Article(id=1156963928124253027, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1156963927277003616, articleNumber=null, orderNo=null, doi=10.12404/j.issn.1671-1815.2403736, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1716134400000, receivedDateStr=2024-05-20, revisedDate=1735747200000, revisedDateStr=2025-01-02, acceptedDate=null, acceptedDateStr=null, onlineDate=1753771295815, onlineDateStr=2025-07-29, pubDate=1742227200000, pubDateStr=2025-03-18, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1753771295815, onlineIssueDateStr=2025-07-29, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1753771295815, creator=13701087609, updateTime=1753771295815, updator=13701087609, issue=Issue{id=1156963927277003616, tenantId=1146029695717560320, journalId=1146123166801305609, year='2025', volume='25', issue='8', pageStart='3079', pageEnd='3528', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=0, createTime=1753771295613, creator=13701087609, updateTime=1753777038876, updator=13701087609, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1156988016305726153, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1156963927277003616, language=EN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1156988016305726154, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1156963927277003616, language=CN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=3268, endPage=3279, ext={EN=ArticleExt(id=1156963928749204329, articleId=1156963928124253027, tenantId=1146029695717560320, journalId=1146123166801305609, language=EN, title=SAR Simulation Data Optimization Based on Target Characteristic Constraints, columnId=1156963928648541032, journalTitle=Science Technology and Engineering, columnName=Electronic and Communicational Technology, runingTitle=null, highlight=null, articleAbstract=

Synthetic aperture radar (SAR) target recognition method based on deep networks requires a large amount of training data, and in practical applications, it is extremely difficult for SAR imaging systems to obtain sufficient and evenly distributed target data. One way to solve the small sample problem in SAR target recognition, is to use electromagnetic simulation technology to generate a large amount of SAR simulation data. However, there are still significant differences between simulated images and measured SAR images, so using simulation data directly cannot bring significant performance improvement for target recognition. A simulation data optimization method based on SAR target characteristic constraints was proposed to address the above issues. On the basis of analyzing the characteristics of SAR targets, a texture structure cycle-consistent generative adversarial network (TS-CycleGAN) based on texture structure and cycle consistency was constructed, in which the structural similarity measure was used to constrain the generation process of CycleGAN. This method can reduce the difference between simulation data and measured data, and can improve the usability of simulation data. The experimental results on the SAR SAMPLE dataset show that, compared to other simulation data optimization methods, the proposed method achieves significant improvements in image quality evaluation and classification performance.

, correspAuthors=Zong-yong CUI, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, authorCompany=null, fund=null, authors=null, authorsList=Hui ZHANG, Li-qiang MOU, Yi-wei LI, Zong-yong CUI), CN=ArticleExt(id=1156964022177325554, articleId=1156963928124253027, tenantId=1146029695717560320, journalId=1146123166801305609, language=CN, title=基于目标特性约束的SAR 仿真数据优化, columnId=1154014169461411904, journalTitle=科学技术与工程, columnName=电子技术、通信技术, runingTitle=null, highlight=null, articleAbstract=

基于深度网络的合成孔径雷达(synthetic aperture radar, SAR)目标识别方法,需要大量训练数据,而实际应用中 SAR 成像系统获取数量充足、分布均匀的目标数据难度极高。解决SAR目标识别小样本问题的途径之一,就是采用电磁仿真技术生成大量 SAR仿真数据。然而仿真图像与实测 SAR图像仍存在较大差异,直接使用仿真数据并不能带来目标识别性能的显著提升。针对上述问题,提出一种基于SAR目标特性约束的仿真数据优化方法。该方法在分析 SAR目标特性的基础上,构建基于纹理结构-循环一致性的生成对抗网络(texture structure cycle-consistent generative adversarial network, TS-CycleGAN),采用结构相似度衡量指标约束 CycleGAN的生成过程,降低仿真数据与实测数据之间的差异,提高仿真数据的可用性。针对SAR SAMPLE 数据集的试验结果表明,所提出的方法相较于其他仿真数据优化方法,在图像质量评估和分类性能方面,都取得了明显的提升。

, correspAuthors=崔宗勇, authorNote=null, correspAuthorsNote=
*崔宗勇(1984—),男,汉族,山东菏泽人,博士,副教授。研究方向:SAR图像解译。E-mail: zycui@uestc.edu.cn。
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=9EDS/gMZKu7TGLYWUwjk+w==, magXml=HRZ4BlNrT0is9rwwzORNmw==, pdfUrl=null, pdf=qwhK9vabWQRaBAR6gRSqUg==, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=cABG5zAj/QXJ86Pi8t/+0g==, mapNumber=null, authorCompany=null, fund=null, authors=

