基于角度引导Transformer融合网络的多站协同目标识别方法

郭帅; 陈婷; 王鹏辉; 丁军; 严俊坤; 王英华; 刘宏伟

doi:10.12000/JR23014

基于角度引导Transformer融合网络的多站协同目标识别方法

doi: 10.12000/JR23014

西安电子科技大学雷达信号处理全国重点实验室西安 710071

基金项目: 国家自然科学基金(62192714, 61701379)，雷达信号处理国家级重点实验室支持计划项目(KGJ202204)，中央高校基本科研业务费(QTZX22160)，中国航天科技集团公司第八研究院产学研合作基金资助项目(SAST2021-011)，陕西省天线与控制技术重点实验室开放基金

详细信息

作者简介:
郭　帅，博士生，主要研究方向为雷达目标识别

陈　婷，博士生，主要研究方向为雷达目标识别

王鹏辉，教授，博士生导师，主要研究方向为雷达目标识别、机器学习理论与应用研究等

丁　军，副研究员，硕士生导师，主要研究方向为雷达目标识别、数据工程

严俊坤，教授，博士生导师，主要研究方向为网络化雷达协同探测、雷达智能化探测等

王英华，教授，博士生导师，主要研究方向为SAR图像目标检测与识别、极化SAR图像处理与解译等

刘宏伟，教授，博士生导师，主要研究方向为雷达目标识别、认知雷达、网络化协同探测、雷达智能化探测等

通讯作者:
王鹏辉 wangpenghui@mail.xidian.edu.cn

刘宏伟 hwliu@xidian.edu.cn

责任主编：冯存前 Corresponding Editor: FENG Cunqian
中图分类号: TN959
计量
- 文章访问数: 941
- HTML全文浏览量: 618
- PDF下载量: 258
- 被引次数: 0
出版历程
- 收稿日期: 2023-02-03
- 修回日期: 2023-04-02
- 网络出版日期: 2023-04-23
- 刊出日期: 2023-06-28

Multistation Cooperative Radar Target Recognition Based on an Angle-guided Transformer Fusion Network

National Key Laboratory of Radar Signal Processing, Xidian University, Xi’an 710071, China

Funds: The National Natural Science Foundation of China (62192714, 61701379), The Stabilization Support of National Radar Signal Processing Laboratory (KGJ202204), The Fundamental Research Funds for the Central Universities (QTZX22160), Industry-University-Research Cooperation of the 8th Research Institute of China Aerospace Science and Technology Corporation (SAST2021-011), Open Fund Shaanxi Key Laboratory of Antenna and Control Technology

More Information

Corresponding author: WANG Penghui, wangpenghui@mail.xidian.edu.cn; LIU Hongwei, hwliu@xidian.edu.cn

摘要

摘要: 多站协同雷达目标识别旨在利用多站信息的互补性提升识别性能。传统多站协同目标识别方法未直接考虑站间数据差异问题，且通常采用相对简单的融合策略，难以取得准确、稳健的识别性能。该文针对多站协同雷达高分辨距离像(HRRP)目标识别问题，提出了一种基于角度引导的Transformer融合网络。该网络以Transformer作为特征提取主体结构，提取单站HRRP的局部和全局特征。并在此基础上设计了3个新的辅助模块促进多站特征融合学习，角度引导模块、前级特征交互模块以及深层注意力特征融合模块。首先，角度引导模块使用目标方位角度对站间数据差异进行建模，强化了所提特征与多站视角的对应关系，提升了特征稳健性与一致性。其次，前级特征交互模块和深层注意力特征融合模块相结合的融合策略，实现了对各站特征的多阶段层次化融合。最后，基于实测数据模拟多站场景进行协同识别实验，结果表明所提方法能够有效地提升多站协同时的目标识别性能。
- 多站协同雷达目标识别 /
- 高分辨距离像(HRRP) /
- 角度引导 /
- 注意力特征融合 /
- Transformer融合网络
Abstract: Multistation cooperative radar target recognition aims to enhance recognition performance by utilizing the complementarity between multistation information. Conventional multistation cooperative target recognition methods do not explicitly consider the issue of interstation data differences and typically adopt relatively simple fusion strategies, which makes it difficult to obtain accurate and robust recognition performance. In this study, we propose an angle-guided transformer fusion network for multistation radar High-Resolution Range Profile (HRRP) target recognition. The extraction of the local and global features of the single-station HRRP is conducted via feature extraction, which employs a transformer as its main structure. Furthermore, three new auxiliary modules are created to facilitate fusion learning: the angle-guided module, the prefeature interaction module, and the deep attention feature fusion module. First, the angle guidance module enhances the robustness and consistency of features via modeling data differences between multiple stations and reinforces individual features associated with the observation perspective. Second, the fusion approach is optimized, and the multilevel hierarchical fusion of multistation features is achieved by combining the prefeature interaction module and the deep attention feature fusion module. Finally, the experiments are conducted on the basis of the simulated multistation scenarios with measured data, and the outcomes demonstrate that our approach can effectively enhance the performance of target recognition in multistation coordination.
- Multistation cooperative radar target recognition /
- High-Resolution Range Profile (HRRP) /
- Angle-guided /
- Attention feature fusion /
- Transformer fusion network

