多任务协同优化学习高分辨SAR稀疏自聚焦成像算法

杨磊; 张苏; 黄博; 盖明慧; 李埔丞

doi:10.11999/JEIT200300

多任务协同优化学习高分辨SAR稀疏自聚焦成像算法

doi: 10.11999/JEIT200300

杨磊^1, ,,
张苏¹,
黄博²,
盖明慧¹,
李埔丞¹

1.
中国民航大学天津市智能信号与图像处理重点实验室天津 300300
2.
中国工程物理研究院电子工程研究所绵阳 621999

基金项目: 国家自然科学基金(61601470)，天津市自然科学基金(16JCYBJC41200)，预研基金(61406190101)

详细信息

作者简介:
杨磊：男，1984年生，副教授，研究方向为高分辨SAR成像及机器学习理论应用

张苏：女，1996年生，硕士生，研究方向为高分辨SAR成像及优化学习理论

黄博：男，1986年生，博士生，研究方向为雷达高度表系统及信号处理

盖明慧：女，1997年生，硕士生，研究方向为高分辨SAR成像及优化学习理论

李埔丞：男，1992年生，博士生，研究方向为高分辨SAR成像及优化学习理论

通讯作者:
杨　磊　yanglei840626@163.com

中图分类号: TN957.52
计量
- 文章访问数: 1406
- HTML全文浏览量: 573
- PDF下载量: 74
- 被引次数: 41
出版历程
- 收稿日期: 2020-04-24
- 修回日期: 2021-02-28
- 网络出版日期: 2021-03-22
- 刊出日期: 2021-09-16

Multi-task Learning of Sparse Autofocusing for High-Resolution SAR Imagery

1.
Tianjin Key Laboratory for Advanced Signal Processing, Civil Aviation University of China, Tianjin 300300, China
2.
Institute of Electronic Engineering, Chinese Academy of Engineering Physics, Mianyang 621999, China

Funds: The National Natural Science Foundation of China(61601470), The Natural Science Foundation of Tianjin, China (16JCYBJC41200), The Equipment Pre-research Fund(61406190101)

摘要

摘要: 针对传统高分辨合成孔径雷达(SAR)稀疏自聚焦成像算法难以有效平衡稀疏与聚焦特征的问题，该文提出一种基于交替方向多乘子方法(ADMM)的多任务协同优化学习稀疏自聚焦(MtL-SA)算法。该算法通过引入熵范数表征SAR成像结果聚焦特征，在ADMM优化框架下，利用近端算法求解聚焦特征解析解。针对原熵范数正则优化目标函数的非凸问题，该文合理设计代价函数，从而保证熵范数近端算子的闭合解析解。同时，应用 $\ell {_1}$ 范数表征成像结果稀疏特征，并建立面向复数SAR成像数据的复数软阈值近端算子。该文所提MtL-SA成像算法可实现对目标场景后向散射场对应稀疏特征和聚焦特征的解析求解，并有效提升自聚焦算法的可靠性和稳健性。两种特征增强处理相互调和，保证了算法运行过程中有效降低误差传播，进而保证联合特征增强精度。仿真及实测机载SAR成像数据实验，验证了算法的有效性和实用性，同时应用相变分析方法分别定量和定性地分析了该文所提算法相比其他传统算法的优越性。
- 合成孔径雷达 /
- 多任务学习 /
- 多特征增强 /
- 熵范数 /
- 近端算子
Abstract: As it is difficult to balance the sparse and focusing features for conventional sparse autofocusing algorithm of Synthetic Aperture Radar (SAR), a Multi-task Learning Sparse Autofocusing (MtL-SA) algorithm is proposed under a novel Alternating Direction Method of Multipliers (ADMM) in this paper. The image entropy norm is introduced to model the focusing feature of the SAR imagery, and it is minimized in a regularized manner using the proximal algorithm. To overcome the non-convexity of the original objective function, a surrogate function under the ADMM framework is designed and optimized accordingly. This ensures closed-form solution of the errors and the focusing feature. Besides, the $\ell {_1}$ -norm is applied to denote the intended sparse feature of the SAR imagery, and a complex-valued proximity operator is derived for the range-compressed SAR data. Due to the cooperative framework, both the features can be solved and achieved with high robustness and acceptable accuracy. Compared with conventions, the computational efficiency improved twice orders in terms of CPU time. The proposed MtL-SA algorithm can realize the analytical solutions of the sparse and focusing features, so as to improve the robustness of the joint enhancement. Experiments using airborne simulated and raw SAR data are performed to verify the effectiveness of the proposed algorithm. Phase transition analysis is applied to examine the superiority of the proposed algorithm compared with the conventions in terms of both quantitative and qualitative.
- Synthetic Aperture Radar (SAR) /
- Multi-task learning /
- Multi-feature enhancement /
- Entropy norm /
- Proximity operator

