基于运动阵列微波成像与多尺度可变形卷积网络的引信目标识别方法

doi:10.12382/bgxb.2024.0282

摘要/Abstract

摘要：

针对传统调频连续波(Frequency Modulated Continuous Wave,FMCW)引信探测维度低、方位分辨能力弱导致目标识别能力不足的问题,提出基于运动阵列微波成像与多尺度可变形卷积网络(Multi-Scale Deformable Convolutional Networks,MSDCN)的引信目标识别方法。在充分分析引信运动过程中回波相位变化规律的基础上建立FMCW运动阵列天线模型,通过运动合成扩充引信天线虚拟阵元数,大幅度提升引信方位向分辨率,实现目标距离-方位的二维高分辨成像。同时,深入分析弹目交会过程中由于目标位置、姿态、距离等状态变化形成的图像多尺度特性,构建MSDCN目标识别模型,提高引信对复杂动态交会场景下目标成像多尺度特性的自适应识别能力。实验结果表明,该方法能够显著提高引信方位分辨能力,在不同目标场景下均取得较好的成像和识别效果,对典型目标多尺度像识别准确率达到94%,-6dB信噪比时目标识别准确率仍能达到88%。

关键词: 引信, 调频连续波, 运动阵列, 距离-方位二维像, 多尺度可变形卷积网络, 目标识别

Abstract:

In response to the challenge of inadequate target recognition capabilities due to the limited detection dimension and weak azimuth resolution of conventional frequency modulated continuous wave (FMCW) fuze,a fuze target recognition method based on motion array microwave imaging and multi-scale deformable convolutional networks (MSDCN) is proposed.A FMCW motion array antenna model is established thorough analysis of the thorough analysis of echo phase variation during the fuze motion.The virtual array elements of fuze antenna are expanded by motion synthesis to significantly enhance the azimuth resolution of the fuze,thus achieving the two-dimensional high-resolution imaging of target distance and azimuth.Simultaneously,a MSDCN target recognition model is constructed by delving into the multi-scale characteristics of the images formed due to the variations in target position,attitude,distance,and other states during the fuze-target encounter process.This enhances the adaptive recognition capability of fuze for the multi-scale characteristics of target imaging in complex dynamic encounter scenarios.The experimental results demonstrate that the proposed method significantly enhances the azimuth resolution of fuze.It achieves satisfactory imaging and recognition results in various target scenarios.The accuracy of multi-scale image recognition for typical targets reaches 94%,and even at -6dB signal-to-noise ratio,the target recognition accuracy remains at 88%.

Key words: fuze, frequency-modulated continuous wave, motion array antenna, range-azimuth two-dimensional image, multi-scale deformable convolutional network, target recognition

中图分类号:

TJ434.1

韩燕文, 闫晓鹏, 高晓峰, 伊光华, 代健. 基于运动阵列微波成像与多尺度可变形卷积网络的引信目标识别方法[J]. 兵工学报, 2025, 46(3): 240282-.

HAN Yanwen, YAN Xiaopeng, GAO Xiaofeng, YI Guanghua, DAI Jian. Fuze Target Recognition Method Based on Motion Array Microwave Imaging and Multi-scale Deformable Convolutional Network[J]. Acta Armamentarii, 2025, 46(3): 240282-.

图/表 17

图1 运动阵列引信成像示意图

Fig.1 Schematic diagram of motion array fuze imaging

图2 阵列运动方案示意图

Fig.2 Array motion scheme

图3 阵列信号距离、方位信息获取流程图

Fig.3 Flow chart of obtaining the distance and azimuth information of array signal

表1 距离-方位角估计的仿真参数

Table 1 Simulation parameters of range-azimuth estimation

载频/ GHz	带宽/ GHz	采样频率/ MHz	采样点数	调制周期/μs	调制斜率/ (MHz·μs^-1)
77	3	10	512	50	80

图4 不同阵列数的成像仿真图

Fig.4 Imaging simulation of different array numbers

图5 MSDCN网络模型

Fig.5 MSDCN network model

表2 MSDCN网络算法流程

Table 2 MSDCN network algorithm flow

算法:多尺度特征融合可变形卷积网络(MSDCN)

输入:距离-方位角灰度图I (200,200)。

1.卷积核conv0(8@3×3),RELU激活层和可变形ROI池化(2×2),输出:特征图0(8×100×100);

2.输入:特征图0;卷积核conv1(16@3×3),RELU和可变形ROI池化(2×2),输出:特征图1(16×50×50);