HTML全文

图 1 融合网络结构示意图

Figure 1. Schematic of different fusion network structures

下载: 全尺寸图片幻灯片

图 2 角度引导Transformer融合网络结构

Figure 2. Angle guided Transformer fusion network framework

下载: 全尺寸图片幻灯片

图 3 切分映射及位置编码

Figure 3. Patch embedding and positional embedding

下载: 全尺寸图片幻灯片

图 4 Transformer模块

Figure 4. The Transformer module

下载: 全尺寸图片幻灯片

图 5 角度引导模块

Figure 5. The angle guided module

下载: 全尺寸图片幻灯片

图 6 前级特征交互模块

Figure 6. The pre-feature interaction module

下载: 全尺寸图片幻灯片

图 7 深层注意力特征融合模块

Figure 7. The deep attention feature fusion module

下载: 全尺寸图片幻灯片

图 8 置换不变Transformer与Transformer特征提取层对比图

Figure 8. Comparison of permutation invariant Transformer in feature fusion and Transformer in feature extraction

下载: 全尺寸图片幻灯片

图 9 模拟多站场景的目标HRRP

Figure 9. Target HRRP examples for simulating multistation scenarios

下载: 全尺寸图片幻灯片

图 10 测试集识别率混淆矩阵(%)

Figure 10. Confusion matrix of the recognition accuracy in test set (%)

下载: 全尺寸图片幻灯片

图 11 测试集数据与本文方法所提特征的二维t-SNE可视化

Figure 11. Visualization of test data and feature via two-dimensional t-SNE

下载: 全尺寸图片幻灯片

图 12 识别率和计算量随着HRRP子序列个数变化的曲线图

Figure 12. Accuracy and calculation amount changing with the number of HRRP subsequences

下载: 全尺寸图片幻灯片

表 1 雷达参数

Table 1. Parameters of radar

参数	数值
信号带宽	400 MHz
距离分辨率	0.375 m

下载: 导出CSV

表 2 目标物理参数

Table 2. Parameters of targets

飞机型号	机身长度(m)	翼展宽度(m)	机高(m)
A320	37.57	34.10	11.76
A321	44.51	34.09	11.76
A330-2	58.80	60.30	17.40
A330-3	63.60	60.30	16.85
A350	66.80	64.75	17.05

下载: 导出CSV

表 3 数据集样本分布

Table 3. Dataset samples distribution

飞机型号	训练样本数	测试样本数
A320	2636	2594
A321	2482	2398
A330-2	2556	2569
A330-3	2785	2572
A350	2890	2181

下载: 导出CSV

表 4 实验参数配置

Table 4. Experimental parameters configuration

实验配置	参数
训练轮次	200
批量大小	64
初始学习率	1E–3
优化器	AdamW
丢弃率	0.1
HRRP子序列的个数N	8
子序列编码维度D	128
Transformer模块数	3
注意力头数	4
角度编码全连接层输出维度	(128, 1152)
特征交互主支路权重	0.6
特征交互其余支路权重	0.2
深层注意力融合模块数	1
损失函数	Cross Entropy Loss

