面向轻量级交叉注意力卷积网络的SAR目标识别

doi:10.12305/j.issn.1001-506X.2025.09.08

摘要/Abstract

摘要：

针对合成孔径雷达（synthetic aperture radar, SAR）飞机目标识别网络消耗部署资源大的问题，提出一种基于轻量级交叉注意力卷积神经网络（lightweight cross-attention convolutional neural network, LCA-CNN）的SAR飞机目标识别方法。一方面，通过交叉注意力机制对目标进行特征提取，使得网络能够更高效地从样本中学习到关键的分类表征，提升飞机细粒度识别的准确率。另一方面，只利用卷积层和注意力模块，从而大幅降低网络的整体参数量。在SAR-AIRcraft-1.0数据集上的对比实验表明：与其他经典的深度学习SAR图像识别算法方法相比，所提方法在更少参数条件下可实现更高的平均识别准确率。

关键词: 合成孔径雷达, 雷达目标识别, 卷积神经网络, 交叉注意力机制

Abstract:

To address the issue of high computational resource consumption for synthetic aperture radar （SAR） aircraft target recognition network, a SAR aircraft target recognition method based on lightweight cross-attention convolutional neural network （LCA-CNN） is proposed. On one hand, the cross-attention mechanism is utilized to extract features of target, enabling the network to learn key classification representations efficiently from samples, thereby improving the accuracy of fine-grained aircraft recognition. On the other hand, the use of only convolutional layers and attention modules significantly reduces the overall number of network parameters. Comparative experiments on the SAR-AIRcraft-1.0 dataset with classical deep learning SAR image recognition algorithms demonstrate that the proposed method achieves higher average recognition accuracy with fewer parameters.

Key words: synthetic aperture radar (SAR), radar target recognition, convolutional neural network (CNN), cross attention mechanism

中图分类号:

TN 95

蒋明煜, 张顺生, 肖思瑶. 面向轻量级交叉注意力卷积网络的SAR目标识别[J]. 系统工程与电子技术, 2025, 47(9): 2853-2861.

Mingyu JIANG, Shunsheng ZHANG, Siyao XIAO. SAR target recognition based on lightweight cross-attention convolutional neural network[J]. Systems Engineering and Electronics, 2025, 47(9): 2853-2861.