3.输入:特征图1;卷积核conv2(32@3×3),RELU和可变形ROI池化(2×2),输出:特征图2(32×25×25);

4.最大池化方法将输入I、特征图0、特征图1和特征图2进行特征融合;

5.全连接层输出FC layer 0、FC layer 1、FC layer 2,加权融合后输入Softmax激活层得到预测概率。

输出:成像灰度图的目标标签。

图6 毫米波FMCW传感器数据采集的实验场景和设备图

Fig.6 Experimental scene and equipment of millimeter wave FMCW sensor data acquisition

表3 毫米波FMCW传感器实验参数

Table 3 Experimental parameters of millimeter wave FMCW sensor

载频/ GHz	带宽/ GHz	调制斜率/ (MHz·μs^-1)	采样频率/ MHz	采样点数	滑轨速度/ (m·s^-1)
77	3.398	68	11.6	512	0.438

图7 不同阵列数的成像结果

Fig.7 Imaging results of different array numbers

图8 不同目标成像和实验场景

Fig.8 Experimental scenes and imaging of different targets

图9 不同角度下的FMCW引信成像图

Fig.9 Imaging of FMCW fuze at different angles

图10 MSDCN模型的训练和验证过程

Fig.10 Training and verification process of MSDCN model

图11 MSDCN模型的混淆矩阵结果

Fig.11 Confusion matrix results of MSDCN model

图12 不同SNR下的FMCW引信成像图

Fig.12 Imaging of FMCW fuzes under different SNRs

图13 不同模型对不同SNR下的毫米波FMCW引信成像图的识别精度曲线图

Fig.13 Recognition accuracies of millimeter wave FMCW fuze images with different SNRs by different models

表4 不同SNR下识别准确率

Table 4 Recognition accuracy under different SNRs

SNR /dB	MSDCN	DCN	U-Net	ResNet-18	Dense Net-21
-14	0.19	0.18	0.19	0.19	0.2
-12	0.21	0.22	0.23	0.22	0.25
-10	0.35	0.29	0.31	0.3	0.26
-8	0.65	0.49	0.62	0.64	0.65
-6	0.88	0.8	0.82	0.87	0.85
-4	0.92	0.85	0.89	0.9	0.9
-2	0.93	0.9	0.92	0.93	0.91
0	0.94	0.92	0.93	0.93	0.93
2	0.95	0.94	0.94	0.93	0.95
4	0.95	0.93	0.94	0.94	0.95

