in URPC 2021. (a) YOLOv5; (b) YOLOv6; (c) YOLOv7; (d) YOLOv8s; (e) This paper
    AlgorithmHSFPNEMS-FPNTDADHPrecision/%Recall/%Params/MmAP0.5/%
    YOLOv8s×××97.396.411.197.9
    YOLOv8s××97.496.17.197.8
    YOLOv8s××96.996.67.398.1
    YOLOv8s××96.996.28.897.8
    YOLOv8s×97.496.88.698.2
    Table 1. Results of ablation experiments on URPC2021 dataset
    AlgorithmHSFPNEMS-FPNTDADHParams/MPrecision/%Recall/%Ball/%Cylinder/%Tyre/%
    YOLOv8s×××11.197.396.498.696.896.7
    YOLOv8s××7.197.496.198.896.897
    YOLOv8s××7.396.996.698.997.497.5
    YOLOv8s××8.896.996.298.396.696.5
    YOLOv8s×8.697.496.899.398.297.5
    Table 2. Results of small target category ablation experiments in the URPC2021 dataset
    AlgorithmCube/%Ball/%Cylinder/%Human body/%Trye/%Circle cage/%Square cage/%Metal bucket/%mAP0.5/%
    RetinaNet[28]92.694.083.086.572.783.392.058.582.8
    PAA[29]95.393.688.294.173.791.595.181.489.1
    CenterNet[30]91.996.884.296.489.585.297.788.191.2
    SparseRCNN[31]97.496.889.097.689.293.198.588.093.7
    YOLOF[32]95.591.987.192.066.772.988.862.082.1
    TOOD[33]96.495.188.693.476.991.296.985.890.5
    VarifocalNet[34]96.494.690.195.079.591.596.785.491.2
    DSA-Net[35]97.998.094.298.494.993.299.294.096.2
    YOLOv597.998.395.998.897.596.099.495.597.4
    YOLOv698.098.796.299.496.996.199.397.497.7
    YOLOv796.297.592.994.991.391.698.598.095.1
    YOLOv8s97.798.696.899.596.797.599.297.297.9
    本文模型97.799.398.299.597.597.398.797.898.2
    Table 3. Comparison of accuracy and mAP@50 for different underwater objects on URPC2021 dataset with other methods
    AlgorithmHSFPNEMS-FPNTDADHParams/MPrecision/%Recall/%Ship(small)/%mAP0.5/%
    YOLOv8s×××11.194.092.493.496.3
    YOLOv8s××7.195.291.595.497.0
    YOLOv8s××7.396.192.195.697.0
    YOLOv8s××8.897.494.197.397.4
    YOLOv8s×8.697.194.497.898.1
    Table 4. Results of comparison experiments on SCTD dataset
    Yan Wang, Honghui Wang, Shudong Liu, Yan Zhang, Zeyu Hao. Small target detection in sonar images with multilevel feature screening and task dynamic alignment[J]. Opto-Electronic Engineering, 2024, 51(10): 240196
    Download Citation