Opto-Electronic Engineering, Volume. 51, Issue 11, 240212-1(2024)

DES-YOLO: a more accurate object detection method

Huawei Zheng... Fei Wang* and Jianbang Gao |Show fewer author(s)
Author Affiliations
  • School of Electronic Engineering, Xi’an Shiyou University, Xi’an, Shaanxi 710065, China
  • show less
    Figures & Tables(17)
    YOLOv5 network structure
    Deformable attention module
    Improved network structure
    Images of part of the NWPU VHR-10 dataset
    Images of part of the fabric defect dataset
    Comparison of detection accuracy of different network models
    Comparison of detection effect of different models
    • Table 1. Experimental environment and configuration

      View table
      View in Article

      Table 1. Experimental environment and configuration

      TypeConfiguration
      GPUNVIDIA GeFore RTX4090
      CPU13th Gen Intel(R) Core(TM) i7-13620H
      CUDA11.7
      Deep learning frameworkPytorch
      Python3.12
    • Table 2. Remote sensing target detection data set

      View table
      View in Article

      Table 2. Remote sensing target detection data set

      PrecisionRecallmAP@0.5mAP@0.5:0.95
      YOLOv5s0.9370.8990.9290.562
      YOLOv5s+CBAM0.9390.8840.9230.511
      YOLOv5s+CA0.9220.8860.9310.551
      YOLOv5s+DA0.9450.8980.9420.561
    • Table 3. Textile defect detection data set

      View table
      View in Article

      Table 3. Textile defect detection data set

      PrecisionRecallmAP@0.5mAP@0.5:0.95
      YOLOv5s0.3500.3220.2760.118
      YOLOv5s+CBAM0.3820.2900.2820.141
      YOLOv5s+CA0.2370.2960.2190.086
      YOLOv5s+DA0.3500.3420.2850.121
    • Table 4. Effect comparison of remote sensing target detection loss function

      View table
      View in Article

      Table 4. Effect comparison of remote sensing target detection loss function

      PrecisionRecallmAP@0.5mAP@0.5:0.95
      YOLOv5s+DA+CIoU0.9450.8980.9420.561
      YOLOv5s+DA+EIoU0.9360.9140.9440.573
      YOLOv5s+DA+SIoU0.9330.9220.9350.571
      YOLOv5s+DA+WIoU0.8480.8420.8840.506
    • Table 5. Effect comparison of textile defect detection loss function

      View table
      View in Article

      Table 5. Effect comparison of textile defect detection loss function

      PrecisionRecallmAP@0.5mAP@0.5:0.95
      YOLOv5s+DA+CIoU0.3500.3420.2850.121
      YOLOv5s+DA+EIoU0.3920.3150.2810.130
      YOLOv5s+DA+SIoU0.3820.2720.2830.144
      YOLOv5s+DA+WIoU0.3570.2920.2580.103
    • Table 6. Remote sensing target detection

      View table
      View in Article

      Table 6. Remote sensing target detection

      Params/MPrecisionRecallmAP@0.5mAP@0.5:0.95Ship
      YOLOv5s+DA8.10.9360.9140.9440.5730.973
      YOLOv5s+DA+STD8.70.9630.9030.9430.6130.98
    • Table 7. Textile defect detection

      View table
      View in Article

      Table 7. Textile defect detection

      Params/MPrecisionRecallmAP@0.5mAP@0.5:0.95Knot head
      YOLOv5s+DA8.10.3920.3150.2810.130.303
      YOLOv5s+DA+STD8.70.4540.2580.2930.140.311
    • Table 8. Results of ablation experiments

      View table
      View in Article

      Table 8. Results of ablation experiments

      Params/MPrecisionRecallmAP@0.5mAP@0.5:0.95
      YOLOv5s7.00.9370.8990.9290.562
      YOLOv5s+DA8.10.9450.8980.9420.561
      YOLOv5s+DA+EIoU8.10.9360.9140.9440.573
      YOLOv5s+DA+EIoU+STD8.70.9630.9030.9430.613
    • Table 9. Results of ablation experiments

      View table
      View in Article

      Table 9. Results of ablation experiments

      Params/MPrecisionRecallmAP@0.5mAP@0.5:0.95
      YOLOv5s7.10.3500.3220.2760.118
      YOLOv5s+DA8.10.3500.3420.2850.121
      YOLOv5s+DA+EIoU8.10.3920.3150.2810.130
      YOLOv5s+DA+EIoU+STD8.70.4540.2580.2930.140
    • Table 10. Results of comparison experiments

      View table
      View in Article

      Table 10. Results of comparison experiments

      Params/MPrecisionRecallmAP@0.5mAP@0.5:0.95GFLOPs/G
      Faster-RCNN41.10.8610.9570.9010.55933.2
      YOLOv3-tiny8.60.9520.8600.9290.54412.9
      YOLOv361.50.9560.9180.9520.602155.4
      YOLOv5s7.00.9370.8990.9290.56215.8
      YOLOv5m20.90.8640.8320.8880.52348.0
      YOLOv7-tiny6.00.7860.6310.7680.38613.3
      YOLOv8s11.10.9060.8670.9320.60128.5
      DES-YOLO8.70.9630.9030.9430.61327.9
    Tools

    Get Citation

    Copy Citation Text

    Huawei Zheng, Fei Wang, Jianbang Gao. DES-YOLO: a more accurate object detection method[J]. Opto-Electronic Engineering, 2024, 51(11): 240212-1

    Download Citation

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

    Category: Article

    Received: Sep. 9, 2024

    Accepted: Oct. 11, 2024

    Published Online: Jan. 24, 2025

    The Author Email: Wang Fei (王飞)

    DOI:10.12086/oee.2024.240212

    Topics