参考文献 23

[1]	赵惠昌. 无线电引信设计原理与方法[M]. 北京: 国防工业出版社, 2012:213-236.
	ZHAO H C. Design principle and method of radio fuze[M]. Beijing: National Defense Industry Press, 2012:213-236. (in Chinese)
[2]	栗苹, 郝新红, 闫晓鹏, 等. 无线电引信抗干扰理论[M]. 北京: 北京理工大学出版社, 2019.
	LI P, HAO X H, YAN X P, et al. Anti-jamming theory of radio fuze[M]. Beijing: Beijing Institute of Technology Press, 2019. (in Chinese)
[3]	ZHANG H Y, LI P, ZHANG G W. Orbital angular momentum radio-wave-based 2-D target angle estimation with mutual coupling[J]. IEEE Sensors Journal, 2023, 23(6):6169-6177.
[4]	CHENG C, GAO M, ZHOU X D. High-resolution forward-looking imaging algorithm for missile-borne detectors[J]. Journal of Systems Engineering and Electronics, 2019, 30(3):456-466. doi: 10.21629/JSEE.2019.03.03
[5]	FELIPE A D Q, MARWAN Y, GERHAND K, et al. Multichannel staggered SAR azimuth processing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(5):2772-2788.
[6]	CHEN H M, LI M, WANG Z Y, et al. Synthetic bandwidth azimuth modulation imaging radar for airborne single-channel forward-looking imaging[J]. Signal Processing, 2017, 130:86-92.
[7]	QIN G, ZHANG Y D, AMIN M G. DOA estimation exploiting moving dilated nested arrays[J]. IEEE Signal Processing Letters, 2019, 26(3):490-494. doi: 10.1109/LSP.2019.2894467
[8]	王新伟, 闫晓鹏, 郝新红, 等. 基于合成虚拟阵列的脉冲多普勒引信高分辨DOA估计方法[J]. 兵工学报, 2024, 45(1):97-104. doi: 10.12382/bgxb.2022.0433
	WANG X W, YAN X P, HAO X H, et al. High resolution DOA estimation method for pulsed Doppler fuze based on synthetic virtual array[J]. Acta Armamentarii, 2019, 45(1):97-104. (in Chinese)
[9]	YANG Y L, JIA F D, JIANG G J, et al. Underdetermined DOA estimation for moving array with reduced mutual coupling in unknown nonuniform noise environment[J]. AEU-International Journal of Electronics and Communications, 2022, 148:154-158.
[10]	LI X L, ZHANG W J. DOA estimation using virtual ESPRIT with successive baselines and coprime baselines[J]. Circuits,Systems,and Signal Processing, 2021, 40:2065-2075.
[11]	YANG Z, CHEN X Y, WU X M. A robust and statistically efficient maximum-likelihood method for DOA estimation using sparse linear arrays[J]. IEEE Transactions on Aerospace and Electronic Systems, 2023, 59(5):6798-6812.
[12]	陈志鹏, 李豪杰, 严秉谦, 等. 基于数据装定的坦克炮毫米波近炸引信炸高控制方法[J]. 兵工学报, 2024, 45(6):2034-2043. doi: 10.12382/bgxb.2023.0028
	CHEN Z P, LI H J, YAN B Q, et al. Control method of millimeter wave proximity fuze blast height for tank gun based on data loading[J]. Acta Armamentarii, 2024, 45(6):2034-2043. (in Chinese)
[13]	刘盛启, 张会强, 滕书华, 等. 基于联合动态稀疏表示的开集距离像目标识别方法[J]. 电子与信息学报, 2023, 45(11):4101-4109.
	LIU S Q, ZHANG H Q, TENG S H, et al. Open set range image target recognition method based on joint dynamic sparse representation[J]. Journal of Electronics and Information, 2023, 45(11):4101-4109. (in Chinese)
[14]	DING C W, ZHANG L, CHEN H Y, et al. Human motion recognition with spatial-temporal-convLSTM network using dynamic range-Doppler frames based on portable FMCW radar[J]. IEEE Transactions on Microwave Theory and Techniques, 2022, 70(11):5029-5038.
[15]	KIM B K, KANG H S, LEE S, et al. Improved drone classification using polarimetric merged-Doppler images[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 18(11):1946-1950.
[16]	熊光明, 罗震, 孙冬, 等. 基于红外相机和毫米波雷达融合的烟雾遮挡无人驾驶车辆目标检测与跟踪[J]. 兵工学报, 2024, 45(3):893-906. doi: 10.12382/bgxb.2022.0602
	XIONG G M, LUO Z, SUN D, et al. Detection and tracking of smoke-obstructed unmanned vehicle targets based on the fusion of infrared cameras and millimeter-wave radars[J]. Acta Armamentarii, 2024, 45(3):893-906. (in Chinese)
[17]	CHEN H S, HE Y, ZHANG L F, et al. A landslide extraction method of channel attention mechanism U-Net network based on Sentinel-2A remote sensing images[J]. International Journal of Digital Earth, 2023, 16(1):552-577.
[18]	ZHANG Y Q, PENG L X, Ma G L, et al. Dynamic gesture recognition model based on millimeter-wave radar with ResNet-18 and LSTM[J]. Frontiers in Neurorobotics, 2022, 16:903197.
[19]	ZHANG J L, LI Y X, SI Y, et al. A low-grade road extraction method using SDG-DenseNet based on the fusion of optical and SAR images at decision level[J]. Remote Sensing, 2022, 14(12):2870.
[20]	丁柏圆, 钟金荣, 马聪慧, 等. 基于属性散射中心模型的SAR目标重构可视化增强方法[J]. 雷达学报, 2013, 2(4):499-506.
	DING B Y, ZHONG J R, MA C H, et al. SAR target reconstruction visualization enhancement method based on attribute scattering center model[J]. Journal of Radar, 2013, 2(4):499-506. (in Chinese)
[21]	CHEN F, WU F, XU J, et al. Adaptive deformable convolutional network[J]. Neurocomputing, 2021, 453:853-864.
[22]	DAI J F, QI H Z, XIONG Y W, et al. Deformable convolutional networks[C]// Proceedings of the IEEE International Conference on Computer Vision. Washington,D.C.,US: IEEE, 2017:764-773.
[23]	YI G H, HAO X H, YAN X P, et al. Automatic modulation recognition of radiation source signals based on two-dimensional data matrix and improved residual neural network[J]. Defence Technology, 2023, 33:364-373.