下载: 导出CSV

表 5 实验结果

Table 5. Experimental results

方法	融合策略	识别率(%)	参数量(M)	计算量(GFLOPs)
CNN单站	雷达1	81.56	4.19	0.30
	雷达2	87.27	4.19	0.30
	雷达3	90.71	4.19	0.30
CNN多站	数据融合	86.35	4.19	0.30
	特征融合	90.08	12.59	1.76
	决策融合	90.96	12.59	1.76
Transformer单站	雷达1	87.12	0.89	0.46
	雷达2	88.03	0.89	0.46
	雷达3	93.21	0.89	0.46
Transformer多站	特征融合	93.60	2.51	1.38
本文方法	方位角度引导+前级特征交互+深层注意力特征融合	96.90	3.39	1.60

下载: 导出CSV

表 6 消融实验结果

Table 6. Results of ablation experiment

方法	角度引导	前级特征交互	深层注意力特征融合	识别率(%)
Transformer 多站特征融合	–	–	–	93.60
	√	–	–	94.50(+0.90)
	–	√	–	93.20(–0.40)
	–	–	√	93.70(+0.10)
	√	√	–	93.77(+0.17)
	–	√	√	94.47(+0.87)
	√	–	√	95.68(+2.08)
本文方法	√	√	√	96.90(+3.30)

下载: 导出CSV

参考文献(27)

