• Opto-Electronic Engineering
  • Vol. 51, Issue 1, 230284-1 (2024)
Dongdong Zhao1, Dunhan Xie1, Peng Chen1、*, Ronghua Liang1, Yi Shen1, and Xinxin Guo2
Author Affiliations
  • 1College of Computer Science and Technology, Zhejiang University of Technology, Hangzhou, Zhejiang 310023, China
  • 2Institute of Deep-sea Science and Engineering, Chinese Academy of Sciences, Sanya, Hainan 572000, China
  • show less
    DOI: 10.12086/oee.2024.230284 Cite this Article
    Dongdong Zhao, Dunhan Xie, Peng Chen, Ronghua Liang, Yi Shen, Xinxin Guo. Lightweight YOLOv5 sonar image object detection algorithm and implementation based on ZYNQ[J]. Opto-Electronic Engineering, 2024, 51(1): 230284-1 Copy Citation Text show less
    Lightweight sonar image YOLOv5 object detection network model
    Fig. 1. Lightweight sonar image YOLOv5 object detection network model
    Depthwise separable convolution structure
    Fig. 2. Depthwise separable convolution structure
    GSConv structure diagram
    Fig. 3. GSConv structure diagram
    CBAM module structure diagram
    Fig. 4. CBAM module structure diagram
    SimAM structure diagram
    Fig. 5. SimAM structure diagram
    Diagram of the sonar system
    Fig. 6. Diagram of the sonar system
    Sonar system workflow diagram
    Fig. 7. Sonar system workflow diagram
    Model conversion process
    Fig. 8. Model conversion process
    Distribution of the number of images in the dataset
    Fig. 9. Distribution of the number of images in the dataset
    Data-enhanced images
    Fig. 10. Data-enhanced images
    Different algorithm image detection result comparison
    Fig. 11. Different algorithm image detection result comparison
    EquipmentFLOPs/GMemory footprint/MBFrame rate/FPSPower/WFrame rate/power/(FPS/W)
    GPU6.0140136.42500.146
    CPU6.021038.8650.135
    ZYNQ70205.8900.63.70.162
    Table 1. Running results of different equipments
    AlgorithmMap50Params(M)
    YOLOv5s0.9127.2
    EfficientDet0.8343.9
    YOLOv70.89037
    Faster-RCNN0.89625.6
    SSD0.91251
    This paper0.9333.2
    Table 2. Comparative experiments
    AlgorithmPrecisionRecallMap50Map50-95Params(M)
    YOLOv5s0.8960.9050.9120.6857.2
    YOLOv5s+SimAM0.8920.9110.9220.7007.0
    YOLOv5s+CBAM0.8960.9270.9180.6977.1
    YOLOv5s+SimAM+CBAM0.8920.9120.9250.7017.1
    YOLOv5s+Focal-CIOU0.8990.8910.9300.7007.2
    YOLOv5s+DWConv0.8800.9110.9190.6943.6
    YOLOv5s+GSConv0.8970.9060.9230.6996.2
    This paper0.9070.8940.9330.7123.2
    Table 3. Ablation experiments
    AlgorithmPrecisionRecallMap50Map50-95
    YOLOv5s0.6780.530.5830.353
    This paper0.6610.5190.590.339
    Table 4. Comparison experiment of classic datasets
    Dongdong Zhao, Dunhan Xie, Peng Chen, Ronghua Liang, Yi Shen, Xinxin Guo. Lightweight YOLOv5 sonar image object detection algorithm and implementation based on ZYNQ[J]. Opto-Electronic Engineering, 2024, 51(1): 230284-1
    Download Citation