HTML全文

图 1 SAR几何模型示意图

下载: 全尺寸图片幻灯片

图 2 仿真实验结果

下载: 全尺寸图片幻灯片

图 3 实测数据实验结果

下载: 全尺寸图片幻灯片

图 4 2维等高线对比图

下载: 全尺寸图片幻灯片

图 5 信噪比-降采样率相变热力图

下载: 全尺寸图片幻灯片

表 1 MtL-SA算法流程

　步骤1 设定初值

${ {\boldsymbol{X} } }^{0}={ {\boldsymbol{Z} } }^{0}={ {\boldsymbol{D} } }^{0}={{{\textit{0}}} },\;k=0,\;G{=2}$ ，设定迭代停止准则，开始循环。

　步骤2 全局优化：

${\boldsymbol{X}}$ 更新运算

${ {\boldsymbol{X} }^{k + 1} } = \left[ { { {\boldsymbol{A} }^{\rm{H} } }{ {\boldsymbol{E} }^{\rm{H} } }{\boldsymbol{Y} } + {\rho _1}\left( { { {\boldsymbol{Z} }_1}^k + { {\boldsymbol{D} }_1}^k} \right) + {\rho _2}\left( { { {\boldsymbol{Z} }_2}^k + { {\boldsymbol{D} }_2}^k} \right)} \right] \cdot {\left( { { {\boldsymbol{A} }^{\rm{H} } }{\boldsymbol{A} } + \rho G{\boldsymbol{I} } } \right)^{ - 1} }\quad \quad \quad \quad$

　步骤3 局部优化：

${{\boldsymbol{Z}}_1}$ ,

${{\boldsymbol{D}}_1}$ ,

${{\boldsymbol{Z}}_2}$ 和

${{\boldsymbol{D}}_2}$ 顺次更新运算

${\rm{for }}\;\;g = G$

${\boldsymbol{Z} }_{\rm{1} }^{k + 1} = {\left( { { {\boldsymbol{E} }^{k + 1} } } \right)^{\rm{H} } }{ {\boldsymbol{A} }^{\rm{H} } }{\boldsymbol{Y} },\;{\boldsymbol{D} }_{\rm{1} }^{k + 1} = {\boldsymbol{D} }_{\rm{1} }^k - { {\boldsymbol{X} }^{k + 1} } + {\boldsymbol{Z} }_{\rm{1} }^{k + 1}\quad \quad \quad \quad \quad$

${\boldsymbol{Z} }_2^{k + 1} = {\rm{pro} }{ {\rm{x} }_{ { { {\lambda _2} } / \rho } } }\left[ { { {\boldsymbol{W} }^{k + 1} } } \right]{\rm{, } }{\boldsymbol{D} }_2^{k + 1} = {\boldsymbol{D} }_2^k - { {\boldsymbol{X} }^{k + 1} } + {\boldsymbol{Z} }_2^{k + 1}\quad \quad \quad \quad$

${\rm{end}}$

${{\boldsymbol{Z}}^{k + 1}} = \left[ {{\boldsymbol{Z}}_1^{k + 1}\;{\boldsymbol{Z}}_2^{k + 1}} \right],\;{{\boldsymbol{D}}^{k + 1}} = \left[ {{\boldsymbol{D}}_1^{k + 1}\;{\boldsymbol{D}}_2^{k + 1}} \right],\;k = k + 1$

　步骤4 当残差小于加性噪声方差时，跳至步骤5，迭代结束。否则，跳至步骤2。

　步骤5 输出联合稀疏聚焦特征增强后的图像数据

${\boldsymbol{X}}$ 。

下载: 导出CSV

参考文献(20)

