基于深度学习网络的航迹分层分类研究  

Research on trajectory hierarchical classification based on deep learning networks

作　　者：王伊凡 吉琳娜[1] 杨风暴[1] WANG Yifan;JI Linna;YANG Fengbao(North University of China,Taiyuan 030000,China)

机构地区：[1]中北大学,山西太原030000

出　　处：《指挥控制与仿真》2025年第2期83-94,共12页Command Control & Simulation

摘　　要：针对现有航迹分类方法无法充分考虑航迹的时间序列特征与空间结构特征,导致分类准确率下降的问题,提出了一种基于深度学习网络的航迹分层分类方法。首先,将船舶航迹转化成图像层,构建基于Swin-Transformer网络的航迹图像层分类模型;其次对于航迹序列层,基于多维信息的航迹压缩算法优化航迹序列的输入,并构建基于Gained-Transformer-Network深度学习网络的航迹序列层分类模型。最后,建立基于置信度的融合层航迹分类模型,提高航迹分层分类的准确率。经过实验验证,平均分类准确率为90%,集成分类器的分类性能相较于其他单分类器平均提高了11%,平均F1分数为0.82。上述结果表明,本文提出的集成分类器对船舶航迹具有较好的分类效果。In response to the issue that existing trajectory classification methods fail to fully consider the time series features and spatial structure features of trajectories,leading to a decline in classification accuracy,this paper proposes a trajectory hierarchical classification method based on deep learning networks.First,ship trajectories are transformed into image layers,and a trajectory image layer classification model based on the Swin-Transformer network is constructed.Next,for the trajectory sequence layer,a multi-dimensional information-based trajectory compression algorithm is utilized to optimize the input of trajectory sequences,and a trajectory sequence layer classification model based on the Gained-Transformer-Network deep learning network is developed.At last,a confidence-based fusion layer trajectory classification model is established to improve the accuracy of layered trajectory classification.Experimental validation shows an average classification accuracy of 90%,with the performance of the ensemble classifier improving by an average of 11%compared to other single classifiers,and an average F1 score of 0.82.The results indicate that the newly proposed method and the new ensemble classifier demonstrate good classification effectiveness for ship trajectories.

关 键 词：航迹分类 深度学习 Swin-Transformer Gained-Transformer-Network 分层分类 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]