图/表 12

图1

图2

图3

图4

图5

图6

表1

表2

表3

表4

图7

图8

参考文献 31

1	DU L, WANG Z C, WANG Y, et al. Survey of research progress on target detection and discrimination of single-channel SAR images for complex scenes[J]. Journal of Radars, 2020, 9 (1): 34- 54.
2	NOVAK L M, OWIRKA G J, BROWER W S. Performance of 10-and 20-target MSE classifiers[J]. IEEE Trans. on Aerospace and Electronic Systems, 2000, 36 (4): 1279- 1289. doi: 10.1109/7.892675
3	唐宁, 高勋章, 黎湘. 基于几何散列法的ISAR像自动目标识别[J]. 系统工程与电子技术, 2012, 34 (4): 692- 697. doi: 10.3969/j.issn.1001-506X.2012.04.10
	TANG N, GAO X Z, LI X. Automatic target recognition of ISAR images based on geometric hash[J]. Journal of Systems Engineering and Electronics, 2012, 34 (4): 692- 697. doi: 10.3969/j.issn.1001-506X.2012.04.10
4	张新征, 黄培康. 基于贝叶斯压缩感知的SAR目标识别[J]. 系统工程与电子技术, 2013, 35 (1): 40- 44. doi: 10.3969/j.issn.1001-506X.2013.01.07
	ZHANG X Z, HUANG P K. SAR ATR based on Bayesian compressive sensing[J]. Journal of Systems Engineering and Electronics, 2013, 35 (1): 40- 44. doi: 10.3969/j.issn.1001-506X.2013.01.07
5	MORGAN D A E. Deep convolutional neural networks for ATR from SAR imagery[C]//Proc. of the SPIE-Algorithms for Synthetic Aperture Radar Imagery XXII, 2015.
6	DING J, CHEN B, LIU H W, et al. Convolutional neural network with data augmentation for SAR target recognition[J]. IEEE Geoscience and Remote Sensing Letters, 2016, 13 (3): 364- 368.
7	CHEN S Z, WANG H P. SAR target recognition based on deep learning[C]//Proc. of the International Conference on Data Science and Advanced Analytics, 2014: 541−547.
8	王彩云, 吴钇达, 王佳宁, 等. 基于改进的CNN和数据增强的SAR目标识别[J]. 系统工程与电子技术, 2022, 44 (8): 2483- 2487. doi: 10.12305/j.issn.1001-506X.2022.08.12
	WANG C Y, WU Y D, WANG J N, et al. SAR target recognition based on improved CNN and data augmentation[J]. Systems Engineering and Electronics, 2022, 44 (8): 2483- 2487. doi: 10.12305/j.issn.1001-506X.2022.08.12
9	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60 (6): 84- 90. doi: 10.1145/3065386
10	SRIVASTAVA N, HINTON G, KRIZHEVSKY A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. The Journal of Machine Learning Research, 2014, 15 (1): 1929- 1958.
11	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. [2024-06-16]. https://arxiv.org/pdf/1409.1556.
12	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.
13	XIE S, GIRSHICK R, DOLLÁR P, et al. Aggregated residual transformations for deep neural networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 1492−1500.
14	SZEGEDY C, LIU W, JIA Y Q, et al. Going deeper with convolutions[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2015.
15	邵嘉琦, 曲长文, 李健伟. 卷积神经网络对SAR目标识别性能分析[J]. 雷达科学与技术, 2018, 16 (5): 525- 532. doi: 10.3969/j.issn.1672-2337.2018.05.010
	SHAO J Q, QU C W, LI J W. Performance analysis of convolutional neural networks for SAR target recognition[J]. Radar Science and Technology, 2018, 16 (5): 525- 532. doi: 10.3969/j.issn.1672-2337.2018.05.010
16	CHEN S Z, WANG H P, XU F, et al. Target classification using the deep convolutional networks for SAR images[J]. IEEE Trans. on Geoscience and Remote Sensing, 2016, 54 (8): 4806- 4817. doi: 10.1109/TGRS.2016.2551720
17	GUO M H, XU T X, LIU J J, et al. Attention mechanisms in computer vision: a survey[J]. Computational Visual Media, 2022, 8 (3): 331- 368. doi: 10.1007/s41095-022-0271-y
18	TANG T, CUI Y T, FENG R, et al. Vehicle target recognition in SAR images with complex scenes based on mixed attention mechanism[J]. Information, 2024, 15 (3): 159- 161. doi: 10.3390/info15030159
19	HUANG Z C, SHI M Y, CHEN Y R, et al. A self-attention armed lightweight optronic convolutional neural network for SAR target recognition[C]//Proc. of the IET International Radar Conference Proceedings, 2023.
20	LANG P, FU X J, FENG C, et al. LW-CMDANet: a novel attention network for SAR automatic target recognition[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2022, 15, 6615- 6630. doi: 10.1109/JSTARS.2022.3195074
21	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017, 30, 178.
22	WU C, HE T Q. A Survey of Applications of Vision Transformer and its Variants[C]//Proc. of the 10th IEEE International Conference on Intelligent Data and Security, 2024: 21−25.
23	MIA M S, ARNOB A B H, NAIM A, et al. ViTs are everywhere: a comprehensive study showcasing vision transformers in different domain[C]//Proc. of the International Conference on the Cognitive Computing and Complex Data, 2023: 101−117.
24	LV P Y, WU W J, ZHONG Y F, et al. Review of vision transformer models for remote sensing image scene classification[C]//Proc. of the IEEE International Geoscience and Remote Sensing Symposium, 2022: 2231−2234.
25	HAN K, WANG Y H, CHEN H T, et al. A survey on vision transformer[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2022, 45 (1): 87- 110.
26	LIU Z, LIN Y T, CAO Y, et al. Swin transformer: hierarchical vision transformer using shifted windows[C]//Proc. of the IEEE/CVF International Conference on Computer Vision, 2021: 10012−10022.
27	王智睿, 康玉卓, 曾璇, 等. SAR-AIRcraft-1.0: 高分辨率SAR飞机检测识别数据集[J]. 雷达学报, 2023, 12 (4): 906- 922. doi: 10.12000/JR23043
	WANG Z R, KANG Y Z, ZENG X, et al. SAR-AIRcraft-1.0: a high-resolution SAR aircraft detection and recognition dataset[J]. Journal of Radars, 2023, 12 (4): 906- 922. doi: 10.12000/JR23043
28	PARK J, WOO S, LEE J Y, et al. Bam: Bottleneck attention module[EB/OL]. [2024-06-16]. https://arxiv.org/abs/1807.06514.
29	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.
30	IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift[C]//Proc. of the International Conference on Machine Learning, 2015.
31	NAIR V, HINTON G E. Rectified linear units improve restricted boltzmann machines[C]//Proc. of the 27th International Conference on Machine Learning, 2010: 807−814.

层类型	卷积核大小	核个数	输出维度	参数量
Conv+BN+ReLU+CBAM	3×3	16	112×112×16	480
Conv+BN+ReLU+CBAM	5×5	32	108×108×32	12 896
Conv+BN+ReLU+CBAM	7×7	64	102×102×64	100 544
Conv+BN+ReLU+CBAM	5×5	128	98×98×128	205 184
MaxPooling	2×2	—	49×49×128	—
Conv+BN+ReLU+CBAM	5×5	256	45×45×256	819 968
MaxPooling	2×2	—	22×22×256	—
Conv+BN+ReLU+CBAM	6×6	128	17×17×128	1 180 032
MaxPooling	2×2	—	8×8×128	—
Conv+BN+ReLU+BAM	5×5	64	4×4×64	204 992
Conv	4×4	7	1×1×7	7 175
总计	—	—	—	2 531 271

