基于改进的CNN和数据增强的SAR目标识别

doi:10.12305/j.issn.1001-506X.2022.08.12

摘要/Abstract

摘要：

针对合成孔径雷达(synthetic aperture radar, SAR)图像目标识别问题, 提出了基于改进的卷积神经网络和数据增强的SAR目标识别方法。首先在训练阶段引入Dropout, 随机删除部分神经元, 增强网络的泛化能力。其次, 在网络中引入L2正则化, 简化模型的同时降低结构风险, 并且能有效地抑制过拟合。然后, 采用Adam优化网络, 提高模型的收敛效率。最后, 采用优选的数据增强方法, 扩充SAR目标数据集, 为网络训练提供更为充足的样本, 进一步提高识别的准确率和模型的泛化性。在运动和静止目标获取与识别(moving and stationary target acquisition and recognition, MSTAR)数据集上进行了实验, 结果表明设计的卷积神经网络识别准确率高, 且具有更好的泛化性。

关键词: 雷达目标识别, 合成孔径雷达, 卷积神经网络, 数据增强, 正则化

Abstract:

For the problem of target recognition in synthetic aperture radar (SAR) image, a method of SAR target recognition based on improved convolution neural network (CNN) and data augmentation is proposed. Firstly, Dropout is brought in the training phase to randomly delete some neurons, so that the generalization ability of the network is enhanced. Secondly, L2 regularization is introduced into the network to reduce the structural risk and effectively restrain the over fitting. Then, Adam is used to optimize the network to improve the convergence efficiency of the model. Finally, the preferred rotation data augmentation method is employed for expanding the data set of SAR target. Through the improved network and increased data, the recognition accuracy and generalization of the model are enhanced. Experiments on moving and stationary target acquisition and recognition (MSTAR) data set show that the proposed method has higher recognition accuracy and better generalization.

Key words: radar target recognition, synthetic aperture radar (SAR), convolutional neural network (CNN), data augmentation, regularization

中图分类号:

TN957.52

王彩云, 吴钇达, 王佳宁, 马璐, 赵焕玥. 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.

Caiyun WANG, Yida WU, Jianing WANG, Lu MA, Huanyue ZHAO. SAR image target recognition based on combinatorial optimization convolutional neural network[J]. Systems Engineering and Electronics, 2022, 44(8): 2483-2487.

