• Optics and Precision Engineering
  • Vol. 32, Issue 24, 3616 (2024)
Zhihao ZHANG, Lixia DU, Yue HOU*, Ziwei HAO, and Jie YIN
Author Affiliations
  • College of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou730000, China
  • show less
    DOI: 10.37188/OPE.20243224.3616 Cite this Article
    Zhihao ZHANG, Lixia DU, Yue HOU, Ziwei HAO, Jie YIN. Multi-feature cross UAV image detection algorithm under cross-layer attentional interaction[J]. Optics and Precision Engineering, 2024, 32(24): 3616 Copy Citation Text show less
    Multi-feature crossover UAV target detection network under cross-layer attentional interaction
    Fig. 1. Multi-feature crossover UAV target detection network under cross-layer attentional interaction
    Sructure of the adaptive cross-layer attention interaction module
    Fig. 2. Sructure of the adaptive cross-layer attention interaction module
    Deformable encoder structure
    Fig. 3. Deformable encoder structure
    Structure diagram of the multi-scale feature fusion module
    Fig. 4. Structure diagram of the multi-scale feature fusion module
    Model visualisation process for the MCAI network (features of interest in red)
    Fig. 5. Model visualisation process for the MCAI network (features of interest in red)
    Receptive field visualization experiment
    Fig. 6. Receptive field visualization experiment
    Image visualisation of a rainy day
    Fig. 7. Image visualisation of a rainy day
    Visualization of the actual application scenario of the VisDrone2019-DET dataset
    Fig. 8. Visualization of the actual application scenario of the VisDrone2019-DET dataset
    Visualization of the actual application scenario of LZ Traffic Video
    Fig. 9. Visualization of the actual application scenario of LZ Traffic Video
    MethodsRPmAP0.5mAP0.5∶0.95mAP
    RT-DETR40.856.542.525.734.2
    G-ACAI41.957.644.427.535.6
    G-DMHSA4358.645.528.236.5
    G-MSCF44.258.846.128.536.8
    Table 1. Improved module ablation experiments
    MethodsClsLocDupeBkgMissFPFN
    RT-DETR20.983.820.212.1213.3110.9237.54
    G-ACAI20.793.300.432.2414.0711.2637.08
    G-DMHSA20.813.260.382.1713.8611.0536.65
    G-MSCF20.253.290.372.2713.7611.1236.70
    Table 2. Category error experiments of the improved module
    MethodsmAP0.5@cleanmAP0.5@rainmAP0.5∶0.95@cleanmAP0.5∶0.95@rain
    YOLOv5-Decoder33.921.519.911.7
    YOLOv8-Deocder32.020.918.311.5
    RT-DETR42.532.625.719
    MCAI46.134.728.520.7
    Table 3. Robustness experiments on rainy weather on the VisDrone2019-DET dataset
    MethodBackbonemAPmAP0.5Param
    Faster-RCNNResNet5021.535.741.7
    RetinaNetResNet5016.127.321.37
    CenterNetDLA-3412.422.7025.6
    SSDVGG-168.616.64.23
    YOLOv3MobileNet8.117.73.75
    YOLOv4ResNet5018.625.386.06
    YOLOV5sCSPDarkNet19.129.87.05
    YOLOv7CSPDarkNet22.934.537.26
    RT-DETRResNet1832.642.521.3
    ATSS-FPN-DyHeadResNet5020.433.838.91
    TOODResNet5020.433.932.4
    DINODETR25.344.547.56
    YOLOX-TinyCSPDarkNet14.827.85.035
    GFLCSPDarkNet19.332.132.279
    RTMDetCSPDarkNet18.431.24.876
    MCAIResNet1836.846.122.4
    Table 4. Comparison of results of different algorithmic models on VisDrone2019-DET
    MethodYearBackBonemAP0.5/%
    Faster-RCNN2016ResNet5034.45
    SSD2017VGG-1628.42
    RetinaNet2018ResNet5037.65
    CenterNet2020DLA-3440.40
    YOLOv32018MobileNet42.4
    YOLOv42020CSPDarkNet45.3
    YOLOv5s2020CSPDarkNet61
    YOLOv72022CSPDarkNet48.7
    RTDETR2023ResNet1870.2
    ATSS-FPN-DyHead2024ResNet5066.9
    TOOD2021ResNet5069.1
    DINO2023ResNet5070.4
    YOLOX-Tiny2021CSPDarkNet62.3
    GFL2020CSPDarkNet66.8
    RTMDet2023CSPDarkNet68.4
    MCAI2024ResNet1872.4
    Table 5. Overall performance comparison of BDD-100K dataset
    BaselineRecallPrecisionmAP0.5mAP0.5∶0.95
    YOLOv341.482.652.130.9
    YOLOv548.588.156.738.6
    YOLOv640.47245.231.7
    YOLOv847.461.352.136.7

    YOLOv8-

    Decoder

    49.576.755.137.2
    RT-DETR64.987.976.151.7
    MCAI75.882.880.659.9
    Table 6. Overall performance comparison of LZ traffic video detection dataset
    Zhihao ZHANG, Lixia DU, Yue HOU, Ziwei HAO, Jie YIN. Multi-feature cross UAV image detection algorithm under cross-layer attentional interaction[J]. Optics and Precision Engineering, 2024, 32(24): 3616
    Download Citation