[1]	DING Beicheng and CHEN Penghui. HRRP feature extraction and recognition method of radar ground target using convolutional neural network[C]. 2019 International Conference on Electromagnetics in Advanced Applications (ICEAA), Granada, Spain, 2019: 658–661.
[2]	WAN Jinwei, CHEN Bo, YUAN Yijun, et al. Radar HRRP recognition using attentional CNN with multi-resolution spectrograms[C]. 2019 International Radar Conference (RADAR), Toulon, France, 2019: 1–4.
[3]	WANG Penghui, DU Lan, PAN Mian, et al. Radar HRRP target recognition based on linear dynamic model[C]. 2011 IEEE CIE International Conference on Radar, Chengdu, China, 2011: 662–665.
[4]	YANG Xiuzhu, ZHANG Xinyue, DING Yi, et al. Indoor activity and vital sign monitoring for moving people with multiple radar data fusion[J]. Remote Sensing, 2021, 13(18): 3791. doi: 10.3390/rs13183791
[5]	YANG Jiachen, ZHANG Zhou, MAO Wei, et al. IoT-based critical infrastructure enabled radar information fusion[J]. Computers & Electrical Engineering, 2022, 98: 107723. doi: 10.1016/j.compeleceng.2022.107723
[6]	WU Hao, DAI Dahai, JI Penghui, et al. High-resolution range profile recognition method of vehicle targets based on accelerated T-SNE with multi-polarization fusion[C]. 2021 IEEE 6th International Conference on Signal and Image Processing (ICSIP), Nanjing, China, 2021: 72–76.
[7]	LIU Chang, ANTYPENKO R, SUSHKO I, et al. Marine distributed radar signal identification and classification based on deep learning[J]. Traitement du Signal, 2021, 38(5): 1541–1548. doi: 10.18280/ts.380531
[8]	HUANG Yuchen, LI Wei, DOU Zhiyang, et al. Activity recognition based on millimeter-wave radar by fusing point cloud and range-Doppler information[J]. Signals, 2022, 3(2): 266–283. doi: 10.3390/signals3020017
[9]	AGUILETA A A, BRENA R F, MAYORA O, et al. Multi-sensor fusion for activity recognition—A survey[J]. Sensors, 2019, 19(17): 3808. doi: 10.3390/s19173808
[10]	周宁宁, 朱士涛, 年毅恒, 等. 一种基于多模态OAM波束的目标特征智能识别方法[J]. 雷达学报, 2021, 10(5): 760–772. doi: 10.12000/JR21056 ZHOU Ningning, ZHU Shitao, NIAN Yiheng, et al. An intelligent target feature recognition method based on multi-mode OAM beams[J]. Journal of Radars, 2021, 10(5): 760–772. doi: 10.12000/JR21056
[11]	邓冬虎, 张群, 罗迎, 等. 双基地ISAR系统中分辨率分析及微多普勒效应研究(英文)[J]. 雷达学报, 2013, 2(2): 152–167. doi: 10.3724/SP.J.1300.2013.13039 DENG Donghu, ZHANG Qun, LUO Ying, et al. Resolution and micro-Doppler effect in Bi-ISAR system (in English)[J]. Journal of Radars, 2013, 2(2): 152–167. doi: 10.3724/SP.J.1300.2013.13039
[12]	冯存前, 李靖卿, 贺思三, 等. 组网雷达中弹道目标微动特征提取与识别综述[J]. 雷达学报, 2015, 4(6): 609–620. doi: 10.12000/JR15084 FENG Cunqian, LI Jingqing, HE Sisan, et al. Micro-Doppler feature extraction and recognition based on netted radar for ballistic targets[J]. Journal of Radars, 2015, 4(6): 609–620. doi: 10.12000/JR15084
[13]	章鹏飞, 李刚, 霍超颖, 等. 基于双雷达微动特征融合的无人机分类识别[J]. 雷达学报, 2018, 7(5): 557–564. doi: 10.12000/JR18061 ZHANG Pengfei, LI Gang, HUO Chaoying, et al. Classification of drones based on micro-Doppler radar signatures using dual radar sensors[J]. Journal of Radars, 2018, 7(5): 557–564. doi: 10.12000/JR18061
[14]	RYKUNOV M, DE GREEF E, KHALID H U R, et al. Multi-radar fusion for failure-tolerant vulnerable road users classification[C]. 2021 18th European Radar Conference (EuRAD), London, United Kingdom, 2022: 337–340.
[15]	SHU Haining and LIANG Qilian. Data fusion in a multi-target radar sensor network[C]. 2007 IEEE Radio and Wireless Symposium, Long Beach, USA, 2007: 129–132.
[16]	DU Lan, WANG Penghui, LIU Hongwei, et al. Bayesian spatiotemporal multitask learning for radar HRRP target recognition[J]. IEEE Transactions on Signal Processing, 2011, 59(7): 3182–3196. doi: 10.1109/TSP.2011.2141664
[17]	DEVLIN J, CHANG M W, LEE K, et al. BERT: Pre-training of deep bidirectional transformers for language understanding[C]. 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), Minneapolis, USA, 2019: 4171–4186.
[18]	DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16×16 words: Transformers for image recognition at scale[C]. 9th International Conference on Learning Representations, Austria, 2021.
[19]	WANG Qiang, LI Bei, XIAO Tong, et al. Learning deep transformer models for machine translation[C]. 57th Annual Meeting of the Association for Computational Linguistics, Florence, Italy, 2019: 1810–1822.
[20]	CAO Kaidi, RONG Yu, LI Cheng, et al. Pose-robust face recognition via deep residual equivariant mapping[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 5187–5196.
[21]	SUN Yuanshuang, WANG Yinghua, LIU Hongwei, et al. SAR target recognition with limited training data based on angular rotation generative network[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 17(11): 1928–1932. doi: 10.1109/LGRS.2019.2958379
[22]	LOSHCHILOV I and HUTTER F. Decoupled weight decay regularization[C]. 7th International Conference on Learning Representations, New Orleans, USA, 2019.
[23]	YUAN Lele. A time-frequency feature fusion algorithm based on neural network for HRRP[J]. Progress in Electromagnetics Research M, 2017, 55: 63–71. doi: 10.2528/PIERM16123002
[24]	QUAN Daying, TANG Zeyu, WANG Xiaofeng, et al. LPI radar signal recognition based on dual-Channel CNN and feature fusion[J]. Symmetry, 2022, 14(3): 570. doi: 10.3390/sym14030570
[25]	ZHU Lijun. Selection of multi-level deep features via spearman rank correlation for synthetic aperture radar target recognition using decision fusion[J]. IEEE Access, 2020, 8: 133914–133927. doi: 10.1109/ACCESS.2020.3010969
[26]	CHEN Wenchao, CHEN Bo, PENG Xiaojun, et al. Tensor RNN with Bayesian nonparametric mixture for radar HRRP modeling and target recognition[J]. IEEE Transactions on Signal Processing, 2021, 69: 1995–2009. doi: 10.1109/TSP.2021.3065847
[27]	GUO Dandan, CHEN Bo, CHEN Wenchao, et al. Variational temporal deep generative model for radar HRRP target recognition[J]. IEEE Transactions on Signal Processing, 2020, 68: 5795–5809. doi: 10.1109/TSP.2020.3027470