基于角度引导注意力的气动目标宽带PD识别方法

doi:10.12305/j.issn.1001-506X.2025.03.13

摘要/Abstract

摘要：

逆合成孔径雷达(inverse synthetic aperture radar, ISAR)图像是雷达自动目标识别的重要手段，获得高分辨率的ISAR图像需要雷达长时间照射，在实际工程应用中存在较大的限制。相比之下，宽带脉冲多普勒(pulse Doppler, PD)图像通过短脉冲积累成像，能够有效节约雷达资源。本文以不同入射视线角下图像中调制现象差异为出发点，设计一种角度引导注意力的卷积神经网络，旨在实现有限资源下更高的识别性能。首先，通过混合注意力残差模块，使网络聚焦于图像空域的差异，从而有效提升目标精细化特征的表征能力。然后，设计角度引导注意力模块，通过角度编码将入射视线角信息嵌入网络，实现目标特征表示与姿态的关联耦合，进一步提升识别准确率。最后，通过3类飞机的实测宽带PD图像进行分类识别，验证所设计网络的有效性。

关键词: 宽带脉冲多普勒图像, 目标识别, 注意力机制, 角度引导

Abstract:

Inverse synthetic aperture radar (ISAR) images are vital tools for radar automatic target recognition. However, acquiring high-resolution ISAR images requires prolonged radar illumination, posing significant limitations in practical engineering applications. In contrast, wideband pulse Doppler (PD) images can effectively save radar resources by accumulating images with short pulses. In this paper, the difference of modulation phenomena in images under different incident view angles is taken as the starting point, aiming to achieve higher recognition performance under limited resources, a convolutional neural network (CNN) with angle-guided attention is proposed. Firstly, through a hybrid attention residual module, the network focuses on the differences in the image space, effectively enhancing the representation capability of fine-grained target features. Subsequently, an angle-guided attention module is designed, embedding incident view angle information into the network through angle encoding, establishing a coupling between target feature representation and pose, further improving recognition accuracy. Finally, the effectiveness of the proposed network is validated through the classification and recognition of measured wideband PD images of three types of aircraft.

Key words: wideband pulse Doppler (PD) image, target recognition, attention mechanism, angle-guided

中图分类号:

TN95

李家宽, 冯博, 刘红亮, 叶春茂, 余继周. 基于角度引导注意力的气动目标宽带PD识别方法[J]. 系统工程与电子技术, 2025, 47(3): 807-816.

Jiakuan LI, Bo FENG, Hongliang LIU, Chunmao YE, Jizhou YU. Angle-guided attention-based wideband PD recognition method for aerodynamic targets[J]. Systems Engineering and Electronics, 2025, 47(3): 807-816.

