• Opto-Electronic Engineering
  • Vol. 51, Issue 4, 240025-1 (2024)
Qinglei Luan1,2, Xinyu Chang1,2, Ye Wu1,2, Conglong Deng1,2,*..., Yanqiong Shi1,2 and Zihua Chen1,2|Show fewer author(s)
Author Affiliations
  • 1School of Mechanical and Electrical Engineering, Anhui Jianzhu University, Hefei, Anhui 230601, China
  • 2Anhui Province Key Laboratory of Intelligent Manufacturing of Construction Machinery, Hefei, Anhui 230601, China
  • show less
    DOI: 10.12086/oee.2024.240025 Cite this Article
    Qinglei Luan, Xinyu Chang, Ye Wu, Conglong Deng, Yanqiong Shi, Zihua Chen. PAW-YOLOv7: algorithm for detection of tiny floating objects in river channels[J]. Opto-Electronic Engineering, 2024, 51(4): 240025-1 Copy Citation Text show less
    YOLOv7 network structure
    Fig. 1. YOLOv7 network structure
    PConv structure diagram
    Fig. 2. PConv structure diagram
    The process of calculating ODConv
    Fig. 3. The process of calculating ODConv
    ACmix structure diagram
    Fig. 4. ACmix structure diagram
    PAW-YOLOv7 network structure diagram
    Fig. 5. PAW-YOLOv7 network structure diagram
    Results of different data expansion methods
    Fig. 6. Results of different data expansion methods
    Object scale distribution of the dataset
    Fig. 7. Object scale distribution of the dataset
    Target detection results of different algorithms in different scenes. Left: detection image, Center: YOLOv7 model, Right: algorithm of this paper
    Fig. 8. Target detection results of different algorithms in different scenes. Left: detection image, Center: YOLOv7 model, Right: algorithm of this paper
    Comparison of detection accuracy of self-built datasets
    Fig. 9. Comparison of detection accuracy of self-built datasets
    Comparison results of heat maps with different algorithms
    Fig. 10. Comparison results of heat maps with different algorithms
    组别HeadACmixODCBSWIoUPC-ELANmAP/%FLOPs/GFPS
    179.9105.4101
    281.1119.586
    385.6101.675
    480.8109.897
    581.5105.4108
    678.183.7124
    787.3115.356
    888.2119.247
    990.8119.248
    1089.797.854
    Table 1. Results of ablation experiments
    算法mAP/%FPSFLOPs/GParams/M
    SSD73.37178.426.3
    Faster R-CNN76.86375.1137.1
    YOLOv385.712596.361.5
    YOLOv5s84.123664.710.7
    TPH-YOLOv582.369125.626.1
    YOLOv779.9101105.437.2
    YOLOv8l86.4117155.443.6
    PAW-YOLOv789.75497.825.4
    Table 2. Comparative experimental data of each algorithm on FloW-Img dataset
    算法mAP/%FPSFLOPs/GParams/M
    SSD57.96778.426.3
    Faster R-CNN61.36175.1137.1
    YOLOv359.713796.361.5
    YOLOv5s66.524264.310.7
    TPH-YOLOv563.774125.626.1
    YOLOv768.198105.137.2
    YOLOv8l65.9106155.443.6
    PAW-YOLOv771.86897.725.4
    Table 3. Comparative experimental data of each algorithm on self-constructed dataset
    Qinglei Luan, Xinyu Chang, Ye Wu, Conglong Deng, Yanqiong Shi, Zihua Chen. PAW-YOLOv7: algorithm for detection of tiny floating objects in river channels[J]. Opto-Electronic Engineering, 2024, 51(4): 240025-1
    Download Citation