[1]	闫贺, 王珏, 黄佳, 等. 基于二维速度搜索的星载SAR运动目标聚焦算法研究[J]. 电子与信息学报, 2019, 41(6): 1287–1293. doi: 10.11999/JEIT180663 YAN He, WANG Jue, HUANG Jia, et al. A moving-targets detection algorithm for spaceborne SAR system based on two-dimensional velocity search method[J]. Journal of Electronics &Information Technology, 2019, 41(6): 1287–1293. doi: 10.11999/JEIT180663
[2]	李煜, 陈杰, 张渊智. 合成孔径雷达海面溢油探测研究进展[J]. 电子与信息学报, 2019, 41(3): 751–762. doi: 10.11999/JEIT180468 LI Yu, CHEN Jie, and ZAHNG Yuanzhi. Progress in research on marine oil spills detection using synthetic aperture radar[J]. Journal of Electronics &Information Technology, 2019, 41(3): 751–762. doi: 10.11999/JEIT180468
[3]	DONOHO D L. Compressed sensing[J]. IEEE Transactions on Information Theory, 2006, 52(4): 1289–1306. doi: 10.1109/TIT.2006.871582
[4]	田鹤, 于海锋, 朱宇, 等. 基于频域稀疏压缩感知的星载SAR稀疏重航过3维成像[J]. 电子与信息学报, 2020, 42(8): 2021–2028. doi: 10.11999/JEJT190638 TIAN He, YU Haifeng, ZHU Yu, et al. Sparse flight 3-D imaging of spaceborne SAR based on frequency domain sparse compressed sensing[J]. Journal of Electronics &Information Technology, 2020, 42(8): 2021–2028. doi: 10.11999/JEJT190638
[5]	YANG Lei, ZHAO Lifan, BI Guoan, et al. SAR ground moving target imaging algorithm based on parametric and dynamic sparse Bayesian learning[J]. IEEE Transactions on Geoscience and Remote Sensing, 2016, 54(4): 2254–2267. doi: 10.1109/TGRS.2015.2498158
[6]	ALONSO M T, LOPEZ-DEKKER P, and MALLORQUI J J. A novel strategy for radar imaging based on compressive sensing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2010, 48(12): 4285–4295. doi: 10.1109/TGRS.2010.2051231
[7]	ZHAO Lifan, WANG Lu, BI Guoan, et al. An autofocus technique for high-resolution inverse synthetic aperture radar imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014, 52(10): 6392–6403. doi: 10.1109/TGRS.2013.2296497
[8]	张群英, 江兆凤, 李超, 等. 太赫兹合成孔径雷达成像运动补偿算法[J]. 电子与信息学报, 2017, 39(1): 129–137. doi: 10.11999/JEIT160201 ZHANG Qunying, JIANG Zhaofeng, LI Chao, et al. Motion compensation imaging algorithm of terahertz synthetic aperture radar[J]. Journal of Electronics &Information Technology, 2017, 39(1): 129–137. doi: 10.11999/JEIT160201
[9]	ZHOU Song, YANG Lei, ZHAO Lifan, et al. Quasi-polar-based FFBP algorithm for miniature UAV SAR imaging without navigational data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(12): 7053–7065. doi: 10.1109/TGRS.2017.2739133
[10]	LINNEHAN R, MILLER J, and ASADI A. Map-drift autofocus and scene stabilization for video-SAR[C]. 2018 IEEE Radar Conference, Oklahoma City, USA, 2018: 1401–1405. doi: 10.1109/RADAR.2018.8378769.
[11]	MAO Xinhua, HE Xueli, and LI Danqi. Knowledge-aided 2-D autofocus for spotlight SAR range migration algorithm imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(9): 5458–5470. doi: 10.1109/TGRS.2018.2817507
[12]	KRAGH T J and KHARBOUCH A A. Monotonic iterative algorithms for SAR image restoration[C]. 2006 IEEE International Conference on Image Processing (ICIP), Atlanta, USA, 2006: 645–648. doi: 10.1109/ICIP.2006.312463.
[13]	张柘, 张冰尘, 洪文, 等. 结合MD自聚焦算法与回波模拟算子的快速稀疏微波成像误差补偿算法[J]. 雷达学报, 2016, 5(1): 25–34. doi: 10.12000/JR15055 ZHANG Zhe, ZHANG Bingchen, HONG Wen, et al. Accelerated sparse microwave imaging phase error compensation algorithm based on combination of SAR raw data simulator and Map-drift autofocus algorithm[J]. Journal of Radars, 2016, 5(1): 25–34. doi: 10.12000/JR15055
[14]	GÜNGÖR A, ÇETIN M, and GÜVEN H E. Autofocused compressive SAR imaging based on the alternating direction method of multipliers[C]. 2017 IEEE Radar Conference (RadarConf), Seattle, USA, 2017: 1573–1576. doi: 10.1109/RADAR.2017.7944458.
[15]	BOYD S, PARIKH N, CHU E, et al. Distributed optimization and statistical learning via the alternating direction method of multipliers[J]. Foundations and Trends ^® in Machine Learning, 2011, 3(1): 1–122. doi: 10.1561/2200000016
[16]	GÜVEN H E, GÜNGÖR A, and ÇETIN M. An augmented Lagrangian method for complex-valued compressed SAR imaging[J]. IEEE Transactions on Computational Imaging, 2016, 2(3): 235–250. doi: 10.1109/TCI.2016.2580498
[17]	ÖNHON N Ö and ÆETIN M. A sparsity-driven approach for SAR image formation and space-variant focusing[C]. 2011 IEEE 19th Signal Processing and Communications Applications Conference (SIU), Antalya, Turkey, 2011: 614–617. doi: 10.1109/SIU.2011.5929725.
[18]	刘碧丹, 王岩飞, 韩松. 距离徙动校正和斜地变换的实时算法研究[J]. 电子与信息学报, 2009, 31(5): 1099–1102. doi: 10.3724/SP.J.1146.2008.00192 LIU Bidan, WANG Yanfei, and HAN Song. A real-time associative algorithm of RCMC and SRGR[J]. Journal of Electronics &Information Technology, 2009, 31(5): 1099–1102. doi: 10.3724/SP.J.1146.2008.00192
[19]	MALEKI A, ANITORI L, YANG Z, et al. Asymptotic analysis of complex LASSO via Complex Approximate Message Passing (CAMP)[J]. IEEE Transactions on Information Theory, 2013, 59(7): 4290–4308. doi: 10.1109/TIT.2013.2252232
[20]	DONOHO D L, MALEKI A, and MONTANARI A. The noise-sensitivity phase transition in compressed sensing[J]. IEEE Transactions on Information Theory, 2011, 57(10): 6920–6941. doi: 10.1109/TIT.2011.2165823

