基于全球AIS的多源航迹关联数据集

Multisource Track Association Dataset Based on the Global AIS

数据、算法和算力是当前人工智能技术发展的3大推力,考虑到智能关联算法研究的迫切需求和多雷达协同观测航迹数据获取困难,针对航迹关联数据集缺失问题,该文公开了多源航迹关联数据集(MTAD),其由全球AIS航迹数据经栅格划分、自动中断和噪声添加处理步骤构建。该数据集包括训练集和测试集两大部分,共有航迹百万余条,其中训练集包含5000个场景样本,测试集包含1000个场景样本,每一个场景样本由几个到几百个数量不等的航迹构成,涵盖多种运动模式、多种目标类型和长度不等的持续时间。同时,进一步对构造的MTAD数据集进行可视化分析,详细研究了各个栅格内航迹的特点,证明了该数据集的丰富性、合理性和有效性。最后,作为参考,给出了关联评价指标和关联基线结果。该数据集目前已被用作海军“金海豚”杯竞赛科目专用数据集。

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.