Multisource Track Association Dataset (MTAD) Based on the Global AIS

Solemnly Declare: when using this data set to publish papers, books and other works, you must formally quote the papers to which this data set belongs:Citation: CUI Yaqi, XU Pingliang, GONG Cheng, YU Zhouchuan, ZHANG Jianting, YU Hongbo, DONG Kai. Multisource Track Association Dataset Based on the Global AIS[J]. Journal of Electronics & Information Technology, 2023, 45(2): 746-756. doi: 10.11999/JEIT221202Authors: Ren Junyu, Yu Ningning, Zhou Chengwei, Shi Zhiguo, Chen JimingAuthor Unit:Institute of Information Fusion, Naval Aviation University91001 Unit91001 UnitCorrespondent: XU Pingliang，[e-mail] link：基于全球AIS的多源航迹关联数据集Funds: The National Natural Science Foundation of China (61790554, 62001499, 62171453)Abstract: Data, algorithms, and hash rates are the three thrust forces for developing artificial intelligence. Considering the urgent demand for research on the intelligent association algorithm and the difficulty of obtaining track data from multi-radar collaborative observation and addressing the problem of missing track association dataset, a Multi-source Track Association Dataset (MTAD) is constructed in this study. MTAD is based on automatic identification system trajectory data after processing grid division, automatic interruption, and error adding. The dataset includes two parts, namely, the training dataset and the test dataset, with more than 1 million tracks. The train and test datasets contain 5000 and 1000 scene samples, respectively. Each scene sample consists of several to hundreds of tracks, covering various movement patterns, target types, and duration times. In addition, the constructed MTAD is further visualized and analyzed, and the characteristics of tracks in each grid are studied in detail, demonstrating the richness, rationality, and effectiveness of the MTAD. The indicators and baseline results of the association are obtained. This dataset has already been used as a dedicated dataset for the Navy’s “Golden Dolphin” Cup competition.

摘要：数据、算法与算力是人工智能发展的三大核心驱动力。针对智能关联算法研究的迫切需求，以及多雷达协同观测航迹数据获取难度大、航迹关联数据集缺失的问题，本研究构建了多源航迹关联数据集（Multi-source Track Association Dataset，简称MTAD）。MTAD基于经过网格划分、自动截断及误差添加处理后的自动识别系统（Automatic Identification System，简称AIS）轨迹数据构建。该数据集包含训练集与测试集两部分，总航迹数量超百万条。其中训练集与测试集分别包含5000组与1000组场景样本，每组场景样本包含数条至数百条航迹，涵盖多种运动模式、目标类型与持续时长。此外，本研究对构建的MTAD进行了可视化分析，详细研究了各网格内航迹的分布特征，验证了该数据集的丰富性、合理性与有效性，并得到了关联任务的评价指标与基准结果。该数据集已作为海军“金海豚”杯赛事的专用数据集投入使用。引用：崔雅琪, 徐平亮, 龚诚, 余周川, 张健挺, 于洪波, 董凯. 基于全球自动识别系统的多源航迹关联数据集[J]. 电子与信息学报, 2023, 45(2): 746-756. doi: 10.11999/JEIT221202（论文链接：https://jeit.ac.cn/cn/article/doi/10.11999/JEIT221202）

搜集汇总

背景与挑战

背景概述

该数据集是一个基于全球自动识别系统（AIS）的多源轨迹关联数据集（MTAD），旨在解决智能关联算法研究中多雷达协同观测轨迹数据获取困难和关联数据集缺失的问题。它包含超过100万条轨迹，分为5000个训练场景样本和1000个测试场景样本，每个样本涵盖多种运动模式、目标类型和持续时间，经过网格划分和误差处理，已用于海军竞赛并验证了其丰富性和有效性。