方法	参数量	显存占用/MB
AlexNet^[11]	57 032 519	245.252
VGG-16^[12]	134 289 223	500.046
ResNet-101^[13]	42 514 503	190.341
A-ConvNets^[16]	325 511	1.3122
ResNeXt-101（64×4d）^[14]	83 455 272	326.549
Swin Transformer^[26]	87 768 224	127.973
所提方法	2 557 374	10.288

CBAM	BAM	准确率
CBAM*0	BAM*0	0.7909
CBAM*7	BAM*0	0.8045
CBAM*0	BAM*7	0.8166
CBAM*6	BAM*1	0.8319
CBAM*5	BAM*2	0.8245
CBAM*4	BAM*3	0.8257
CBAM*1	BAM*6	0.8183

方法	平均识别准确率	A220	A330	A320/321	ARJ21	Boeing737	Boeing787	其他
AlexNet^[11]	0.670 6	0.68	0.98	0.77	0.64	0.54	0.60	0.81
VGG-16^[12]	0.724 0	0.72	0.96	1.00	0.66	0.61	0.66	0.84
ResNet-101^[13]	0.771 7	0.81	0.96	1.00	0.85	0.68	0.68	0.81
A-ConvNets^[16]	0.782 5	0.84	1.00	0.97	0.76	0.75	0.66	0.83
ResNeXt-101（64×4d）^[14]	0.781 9	0.79	0.96	0.97	0.83	0.74	0.70	0.81
Swin Transformer^[26]	0.279 4	0.49	0.21	0.00	0.02	0.40	0.22	0.22
所提方法	0.831 9	0.90	0.98	1.00	0.85	0.78	0.76	0.84

[1]	付卫红, 彭文洪, 刘乃安. 混合注意力优化的SAR图像小目标检测方法[J]. 系统工程与电子技术, 2025, 47(8): 2519-2526.
[2]	倪康, 贾文杰, 邹旻瑞, 郑志忠. 基于动态聚合网络的SAR目标检测[J]. 系统工程与电子技术, 2025, 47(8): 2527-2539.
[3]	黄丰卓, 冯东, 华洋晟, 黄晓涛. 一种改进的全息SAR相位梯度自聚焦算法[J]. 系统工程与电子技术, 2025, 47(6): 1786-1795.
[4]	张新征, 闫梦可, 朱晓林. 噪声伪标签容忍的半监督SAR目标识别[J]. 系统工程与电子技术, 2025, 47(6): 1796-1805.
[5]	贾竣皓, 陈学斌, 李璋峰, 叶春茂. 空间群目标ISAR成像与检测方法[J]. 系统工程与电子技术, 2025, 47(6): 1843-1854.
[6]	刘宁, 李芳芳, 李新武, 洪文. 结合足迹和相位信息的SAR高层建筑三维重建[J]. 系统工程与电子技术, 2025, 47(5): 1469-1486.
[7]	段阿敏, 张朝辉. 基于二次分解的混合神经网络蜂窝流量预测[J]. 系统工程与电子技术, 2025, 47(5): 1687-1697.
[8]	贾钰嘉, 张思乾, 唐涛, 匡纲要. 强化散射特征的机载SAR实传图像盲超分辨重建[J]. 系统工程与电子技术, 2025, 47(3): 753-767.
[9]	杨志伟, 杨安东, 梁庚辰, 李相海, 李晓蕊, 刘杰. 多子带融合解动目标径向速度模糊方法[J]. 系统工程与电子技术, 2025, 47(3): 788-796.
[10]	陈世龙, 刘霖, 王晓蓓, 曾翰森, 刘亚波, 刘翔. 脉间捷变频SAR快速补偿频域成像处理算法[J]. 系统工程与电子技术, 2025, 47(3): 797-806.
[11]	贾蕾蕾, 刘利民, 董健. 基于图像结构信息的可见光和SAR图像快速配准[J]. 系统工程与电子技术, 2025, 47(2): 428-441.
[12]	付卫红, 张鑫钰, 刘乃安. 基于多尺度融合神经网络的同频同调制单通道盲源分离算法[J]. 系统工程与电子技术, 2025, 47(2): 641-649.
[13]	蒋李兵, 杨庆伟, 郑舒予, 王壮. 基于拍卖理论的组网雷达多轨道目标ISAR成像资源分配算法[J]. 系统工程与电子技术, 2025, 47(1): 81-93.
[14]	孟洋, 周国如, 李洁, 张冰尘. 基于结构化字典学习的判别稀疏微波成像方法[J]. 系统工程与电子技术, 2025, 47(1): 94-100.
[15]	陈洪猛, 李军, 刘京, 黄伟, 张英杰, 陈燕, 鲁耀兵. 基于Radon时频分析的海面舰船目标SAR-ISAR混合成像方法[J]. 系统工程与电子技术, 2025, 47(1): 109-116.