1148 工程科学学报，第43卷，第9期 Lake City,2018:8697 [40]Pang J M,Chen K,Shi J P,et al.Libra R-CNN:Towards balanced [36]Ma NN,Zhang X Y,Zheng H T,et al.ShuffleNet V2:practical leaming for object detection /2019 IEEE/CVF Conference on guidelines for efficient CNN architecture design /Computer Computer Vision and Pattern Recognition(CVPR).Long Beach, Vision-ECCV 2018.Munich,2018:122 2019:821 [37]Szegedy C,loffe S,Vanhoucke V,et al.Inception-v4,inception- [41]Li Y H,Chen Y T,Wang N Y,et al.Scale-aware trident networks ResNet and the impact of residual connections on learning / for object detection /2019 IEEE/CVF International Conference Proceedings of the AAAI Conference on Artificial Intelligence.San on Computer Vision (ICCV).Seoul,2019:6053 Francisco,2017(31片1 [42]KongT,Sun FC,Liu H P,et al.FoveaBox:beyound anchor-based [38]Chollet F.Xception:deep learning with depthwise separable object detection.IEEE Trans Image Process,2020,29:7389 convolutions II 2017 IEEE Conference on Computer Vision and [43]Zhang S F,Chi C,Yao Y Q,et al.Bridging the gap between Pattern Recognition (CVPR).Honolulu,2017:1800 anchor-based and anchor-free detection via adaptive training [39]Cai Z W,Vasconcelos N.Cascade R-CNN:Delving into high sample selection I 2020 IEEE/CVF Conference on Computer quality object detection 2018 IEEE/CVF Conference on Vision and Pattern Recognition(CVPR).Seattle,2020:9756 Computer Vision and Pattern Recognition.Salt Lake City,2018: [44]Ultralytics.YoloV5[EB/OL].Github(2020-10-12)[2021-06-24]. 6154 https://github.com/ultralytics/YoloV5Lake City, 2018: 8697 Ma N N, Zhang X Y, Zheng H T, et al. ShuffleNet V2: practical guidelines  for  efficient  CNN  architecture  design  // Computer Vision – ECCV 2018. Munich, 2018: 122 [36] Szegedy C, Ioffe S, Vanhoucke V, et al. Inception-v4, inception￾ResNet  and  the  impact  of  residual  connections  on  learning  // Proceedings of the AAAI Conference on Artificial Intelligence. San Francisco, 2017(31): 1 [37] Chollet  F.  Xception:  deep  learning  with  depthwise  separable convolutions  //  2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, 2017: 1800 [38] Cai  Z  W,  Vasconcelos  N.  Cascade  R-CNN:  Delving  into  high quality  object  detection  //  2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.  Salt  Lake  City,  2018: 6154 [39] Pang J M, Chen K, Shi J P, et al. Libra R-CNN: Towards balanced learning  for  object  detection  //  2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).  Long  Beach, 2019: 821 [40] Li Y H, Chen Y T, Wang N Y, et al. Scale-aware trident networks for  object  detection  //  2019 IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, 2019: 6053 [41] Kong T, Sun F C, Liu H P, et al. FoveaBox: beyound anchor-based object detection. IEEE Trans Image Process, 2020, 29: 7389 [42] Zhang  S  F,  Chi  C,  Yao  Y  Q,  et  al.  Bridging  the  gap  between anchor-based  and  anchor-free  detection  via  adaptive  training sample  selection  //  2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, 2020: 9756 [43] Ultralytics.  YoloV5[EB/OL].  Github(2020-10-12)[2021-06-24]. https://github.com/ultralytics/YoloV5 [44] · 1148 · 工程科学学报，第 43 卷，第 9 期