Journal of Applied Optics, Volume. 45, Issue 2, 346(2024)

Infrared target detection algorithm based on improved Faster R-CNN

Xichen WANG1... Fulun PENG2, Yexun LI3 and Junju ZHANG1,* |Show fewer author(s)
Author Affiliations
  • 1School of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
  • 2Xi'an Institute of Applied Optics, Xi'an 710065, China
  • 3Jiangsu North Huguang Photoelectric Co.,Ltd., Wuxi 214100, China
  • show less
    Figures & Tables(15)
    Structure diagram of Faster R-CNN algorithm
    Structure diagram of image preprocessing
    Image enhancement contrast
    Residual structure diagram of FCA-ResNet
    Schematic diagram of SE and FCA attention mechanism
    Two-dimensional discrete cosine transform
    PAFPN structure diagram with FCA-ResNet50 as main stem
    Loss curve
    Detection results of algorithm
    Visualization of heat map of P2,P3 and P4 layers in different backbones
    Comparison of detection results of different algorithms
    • Table 1. Parameters setting

      View table
      View in Article

      Table 1. Parameters setting

      参数设置
      输入图像大小/像素800×800
      优化器SGD
      权重衰减系数0.0001
      初始学习率0.005
      批大小(batch_size)4
      训练周期/epoch20
      迭代次数/次8000
    • Table 2. Detection results of different image enhancement methods

      View table
      View in Article

      Table 2. Detection results of different image enhancement methods

      模型图像增强AP
      直方图均衡伽马变换
      ResNet50+FPN84.6
      84.9
      84.7
      85.1
    • Table 3. Detection results of different models

      View table
      View in Article

      Table 3. Detection results of different models

      BackboneNeckAPFLOPs/G参数量/M
      ResNet50FPN85.1134.3841.12
      SE-ResNet50FPN88.7134.4543.63
      FCA-ResNet50FPN90.0134.3843.61
      FCA-ResNet50PAFPN92.7149.8747.15
    • Table 4. Detection results for mainstream algorithmic frameworks

      View table
      View in Article

      Table 4. Detection results for mainstream algorithmic frameworks

      Model图像 增强BackboneNeckAPFLOPs/GParam/M
      One-stage
      YOLOv3×DarkNet53FPN82.1121.1561.52
      RetinaNet×ResNet50FPN83.2127.8236.1
      Fcos×ResNet50FPN84.7125.4132.02
      Two-stage
      Casade R-CNNResNet50FPN86.7162.1868.93
      Mask R-CNNResNet50FPN86.1187.4543.77
      Faster R-CNNResNet50FPN85.1134.3841.12
      本文算法FCA- ResNet50PAFPN92.7134.3843.61
    Tools

    Get Citation

    Copy Citation Text

    Xichen WANG, Fulun PENG, Yexun LI, Junju ZHANG. Infrared target detection algorithm based on improved Faster R-CNN[J]. Journal of Applied Optics, 2024, 45(2): 346

    Download Citation

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

    Category: Research Articles

    Received: Apr. 11, 2023

    Accepted: --

    Published Online: May. 28, 2024

    The Author Email: Junju ZHANG (张俊举(1979—))

    DOI:10.5768/JAO202445.0202001

    Topics