基于两阶段注意力层Transformer的弹道目标多站融合识别
Multi-station Fusion Recognition of Ballistic Targets Based on Two-stage Attention Transformer
翟相龙 1王旋 1王雁冰 1王峰1
作者信息
- 1. 复旦大学信息科学与工程学院,上海 200433
- 折叠
摘要
多站弹道目标融合识别旨在利用多个雷达站点信息的互补性提升弹道目标识别性能,而传统多站下弹道目标识别方法未直接考虑多站数据间的关联特性,难以取得准确、稳健的识别性能.针对多站下基于雷达散射截面积(RCS)高速飞行目标的识别问题,提出了一种基于两阶段注意力的弹道目标多站融合识别方法.首先,在现有的Transformer模型上添加维度分段模块将多站雷达数据嵌入于二维向量中,保留站内数据时序及站间关联信息;然后,添加了两阶段注意力层,有效地捕获站内时序信息及跨站维度的依赖关系;最后,基于仿真动态RCS数据模拟多站场景开展了融合识别实验.实验结果表明该方法能够有效提升多站条件下的弹道目标识别性能.
Abstract
Multi-station ballistic target fusion recognition aims to enhance the recognition performance of ballistic targets by levera-ging the complementarity of radar station information from multiple radar stations.Traditional methods for multi-station ballistic tar-get recognition do not directly consider the inherent data characteristics between stations,typically overlooking inter-station data correlations during the decision-making process.This paper addresses the problem of high-speed flying target recognition in a multi-station context based on dynamic radar cross-section(RCS).A two-stage attention-based ballistic target multi-station fusion recog-nition approach is proposed.Firstly,a dimension segmentation module is added to the existing Transformer model to embed multi-station radar data into 2-dimensional vectors,preserving both intra-station temporal and inter-station correlation information.Sec-ondly,two-stage attention layers are incorporated to effectively capture intra-station temporal information and inter-station dimen-sional dependencies.Finally,fusion recognition experiments are conducted using simulated dynamic RCS data to simulate multi-station scenarios,demonstrating that the proposed approach can significantly enhance the recognition performance of ballistic targets under multi-station condition.
关键词
雷达散射截面/多站融合弹道目标识别/Transformer模型/两阶段注意力Key words
radar cross section/multi-station ballistic target fusion recognition/Transformer model/two-stage attention引用本文复制引用
基金项目
国家自然科学基金资助项目(61991421)
出版年
2024
国家科技期刊平台
NSTL
万方数据