图/表 18

图1

图2

图3

图4

图5

图6

表1

图7

图8

图9

图10

图11

表2

表3

表4

表5

图12

图13

参考文献 33

1	SAIDI M N, DAOUDI K, KHENCHAF A, et al. Automatic target recognition of aircraft models based on ISAR images[C]//Proc. of the IEEE International Geoscience and Remote Sensing Symposium, 2009.
2	尚士泽, 李明, 侯颖妮. 基于对称Gabor小波滤波的舰船目标识别方法[J]. 现代雷达, 2017, 39 (10): 43- 48.
	SHANG S Z , LI M , HOU Y N . Ship target recognition method based on symmetric Gabor wavelet filtering[J]. Modern Radar, 2017, 39 (10): 43- 48.
3	WANG N, MIAO Y Z, TONG C, et al. Space target recognition of ISAR image based on the PCA method[C]//Proc. of the International Conference on Image, Signal Processing, and Pattern Recognition, 2023: 963-970.
4	CEXUS J C, TOUMI A, RIAHI M. Target recognition from ISAR image using polar mapping and shape matrix[C]//Proc. of the 5th International Conference on Advanced Technologies for Signal and Image Processing, 2020.
5	JIN X F, SU F L. Aircraft recognition using ISAR image based on quadrangle-points affine transform[C]//Proc. of the 15th International Congress on Image and Signal Processing, 2022.
6	LI X Y , HE Y , JING X J . A survey of deep learning-based human activity recognition in radar[J]. Remote Sensing, 2019, 11 (9): 1068. doi: 10.3390/rs11091068
7	GURBUZ S Z , AMIN M G . Radar-based human-motion recognition with deep learning: Promising applications for indoor monitoring[J]. IEEE Signal Processing Magazine, 2019, 36 (4): 16- 28. doi: 10.1109/MSP.2018.2890128
8	YANG Y, GAO X Z, SHEN Q M. Learning embedding adaptation for ISAR image recognition with few samples[C]//Proc. of the 2nd Information Communication Technologies Conference, 2021: 86-89.
9	曾磊. 基于深度卷积神经网络的ISAR图像小样本识别方法研究[D]. 西安: 西安电子科技大学, 2021.
	ZENG L. Research on small sample recognition of ISAR images based on deep convolutional neural networks[D]. Xi'an: Xidian University, 2021.
10	LU W , ZHANG Y S , YIN C B , et al. A deformation robust ISAR image satellite target recognition method based on PT-CCNN[J]. IEEE Access, 2021, 9, 23432- 23453. doi: 10.1109/ACCESS.2021.3056671
11	BAI X R , ZHOU X N , ZHANG F , et al. Robust Pol-ISAR target recognition based on ST-MC-DCNN[J]. IEEE Trans. on Geoscience and Remote Sensing, 2019, 57 (12): 9912- 9927. doi: 10.1109/TGRS.2019.2930112
12	ZHOU X N, BAI X R. Robust ISAR target recognition based on IC-STNs[C]//Proc. of the 6th Asia-Pacific Conference on Synthetic Aperture Radar, 2019.
13	ZHOU X N , BAI X R , WANG L , et al. Robust ISAR target recognition based on ADRISAR-Net[J]. IEEE Trans. on Aerospace and Electronic Systems, 2022, 58 (6): 5494- 5505. doi: 10.1109/TAES.2022.3174826
14	任永梅, 杨杰, 郭志强, 等. 基于多尺度卷积神经网络的自适应熵加权决策融合船舶图像分类方法[J]. 电子与信息学报, 2021, 43 (5): 1424- 1431.
	REN Y M , YANG J , GUO Z Q , et al. Adaptive entropy-weighted decision fusion ship image classification method based on multiscale convolutional neural network[J]. Journal of Electronics & Information Technology, 2021, 43 (5): 1424- 1431.
15	LIN W H, GAO X Z, SHEN Q M. A novel recognition method of ISAR image fusion based on Botnet and Embracenet[C]//Proc. of the IEEE 6th International Conference on Signal and Image Processing, 2021: 841-845.
16	LI X H , RAN J H , WEN Y B , et al. MVFRnet: a novel high-accuracy network for ISAR air-target recognition via multi-view fusion[J]. Remote Sensing, 2023, 15 (12): 3052. doi: 10.3390/rs15123052
17	HAN C L, LU Y, YANG D, et al. Multi-source collaborative target classification based on ISAR and infrared image[C]//Proc. of the 6th International Conference on Image, Vision and Computing, 2021: 149-153.
18	WANG J, HUO C Y, WANG F, et al. Analysis of space target fusion reconstruction based on optical image and ISAR image[C]//Proc. of the 8th Symposium on Novel Photoelectronic Detection Technology and Applications, 2022: 520-525.
19	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]// Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7132-7141.
20	WOO S, PARK J, LEE J Y, et al. Cbam: convolutional block attention module[C]//Proc. of the European Conference on Computer Vision, 2018: 3-19.
21	LI X, HU X L, YANG J. Spatial group-wise enhance: improving semantic feature learning in convolutional networks[EB/OL]. [2023-10-21]. https://arxiv.org/abs/1905.09646.
22	张一凡, 张双辉, 刘永祥, 等. 基于注意力机制的堆叠LSTM网络雷达HRRP序列目标识别方法[J]. 系统工程与电子技术, 2021, 43 (10): 2775- 2781. doi: 10.12305/j.issn.1001-506X.2021.10.09
	ZHANG Y F , ZHANG S H , LIU Y X , et al. Radar HRRP sequence target recognition method of attention mechanism based stacked LSTM network[J]. Systems Engineering and Electronics, 2021, 43 (10): 2775- 2781. doi: 10.12305/j.issn.1001-506X.2021.10.09
23	曹鹏宇, 杨承志, 陈泽盛, 等. 基于深度残差收缩注意力网络的雷达信号识别方法[J]. 系统工程与电子技术, 2023, 45 (3): 717- 725. doi: 10.12305/j.issn.1001-506X.2023.03.12
	CAO P Y , YANG C Z , CHEN Z S , et al. Radar signal recognition method based on deep residual shrinkage attention network[J]. Systems Engineering and Electronics, 2023, 45 (3): 717- 725. doi: 10.12305/j.issn.1001-506X.2023.03.12
24	赵庆媛, 赵志强, 叶春茂, 等. 基于自注意力的双波段预警雷达微动融合识别[J]. 系统工程与电子技术, 2023, 45 (3): 708- 716. doi: 10.12305/j.issn.1001-506X.2023.03.11
	ZHAO Q Y , ZHAO Z Q , YE C M , et al. Micro-motion fusion recognition of double-band early warning radar based on self-attention mechanism[J]. Systems Engineering and Electronics, 2023, 45 (3): 708- 716. doi: 10.12305/j.issn.1001-506X.2023.03.11
25	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
26	CHEN V C. Radar signatures of rotor blades[C]//Proc. of the Conference on Wavelet Applications, 2001: 63-70.
27	CHEN V C , LI F , HO S S , et al. Micro-Doppler effect in radar: phenomenon model and simulation study[J]. IEEE Trans. on Aerospace and Electronic Systems, 2006, 42 (1): 2- 21.
28	BELL M R , GRUBBS R A . JEM modeling and measurement for radar target identification[J]. IEEE Trans. on Aerospace and Electronic Systems, 1993, 29 (1): 73- 87. doi: 10.1109/7.249114
29	MARTIN J, MULGREW B. Analysis of the theoretical radar return signal form aircraft propeller blades[C]//Proc. of the IEEE International Conference on Radar, 1990: 569-572.
30	郭帅, 陈婷, 王鹏辉, 等. 基于角度引导Transformer融合网络的多站协同目标识别方法[J]. 雷达学报, 2023, 12 (3): 516- 528.
	GUO S , CHEN T , WANG P H , et al. Multi-station cooperative target recognition method based on angle-guided transformer fusion network[J]. Journal of Radars, 2023, 12 (3): 516- 528.
31	SUN Y S , WANG Y H , LIU H W , et al. SAR target recognition with limited training data based on angular rotation generative network[J]. IEEE Geoscience and Remote Sensing Letters, 2019, 17 (11): 1928- 1932.
32	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[EB/OL]. [2023-10-21]. https://arxiv.org/abs/1706.03762.
33	KIRANYAZ S , AVCI O , ABDELJABER O , et al. 1D convolutional neural networks and applications: a survey[J]. Mechanical Systems and Signal Processing, 2021, 151, 107398. doi: 10.1016/j.ymssp.2020.107398