施引文献

期刊类型引用(20)

1.	周晨，周乾伟，陈翰墨，管秋，胡海根，吴延壮. 面向RGBD图像显著性检测的循环逐尺度融合网络. 小型微型计算机系统. 2023(10): 2276-2283 . 百度学术
2.	叶海峰，赵玉琛. 视觉位置识别中代表地点的标识牌算法. 小型微型计算机系统. 2021(04): 823-828 . 百度学术
3.	王慧玲，宋鑫怡，杨颖. 基于优化查询的改进显著性检测算法. 吉林大学学报(信息科学版). 2020(03): 319-324 . 百度学术
4.	郭迎春，李卓. 基于边缘特征和自适应融合的视频显著性检测. 河北工业大学学报. 2019(01): 1-7 . 百度学术
5.	鲁文超，段先华，徐丹，王万耀. 基于多尺度下凸包改进的贝叶斯模型显著性检测算法. 计算机科学. 2019(06): 295-300 . 百度学术
6.	王宝艳，张铁，李凯，杜松林. DEL分割算法对SSLS算法的改进. 小型微型计算机系统. 2019(10): 2052-2057 . 百度学术
7.	张巧荣，徐国愚，张俊峰. 利用视觉显著性的前景目标分割. 兰州大学学报(自然科学版). 2019(06): 833-840 . 百度学术
8.	杨俊丰，林亚平，欧博，蒋军强，李强. 基于显著性加权随机优化的快速响应码美化方法. 电子与信息学报. 2018(02): 289-297 . 本站查看
9.	邓晨，谢林柏. 全局对比和背景先验驱动的显著目标检测. 计算机工程与应用. 2018(03): 212-216 . 百度学术
10.	刘亚宁，吴清，魏雪. 基于流行排序的前景背景显著性检测算法. 科学技术与工程. 2018(18): 74-81 . 百度学术
11.	闫钧华，肖勇旗，姜惠华，杨勇，张寅. 融合区域像素显著性和时域信息的地面动目标检测及其DSP实现. 电子设计工程. 2018(19): 178-183+193 . 百度学术
12.	陈厚仁，蔡延光. 基于视频的干线交通流检测系统的研究与实现. 工业控制计算机. 2017(07): 85-87 . 百度学术
13.	赵艳艳，沈西挺. 基于同步更新的背景检测显著性优化. 计算机工程. 2017(10): 264-267 . 百度学术
14.	田畅，姜青竹，吴泽民，刘涛，胡磊. 基于区域协方差的视频显著度局部空时优化模型. 电子与信息学报. 2016(07): 1586-1593 . 本站查看
15.	罗会兰，万成涛，孔繁胜. 基于KL散度及多尺度融合的显著性区域检测算法. 电子与信息学报. 2016(07): 1594-1601 . 本站查看
16.	张晴，林家骏，戴蒙. 基于图的流行排序的显著目标检测改进算法. 计算机工程与应用. 2016(22): 26-32+38 . 百度学术
17.	杜永强. 过度曝光图像缺失信息修复算法. 科技通报. 2016(08): 146-149 . 百度学术
18.	郎波，樊一娜，黄静. 利用混合高斯进行物体成分拟合匹配的算法. 科学技术与工程. 2016(20): 73-80 . 百度学术
19.	项导，侯赛辉，王子磊. 基于背景学习的显著物体检测. 中国图象图形学报. 2016(12): 1634-1643 . 百度学术
20.	吕建勇，唐振民. 一种基于图的流形排序的显著性目标检测改进方法. 电子与信息学报. 2015(11): 2555-2563 . 本站查看