现代雷达2025，Vol.47Issue(11)：15-21,7.DOI:10.16592/j.cnki.1004-7859.2025092202
基于时序特征与注意力机制的多模态高动态场景空间目标识别方法
A Multi-modal Space Target Recognition Method for Highly Dynamic Scenarios Based on Temporal Features and Attention Mechanism
穆思远 1姜媛媛 2庞钰宁 3周嘉琪 1李东瀛 1魏飞鸣 1郁文贤1
作者信息
- 1. 上海市北斗导航与位置服务重点实验室,上海 200240||上海交通大学集成电路学院,上海 200240
- 2. 上海航天技术研究院,上海 201109
- 3. 太原卫星发射中心,宁夏 银川 750004
- 折叠
摘要
Abstract
In order to address the limitation that single-modality radar features are highly susceptible to noise and attitude variations in high-dynamic space target recognition,a multi-modal recognition framework based on temporal alignment and attention mecha-nisms is presented in this paper,which leverages the complementary nature of motion features and radar cross-section(RCS)se-quences.Firstly,a temporal convolutional network is employed to extract motion patterns from multi-frame trajectories,while mod-eling of the RCS sequence is performed using sliding-window convolution,so as to capture local temporal variations.Next,prior to fusion,temporal alignment and attentive pooling are introduced to associate high-frequency motion features with low-frequency RCS features,thereby reducing temporal mismatches.Finally,a cross-attention mechanism is utilized to achieve adaptive fusion.Ex-periments on multi-type simulated datasets demonstrate that the proposed approach consistently outperforms single-modality and tra-ditional methods across both 24-class and 5-class recognition tasks,achieving an accuracy of 96.43%for the latter.These results confirm the effectiveness and robustness of the method under narrowband radar conditions,and highlight its potential as a promising solution for multi-modal feature fusion and recognition of complex space targets.关键词
窄带雷达/高动态目标识别/时序卷积网络/多模态融合/交叉注意力Key words
narrowband radar/high-dynamic target recognition/temporal convolutional network(TCN)/multi-modal fusion/cross-attention分类
信息技术与安全科学引用本文复制引用
穆思远,姜媛媛,庞钰宁,周嘉琪,李东瀛,魏飞鸣,郁文贤..基于时序特征与注意力机制的多模态高动态场景空间目标识别方法[J].现代雷达,2025,47(11):15-21,7.