网络结构	参数设置	输出尺寸
阶段0	$\left[\begin{array}{c}7 \times 7 @ 64 \\\text { 最大池化 }\end{array}\right] \times 1$	16×64@64
阶段1	$\left[\begin{array}{c}3 \times 3 @ 64, 2 \\\text { 混合注意力 }\end{array}\right] \times 3$	16×64@64
阶段2	$\left[\begin{array}{c}3 \times 3 @ 128, 2 \\\text { 混合注意力 }\end{array}\right] \times 4$	8×32@128
角度引导注意力模块	$\left[\begin{array}{c}\text { 角度编码, MLP } \\3 \times 3 @ 64 \\3 \times 3 @ 128\end{array}\right] \times 1$	8×32@128
阶段3	$\left[\begin{array}{c}3 \times 3 @ 256, 2 \\\text { 混合注意力 }\end{array}\right] \times 6$6	4×16@256
阶段4	$\left[\begin{array}{c}3 \times 3 @ 512, 2 \\\text { 混合注意力 }\end{array}\right] \times 3$	2×8@512
全连接层	平均池化, softmax	1×1

数据集	目标类别	数量	覆盖角度/(°)
训练集	安26	4 830	12.1~165.7
	奖状	4 830	14.6~165.3
	雅克42	4 830	114.3~174.3
测试集	安26	4 528	22.3~89.1
	安26	4 528	132.2~167.2
	奖状	4 830	31.1~138.2
	雅克42	2 722	161.3~177.8

网络输入	识别准确率/%			平均识别率/%
网络输入	安26	奖状	雅克42	平均识别率/%
高分辨距离像	92.1	84.8	98.2	91.70
宽带PD图像	95.3	90.5	98.2	94.67

网络模型	识别准确率/%			平均识别率/%	参数量(×10⁶)	计算量(浮点运算次数×10⁹/s)
网络模型	安26	奖状	雅克42	平均识别率/%	参数量(×10⁶)	计算量(浮点运算次数×10⁹/s)
ResNet34	95.3	90.5	98.2	94.67	21.28	38.43
ResNet34+通道注意力模块	92.7	94.7	97.6	95.00	21.44	38.46
ResNet34+空间注意力模块	97.3	93.9	93.7	94.96	21.28	38.44
ResNet34+角度引导注意力模块	96.8	95.1	100.0	97.30	21.31	38.59
ResNet34+混合注意力残差模块	98.9	94.6	98.8	97.43	21.44	38.46
ResNet34+角度引导注意力模块+ 混合注意力残差模块	96.6	98.1	100.0	98.23	21.46	38.61

角度引导注意力模块位置	平均识别率/%	参数量(×10⁶)
阶段1末端	97.00	21.29
阶段2末端	97.30	21.32
阶段3末端	96.47	21.42
阶段4末端	96.37	21.84