图/表 8

图1

图2

图3

表1

表2

表3

表4

表5

参考文献 21

1	贺丰收, 何友, 刘准钆, 等. 卷积神经网络在雷达自动目标识别中的研究进展[J]. 电子与信息学报, 2020, 42 (1): 119- 131.
	HE F S , HE Y , LIU H G , et al. Research progress of convolution neural network in radar automatic target recognition[J]. Journal of Electronics and Information Technology, 2020, 42 (1): 119- 131.
2	KRIZHEVSKY A , SUTSKEVER I , HINTON G . ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60 (6): 84- 90. doi: 10.1145/3065386
3	田壮壮, 占荣辉, 胡杰民, 等. 基于卷积神经网络的SAR图像目标识别研究[J]. 雷达学报, 2016, 5 (3): 320- 325.
	TIAN Z Z , ZHAN R H , HU J M , et al. Research on SAR image target recognition based on convolution neural network[J]. Journal of Radars, 2016, 5 (3): 320- 325.
4	王彩云, 胡允侃, 李晓飞, 等. 基于卷积稀疏编码与多分类器融合的雷达HRRP目标识别方法[J]. 系统工程与电子技术, 2018, 40 (11): 2433- 2437. doi: 10.3969/j.issn.1001-506X.2018.11.07
	WANG C Y , HU Y K , LI X F , et al. Radar HRRP target re-cognition method based on convolutional sparse coding and multi classifier fusion[J]. Systems Engineering and Electronics, 2018, 40 (11): 2433- 2437. doi: 10.3969/j.issn.1001-506X.2018.11.07
5	李君宝, 杨文慧, 许剑清, 等. 基于深度卷积网络的SAR图像目标检测识别[J]. 导航定位与授时, 2017, 4 (1): 60- 66.
	LI J B , YANG W H , XU J Q , et al. SAR image target detection and recognition based on deep convolution network[J]. Navigation Positioning and Timing, 2017, 4 (1): 60- 66.
6	刘晨, 曲长文, 周强, 等. 基于CNN的SAR图像目标分类优化算法[J]. 雷达科学与技术, 2017, 15 (4): 362- 367.
	LIU C , QU C W , ZHOU Q , et al. Target classification optimization algorithm of SAR image based on CNN[J]. Radar Science and Technology, 2017, 15 (4): 362- 367.
7	高志帮, 刘以安. 多尺度分解结合卷积神经网络的SAR图像识别[J]. 火控雷达技术, 2019, 48 (1): 1- 6.
	GAO Z B , LIU Y A . SAR image recognition based on multiscale decomposition and convolutional neural network[J]. Fire Control Radar Technology, 2019, 48 (1): 1- 6.
8	任硕良, 索继东, 佟禹. 卷积神经网络结合迁移学习的SAR目标识别[J]. 电光与控制, 2020, 27 (10): 37- 41. doi: 10.3969/j.issn.1671-637X.2020.10.008
	REN S L , SUO J D , TONG Y . SAR target recognition based on convolution neural network and transfer learning[J]. Electronics Opitics & Control, 2020, 27 (10): 37- 41. doi: 10.3969/j.issn.1671-637X.2020.10.008
9	LI J W , QU C W , PENG S J . Ship classification for unbalanced SAR dataset based on convolutional neural network[J]. Journal of Applied Remote Sensing, 2018,
10	陈欣, 陈明逊. 基于增强数据集卷积神经网络的SAR目标识别方法[J]. 重庆理工大学学报(自然科学), 2020, 34 (1): 86- 93.
	CHEN X , CHEN M X . SAR target recognition method based on convolutional neural network of enhanced dataset[J]. Journal of Chongqing University of Technology(Natural Science), 2020, 34 (1): 86- 93.
11	满凤环, 陈秀宏, 何佳佳. 改进的Dropout正则化卷积神经网络[J]. 传感器与微系统, 2018, 37 (4): 44- 47.
	MAN F H , CHEN X H , HE J J . Improved dropout regularized convolutional neural network[J]. Transducer and Microsystem Technologies, 2018, 37 (4): 44- 47.
12	王博威, 潘宗序, 胡玉新, 等. 少量样本下基于孪生CNN的SAR目标识别[J]. 雷达科学与技术, 2019, 17 (6): 603- 609.603-609, 615 doi: 10.3969/j.issn.1672-2337.2019.06.003
	WANG B W , PAN Z X , HU Y X , et al. SAR target recognition based on twin CNN with few samples[J]. Radar Science and Technology, 2019, 17 (6): 603- 609.603-609, 615 doi: 10.3969/j.issn.1672-2337.2019.06.003
13	周飞燕, 金林鹏, 董军. 卷积神经网络研究综述[J]. 计算机学报, 2017, 40 (6): 1229- 1251.
	ZHOU F Y , JIN L P , DONG J . Review of convolutional neural networks[J]. Chinese Journal of Computers, 2017, 40 (6): 1229- 1251.
14	张琴. 基于CNN的图像分类中激活函数的研究[J]. 现代计算机, 2020, (32): 43- 47. doi: 10.3969/j.issn.1007-1423.2020.32.009
	ZHANG Q . Research on activation function in image classification based on CNN[J]. Modern Computer, 2020, (32): 43- 47. doi: 10.3969/j.issn.1007-1423.2020.32.009
15	姬壮伟. 神经网络过拟合问题的Dropout优化[J]. 长治学院学报, 2020, 37 (2): 37- 39. doi: 10.3969/j.issn.1673-2014.2020.02.011
	JI Z W . Dropout optimization of neural network over fitting problem[J]. Journal of Changzhi University, 2020, 37 (2): 37- 39. doi: 10.3969/j.issn.1673-2014.2020.02.011
16	BURDEN F , WINKLER D . Bayesian regularization of neural networks[J]. Methods in Molecular Biology, 2008, 458, 25- 44.
17	TICKNOR J L . A Bayesian regularized artificial neural network for stock market forecasting[J]. Expert Systems with Applications, 2013, 40 (14): 5501- 5506. doi: 10.1016/j.eswa.2013.04.013
18	高友文, 周本君, 胡晓飞. 基于数据增强的卷积神经网络图像识别研究[J]. 计算机技术与发展, 2018, (8): 62- 65.
	GAO Y W , ZHOU B J , HU X F . Convolution neural network image recognition based on data enhancement[J]. Computer Technology and Development, 2018, (8): 62- 65.
19	XIA S , YU Z , YU J D . Adaptive convolutional network for SAR image classification[J]. The Journal of Engineering, 2019, 2019 (20): 6868- 6872. doi: 10.1049/joe.2019.0565
20	王源源. 一种基于多分辨率表示的SAR图像识别方法[J]. 电光与控制, 2020, 27 (10): 31- 36.
	WANG Y Y . SAR image recognition method based on multi resolution representation[J]. Electronics Optics & Control, 2020, 27 (10): 31- 36.
21	应自炉, 宣晨, 翟懿奎, 等. 面向小样本SAR图像识别的自注意力多尺度特征融合网络[J]. 信号处理, 2020, 36 (11): 1846- 1858.
	YING Z L , XUAN C , ZHAI Y K , et al. Self attention multi scale feature fusion network for small sample SAR image recognition[J]. Journal of Signal Processing, 2020, 36 (11): 1846- 1858.

概率	尺寸
概率	32×32	64×64	128×128
0	0.983 4	0.979 4	0.977 4
0.3	0.985 2	0.980 2	0.972 0
0.5	0.971 8	0.972 8	0.959 1

正则化方案	平均识别准确率
无正则化	0.985 2
L1正则化λ=0.001	0.982 9
L2正则化λ=0.001	0.980 3
L2正则化λ=0.005	0.988 2

数据增强方法	平均识别准确率
无数据增强	0.985 2
高斯模糊	0.980 4
上下翻转	0.974 5
左右翻转	0.980 2
裁剪	0.964 2
旋转	0.990 2
旋转+左右翻转	0.984 3

目标类别	识别输出类别										识别率	平均识别率
目标类别	BMP2	BTR70	T72	2S1	BRDM2	BTR60	D7	T62	ZIL 131	ZSU 234	识别率	平均识别率
BMP2	193	2	0	0	0	0	0	0	0	0	0.989 7	0.992 0
BTR70	0	196	0	0	0	0	0	0	0	0	1
T72	1	0	195	0	0	0	0	0	0	0	0.994 9
2S1	0	0	0	274	0	0	0	0	0	0	1
BRDM2	0	0	0	0	274	0	0	0	0	0	1
BTR60	0	0	0	0	4	191	0	0	0	0	0.979 5
D7	0	0	0	0	2	0	272	0	0	0	0.992 7
T62	0	0	0	4	2	0	0	267	0	0	0.978 0
ZIL131	0	0	0	0	1	0	1	0	272	0	0.992 7
ZSU234	0	0	0	0	0	0	0	1	1	272	0.992 7

方法	平均识别率/%
自适应CNN法^[19]	97.83
多分辨率表示法^[20]	98.72
自注意力多尺度特征融合法^[21]	98.22
本文方法	99.20