航空学报 > 2024, Vol. 45 Issue (2): 328672-328672   doi: 10.7527/S1000-6893.2023.28672

AIS和光学遥感图像引导的星载SAR舰船目标识别网络

王子玲1, 熊振宇1(), 杨璐铖2, 杨蕊宁3, 黄林周4   

  1. 1.海军航空大学 信息融合研究所,烟台 264001
    2.中国人民解放军 91033部队,青岛 266000
    3.重庆市勘测院,重庆 401120
    4.重庆市地理信息和遥感应用中心,重庆 401120
  • 收稿日期:2023-03-09 修回日期:2023-04-12 接受日期:2023-05-30 出版日期:2024-01-25 发布日期:2023-05-31
  • 通讯作者: 熊振宇 E-mail:x_zhen_yu@163.com
  • 基金资助:
    青年科学基金(62001499);国家自然科学基金(61790554)

Spaceborne SAR ship target recognition network guided by AIS and optical remote sensing images

Ziling WANG1, Zhenyu XIONG1(), Lucheng YANG2, Ruining YANG3, Linzhou HUANG4   

  1. 1.Institute of Information Fusion,Naval Aviation University,Yantai 264001,China
    2.No. 91033 Unit of the People’s Liberation Army of China,Qingdao 266000,China
    3.Chongqing Survey Institute,Chongqing 401120,China
    4.Chongqing Geomatics and Remote Sensing Center,Chongqing 401120,China
  • Received:2023-03-09 Revised:2023-04-12 Accepted:2023-05-30 Online:2024-01-25 Published:2023-05-31
  • Contact: Zhenyu XIONG E-mail:x_zhen_yu@163.com
  • Supported by:
    National Science Fund for Young Scholars(62001499);National Natural Science Foundation of China(61790554)

摘要:

星载SAR作为全天时、全天候的感知手段广泛应用于海洋目标识别任务中,由于SAR图像分辨率低、难解译、样本不均衡导致现有单一模态目标识别算法识别精度低。提出了一种AIS和光学遥感图像引导的星载SAR舰船目标识别网络,针对不同模态数据特征维度不同导致难度量问题,利用特征迁移模块在保留各自模态独有特征属性前提下将异构特征映射到共同的空间中度量;针对不同模态不同类别数据存在样本不均衡问题,利用异构特征对齐模块充分挖掘不同模态的互补信息,以细粒度的方式进一步对齐不同模态的异构特征,同时将各个模态的判别性特征作为先验信息迁移至SAR图像模态中。实验部分利用AIS历史数据和光学遥感数据集作为辅助信息,在2个公开的SAR图像舰船目标数据集中进行测试,实验结果表明所提算法通过迁移不同模态信息,有效提升了SAR图像舰船目标的识别准确率。

关键词: 多源目标, 遥感图像, 舰船目标识别, AIS数据, 异构特征

Abstract:

Spaceborne SAR is widely used in marine target recognition tasks as an all-season and all-weather sensing means. Due to the low resolution, difficult interpretation and uneven samples of SAR images, the existing single-mode target recognition algorithms have low recognition accuracy. In this paper, a spaceborne SAR ship target recognition network guided by AIS and optical remote sensing images is proposed. To overcome the difficulty caused by different feature dimensions of different modal data, the heterogeneous features are mapped into the common space measurement by using the feature migration module on the premise of preserving the unique feature attributes of each modal. For the problem of sample imbalance in different modes and different categories of data, the heterogeneous feature alignment module is used to fully mine the complementary information of different modes, further align the heterogeneous features of different modes in a fine-grained way, and migrate the discriminant features of each mode as a priori information to SAR image modes. In the experimental part, AIS historical data and optical remote sensing data set are used as auxiliary information on two public SAR image ship target data sets. The experimental results show that the network proposed can effectively improve the recognition accuracy of ship targets in SAR images by fusing different modal information.

Key words: multi-source ship target, remote sensing image, ship target recognition, AIS data, heterogeneous characteristic

中图分类号: