留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于EfficientDet的无预训练SAR图像船舶检测器

包壮壮 赵学军

包壮壮, 赵学军. 基于EfficientDet的无预训练SAR图像船舶检测器[J]. 北京航空航天大学学报, 2021, 47(8): 1664-1672. doi: 10.13700/j.bh.1001-5965.2020.0255
引用本文: 包壮壮, 赵学军. 基于EfficientDet的无预训练SAR图像船舶检测器[J]. 北京航空航天大学学报, 2021, 47(8): 1664-1672. doi: 10.13700/j.bh.1001-5965.2020.0255
BAO Zhuangzhuang, ZHAO Xuejun. Ship detector in SAR images based on EfficientDet without pre-training[J]. Journal of Beijing University of Aeronautics and Astronautics, 2021, 47(8): 1664-1672. doi: 10.13700/j.bh.1001-5965.2020.0255(in Chinese)
Citation: BAO Zhuangzhuang, ZHAO Xuejun. Ship detector in SAR images based on EfficientDet without pre-training[J]. Journal of Beijing University of Aeronautics and Astronautics, 2021, 47(8): 1664-1672. doi: 10.13700/j.bh.1001-5965.2020.0255(in Chinese)

基于EfficientDet的无预训练SAR图像船舶检测器

doi: 10.13700/j.bh.1001-5965.2020.0255
详细信息
    通讯作者:

    赵学军. E-mail: 292457155@qq.com

  • 中图分类号: TN957.51;TP751

Ship detector in SAR images based on EfficientDet without pre-training

More Information
  • 摘要:

    针对多尺度、多场景的合成孔径雷达(SAR)图像船舶检测问题,提出了一种基于EfficientDet的无预训练目标检测器。现有的基于卷积神经网络的SAR图像船舶检测器并没有表现出其应有的出色性能。重要原因之一是依赖分类任务的预训练模型,没有有效的方法来解决SAR图像与自然场景图像之间存在的差异性;另一个重要原因是没有充分利用卷积神经网络各层的信息,特征融合能力不够强,难以处理包括海上和近海在内的多场景船舶检测,尤其是无法排除近海复杂背景的干扰。SED就这2个方面改进方法,在公开SAR船舶检测数据集上进行实验,检测精度指标平均准确率(AP)达到94.2%,与经典的深度学习检测器对比,超过最优的RetineNet模型1.3%,在模型大小、算力消耗和检测速度之间达到平衡,验证了所提模型在多场景条件下多尺度SAR图像船舶检测具有优异的性能。

     

  • 图 1  数据归一化的方式

    Figure 1.  Methods of data normalization

    图 2  两种卷积对比

    Figure 2.  Comparison of two convolutions

    图 3  残差模块和倒残差模块数据流图对比

    Figure 3.  Comparison of data flow graph between residual blocks and inverted residual blocks

    图 4  特征融合网络设计

    Figure 4.  Design of feature fusion network

    图 5  网络结构示意图

    Figure 5.  Schematic diagram of network structure

    图 6  复杂背景下的船舶数据集可视化

    Figure 6.  Visualization of a ship dataset in complex background

    图 7  不同模型预测结果可视化

    Figure 7.  Visualized prediction results of different models

    表  1  消融实验

    Table  1.   Ablation experiment

    条件 组成
    预训练
    BN
    GN
    BiFPN
    改进BiFPN
    AP0.5/% 92.3 93.4 93.5 93.7 94.2 Nan
    AP0.5∶0.95/% 60.0 59.9 60.6 63.3 64.7 Nan
    下载: 导出CSV

    表  2  不同模型结果对比

    Table  2.   Comparison of results among different models

    指标 SSD300 SSD512 Faster R-CNN
    (R50)
    RetinaNet
    (R50)
    EfficientDet-D0
    (预训练)
    EfficientDet-D4
    (预训练)
    SED
    AP0.5/% 88.5 89.6 91.8 92.9 92.3 93.4 94.2
    AP0.5∶0.95/% 49.1 51.4 54.9 57.1 60.0 62.7 64.7
    训练时长/min 10 43 23 15 14 195 19
    测试时长/s 77 126 114 115 144 326 227
    图像处理速度/(fp·s-1) 56.6 36.3 38.6 38.0 30.5 13.5 19.3
    模型大小/MB 190.0 195.0 247.6 303.2 15.7 83.2 15.4
    下载: 导出CSV
  • [1] KANJIR U, GREIDANUS H, KRIS ˇTOF O. Vessel detection and classification from spaceborne optical images: A literature survey[J]. Remote Sensing of Envioronment, 2018, 207: 1-26. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC5877374/
    [2] WANG Y, WANG C, ZHANG H, et al. Automatic ship detection based on retinanet using multi-resolution gaofen-3 imagery[J]. Remote Sensing, 2019, 11(5): 531. doi: 10.3390/rs11050531
    [3] EL-DARYMLI K, GILL E W, MCGUIRE P, et al. Automatic target recognition in synthetic aperture tadar imagery: A state-of-the-art review[J]. IEEE Access, 2016, 4: 6014-6058. doi: 10.1109/ACCESS.2016.2611492
    [4] YANG C S, PARK J H, RASHID A. An improved method of land masking for synthetic aperture radar-based ship detection[J]. Journal of Navigation, 2018, 71(4): 788-804. doi: 10.1017/S037346331800005X
    [5] MOLINA D E, GLEICH D, DATCU M, et al. Gibbs random field models for model-based despeckling of SAR images[J]. IEEE Geoscience and Remote Sensing Letters, 2009, 7(1): 73-77. http://ieeexplore.ieee.org/document/5071276
    [6] QIN X X, ZHOU S L, ZOU H X, et al. A CFAR detection algorithm for generalized gamma distributed background in high-resolution SAR images[J]. IEEE Geoscience and Remote Sensing Letters, 2013, 10(4): 806-810. doi: 10.1109/LGRS.2012.2224317
    [7] ZHAO J, ZHANG Z, YU W, et al. A cascade coupled convolutional neural network guided visual attention method for ship detection from SAR images[J]. IEEE Access, 2018, 6: 50693-50708. doi: 10.1109/ACCESS.2018.2869289
    [8] LI J, QU C, SHAO J. Ship detection in SAR images based on an improved faster R-CNN[C]//2017 SAR in Big Data Era: Models, Methods and Applications. Piscataway: IEEE Press, 2017: 1-6.
    [9] WANG Y, WANG C, ZHANG H, et al. A SAR dataset of ship detection for deep learning under complex backgrounds[J]. Remote Sensing, 2019, 11(7): 765. doi: 10.3390/rs11070765
    [10] LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single shot multiBox detector[C]//European Conference on Computer Vision. Berlin: Springer, 2016: 21-37.
    [11] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[C]//Advances in Neural Information Processing Systems, 2015: 91-99.
    [12] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//2017 IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 2999-3007.
    [13] NOGUEIRA K, PENATTI O A, DOS SANTOS J A. Towards better exploiting convolutional neural networks for remote sensing scene classification[J]. Pattern Recognition, 2017, 61: 539-556. doi: 10.1016/j.patcog.2016.07.001
    [14] SHEN Z, LIU Z, LI J, et al. DSOD: Learning deeply supervised object detectors from scratch[C]//2017 IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 1919-1927.
    [15] ZHU R, ZHANG S, WANG X, et al. ScratchDet: Training single-shot object detectors from scratch[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2019: 2268-2277.
    [16] LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2017: 2117-2125.
    [17] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2016: 770-778.
    [18] TAN M, PANG R, LE Q V. EfficientDet: Scalable and efficient object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE Press, 2020: 10781-10790.
    [19] GIRSHICK R. Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2015: 1440-1448.
    [20] HE K, GKIOXARI G, DOLLÁR P, et al. Mask R-CNN[C]//2017 IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 2961-2969.
    [21] REDMON J, FARHADI A. YOLOv3: An incremental improvement[EB/OL]. [2020-05-13]. https://arxiv.org/abs/1804.02767.
    [22] TIAN Z, SHEN C, CHEN H, et al. FCOS: Fully convolutional one-stage object detection[C]//2019 IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2019: 9627-9636.
    [23] LIU S, QI L, QIN H, et al. Path aggregation network for instance segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 8759-8768.
    [24] GHIASI G, LIN TY, LE Q V. NAS-FPN: Learning scalable feature pyramid architecture for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2019: 7036-7045.
    [25] IOFFE S, SZEGEDY C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the 32nd International Conference on International Conference on Machine Learning. New York: ACM Press, 2015: 448-456.
    [26] WU Y, HE K. Group normalization[C]//European Conference on Computer Vision. Berlin: Springer, 2018: 3-19.
    [27] HOWARD A G, ZHU M, CHEN B, et al. MobileNets: Efficient convolutional neural networks for mobile vision applications[EB/OL]. [2020-05-13]. https://arxiv.org/abs/1704.04861.
    [28] SANDLER M, HOWARD A, ZHU M, et al. MobileNetV2: Inverted residuals and linear bottlenecks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 4510-4520.
    [29] CHEN X, FANG H, LIN T, et al. Microsoft COCO captions: Data collection and evaluation server[J]. (2015-04-03)[2020-05-13]. https://arxiv.org/abs/1504.00325.
    [30] KINGMA D P, BA J. Adam: A method for stochastic optimization[EB/OL]. (2017-01-30)[2020-05-13]. https://arxiv.org/abs/1412.6980v9.
  • 加载中
图(7) / 表(2)
计量
  • 文章访问数:  178
  • HTML全文浏览量:  7
  • PDF下载量:  157
  • 被引次数: 0
出版历程
  • 收稿日期:  2020-06-11
  • 录用日期:  2020-09-04
  • 刊出日期:  2021-08-20

目录

    /

    返回文章
    返回
    常见问答