航空学报 > 2026, Vol. 47 Issue (10): 533060-533060   doi: 10.7527/S1000-6893.2026.33060

航天遥感图像智能处理与分析专刊

基于多视角低空遥感图像的船舶目标关联

刘云鹤1, 姜智卓2(), 刘瑜3, 孙显4,5, 何友3   

  1. 1.清华大学 深圳国际研究生院,深圳 518000
    2.南开大学 计算机学院,天津 300071
    3.清华大学 电子工程系,北京 100084
    4.中国科学院 空天信息创新研究院,北京 100094
    5.中国科学院 网络信息体系技术重点实验室,北京 100190
  • 收稿日期:2025-11-07 修回日期:2025-12-03 接受日期:2026-01-04 出版日期:2026-01-16 发布日期:2026-01-15
  • 通讯作者: 姜智卓 E-mail:jiangzhizhuo@nankai.edu.cn
  • 基金资助:
    国家自然科学基金(62401335)

Vessel target association based on multi-view low-altitude remote sensing images

Yunhe LIU1, Zhizhuo JIANG2(), Yu LIU3, Xian SUN4,5, You HE3   

  1. 1.Tsinghua Shenzhen International Graduate School,Tsinghua University,Shenzhen 518000,China
    2.College of Computer Science,Nankai University,Tianjin 300071,China
    3.Department of Electronic Engineering,Tsinghua University,Beijing 100084,China
    4.Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China
    5.Key Laboratory of Network Information System Technology,Chinese Academy of Sciences,Beijing 100190,China
  • Received:2025-11-07 Revised:2025-12-03 Accepted:2026-01-04 Online:2026-01-16 Published:2026-01-15
  • Contact: Zhizhuo JIANG E-mail:jiangzhizhuo@nankai.edu.cn
  • Supported by:
    National Natural Science Foundation of China(62401335)

摘要:

低空遥感场景下的船舶目标关联技术是推动海上监测及其智能感知系统发展的重要支撑。然而,现有方法多直接迁移行人或车辆重识别算法,难以有效应对船舶图像中的特有问题,尤其是因无人机等低空遥感平台成像视角多变导致的类内差异大、局部信息缺失等挑战,这往往导致同一船舶目标出现异常样本,极大影响关联精度。为了解决上述问题,提出一种基于多尺度相关性Transformer网络的船舶目标关联算法。与现有方法不同,该算法能够同时对输入图像集合进行多尺度显式的全局和局部相关性建模,且在模型训练时,不只依赖单幅图像的孤立特征进行学习,而是融合利用图像间的互补信息,抑制由类内差异或局部缺失引起的异常样本影响。具体而言,设计了全局关联模块,构建完整输入图像间的全局相似性关联矩阵,基于图像间一致性进行特征聚合,实现显式全局相关性建模;同时设计了局部关联模块,构建一个基于动态更新机制的记忆库,挖掘并对齐正样本的局部特征,通过上下文相似性提取局部相关性。在4个公开实测数据集上的实验结果表明:所提方法在目标关联准确度的性能指标上均优于现有主流方法,验证了其有效性、鲁棒性与工程实用潜力。

关键词: 低空遥感, 船舶目标关联, 相关性建模, 特征增强, 特征融合

Abstract:

Vessel target association under low-altitude remote-sensing scenarios is a crucial component supporting the development of maritime monitoring and intelligent perception systems. However, most existing approaches directly migrate pedestrian or vehicle re-identification algorithms, which fail to effectively handle the unique challenges of vessel imagery-particularly the large intra-class variations and local information loss caused by the diverse imaging perspectives of UAV-based low-altitude imaging platforms. These issues often lead to outlier samples within the same vessel identity, significantly degrading association accuracy. To overcome these limitations, this paper proposes a Multi-scale Correlation-aware Transformer network (MCFormer) for vessel target association. Unlike conventional methods that learn from isolated features of single images, MCFormer performs explicit global and local correlation modeling across multi-scale image collections, leveraging inter-image complementary information to suppress the effects of intra-identity variance and partial occlusion. Specifically, a Global Correlation Module (GCM) constructs a comprehensive inter-image similarity matrix to achieve explicit global correlation modeling through consistency-based feature aggregation, while a Local Correlation Module (LCM) builds a dynamically updated memory bank to mine and align positive local features, capturing fine-grained contextual correlations. Experiments conducted on four publicly available real-world datasets demonstrate that the proposed method consistently outperforms mainstream method in performance metrics related to target association accuracy, verifying its effectiveness, robustness, and engineering potential.

Key words: low-altitude remote sensing, vessel target association, correlation modeling, feature enhancement, feature fusion

中图分类号: