基于SAR和AIS的角度最近邻数据关联方法

提升海上态势感知能力是构建智慧海洋的重要环节。针对目前海上目标研究单源传感器存在感知盲区，多源传感器数据关联易受杂波干扰、在密集区表现不佳等问题，本文基于合成孔径雷达(SAR)和船舶自动识别系统(AIS)数据，提出一种抗干扰性强的角度最近邻数据关联方法，充分利用SAR与AIS船舶目标的空间角度关系，提高船舶目标在密集区域点迹关联的准确性。首先，对AIS数据进行时空滤波，实现数据粗关联，构建关联分析的数据候选集；然后，从时空数据的空间关系角度出发，在灰狼优化和匈牙利算法的启发下，利用点迹对特征向量矩阵进行运算，实现对多源空间数据的优化关联；最后结合数据几何关系对结果进行置信度评估。本文选取5幅SAR影像与AIS数据进行实验，并基于SAR影像数据及船舶轨迹点分布密度设计仿真实验，结果表明，本文所提出的角度最近邻数据关联方法，在密集分布情况下，关联精度为传统NN、GNN算法的3.62和4.61倍，运行时间为1.69 s，相较于NN算法仅增长1.36 s，仅占GNN运行时间的0.49%，在运行时间增长不大的情况下具有更强的抗干扰能力，在密集区域仍能取得较好的关联效果。

A method of Angular Nearest Neighbor Data Association based on SAR and AIS

Improving maritime situation awareness is important for building an intelligent ocean. In order to solve the problems of perception blind area of single source sensor and poor performance of multi-source sensor data association in dense area, this paper proposes an angle nearest neighbor data association method with strong anti-interference based on Synthetic Aperture Radar (SAR) and Automatic Identification System (AIS) data. The spatial angle relationship between SAR and AIS is utilized to improve the accuracy of point trace correlation of ship targets in dense areas. Firstly, the spatiotemporal filtering is performed on AIS data to realize data coarse association, and a data candidate set for association analysis is constructed. Then, from the perspective of spatial relationship of spatiotemporal data, inspired by gray wolf optimization and Hungarian algorithm, the eigenvector matrix of point trace pair is operated to achieve the optimal association of multi-source spatial data. Finally, the confidence of the results is evaluated by combining the geometric relationship of the data. In this paper, five SAR images and AIS data are selected for experiments, and simulation experiments are designed based on SAR image data and the distribution density of ship trajectory points. The results show that, in the case of dense distribution, the correlation accuracy of the angular nearest neighbor data association method proposed in this paper is 3.62 and 4.61 times that of the traditional nearest neighbor algorithm and global nearest neighbor algorithm, respectively. The running time of our proposed method is 1.69 s, which is only 1.36 s longer than nearest neighbor algorithm and only accounts for 0.49 percent of global nearest neighbor algorithm’s running time. By comparison, our proposed method has stronger anti-interference ability and can achieve good correlation accuracy in dense areas with little increase in the running time.