张慧(1984-),女,汉族,山东菏泽人,硕士,副教授。研究方向:模式识别、人工智能。E-mail: zh21y@163.com。

, authorsList=张慧, 牟立强, 李沂蔚, 崔宗勇)}, authors=[Author(id=1156986835080663873, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=zh21y@163.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1156986835688837973, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, authorId=1156986835080663873, language=EN, stringName=Hui ZHANG, firstName=Hui, middleName=null, lastName=ZHANG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 School of Engineering Chengdu College of University of Electronic Science and Technology of China Chengdu 611731 China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1156986835751752535, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, authorId=1156986835080663873, language=CN, stringName=张慧, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 电子科技大学成都学院 工学院 成都 611731, bio={"content":"

张慧(1984-),女,汉族,山东菏泽人,硕士,副教授。研究方向:模式识别、人工智能。E-mail: zh21y@163.com。

"}, bioImg=null, bioContent=

张慧(1984-),女,汉族,山东菏泽人,硕士,副教授。研究方向:模式识别、人工智能。E-mail: zh21y@163.com。

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1156986832723464927, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, xref=1, ext=[AuthorCompanyExt(id=1156986832731853536, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986832723464927, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Engineering Chengdu College of University of Electronic Science and Technology of China Chengdu 611731 China), AuthorCompanyExt(id=1156986834921280311, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986832723464927, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 电子科技大学成都学院 工学院 成都 611731)])]), Author(id=1156986835823055705, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1156986836502532974, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, authorId=1156986835823055705, language=EN, stringName=Li-qiang MOU, firstName=Li-qiang, middleName=null, lastName=MOU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2 School of Information and Communication Engineering University of Electronic Science and Technology of China Chengdu 611731 China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1156986836561253234, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, authorId=1156986835823055705, language=CN, stringName=牟立强, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2 电子科技大学 信息与通信工程学院 成都 611731, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1156986834988389179, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, xref=2, ext=[AuthorCompanyExt(id=1156986834992583483, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986834988389179, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 School of Information and Communication Engineering University of Electronic Science and Technology of China Chengdu 611731 China), AuthorCompanyExt(id=1156986835000972092, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986834988389179, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 电子科技大学 信息与通信工程学院 成都 611731)])]), Author(id=1156986836619973494, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1156986836699665275, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, authorId=1156986836619973494, language=EN, stringName=Yi-wei LI, firstName=Yi-wei, middleName=null, lastName=LI, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2 School of Information and Communication Engineering University of Electronic Science and Technology of China Chengdu 611731 China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1156986836800328576, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, authorId=1156986836619973494, language=CN, stringName=李沂蔚, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2 电子科技大学 信息与通信工程学院 成都 611731, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1156986834988389179, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, xref=2, ext=[AuthorCompanyExt(id=1156986834992583483, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986834988389179, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 School of Information and Communication Engineering University of Electronic Science and Technology of China Chengdu 611731 China), AuthorCompanyExt(id=1156986835000972092, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986834988389179, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 电子科技大学 信息与通信工程学院 成都 611731)])]), Author(id=1156986836892603271, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=zycui@uestc.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1156986837249119124, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, authorId=1156986836892603271, language=EN, stringName=Zong-yong CUI, firstName=Zong-yong, middleName=null, lastName=CUI, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2 School of Information and Communication Engineering University of Electronic Science and Technology of China Chengdu 611731 China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1156986837328810903, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, authorId=1156986836892603271, language=CN, stringName=崔宗勇, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2 电子科技大学 信息与通信工程学院 成都 611731, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1156986834988389179, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, xref=2, ext=[AuthorCompanyExt(id=1156986834992583483, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986834988389179, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 School of Information and Communication Engineering University of Electronic Science and Technology of China Chengdu 611731 China), AuthorCompanyExt(id=1156986835000972092, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986834988389179, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 电子科技大学 信息与通信工程学院 成都 611731)])])], keywords=[Keyword(id=1156986837890847651, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, orderNo=1, keyword=synthetic aperture radar (SAR)), Keyword(id=1156986837962150821, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, orderNo=2, keyword=simulation data), Keyword(id=1156986838029259687, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, orderNo=3, keyword=target characteristic), Keyword(id=1156986838083785641, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, orderNo=4, keyword=generative adversarial network(GAN)), Keyword(id=1156986838150894507, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, orderNo=1, keyword=合成孔径雷达(SAR)), Keyword(id=1156986838209614765, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, orderNo=2, keyword=仿真数据), Keyword(id=1156986838268335023, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, orderNo=3, keyword=目标特性), Keyword(id=1156986838318666673, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, orderNo=4, keyword=生成对抗网络(GAN))], refs=[Reference(id=1156986842164842521, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2008, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[1], rfOrder=0, authorNames=宋建社, 郑永安, 袁礼海, journalName=null, refType=null, unstructuredReference=宋建社, 郑永安, 袁礼海. 合成孔径雷达图像理解与应用[M]. 北京: 科学出版社, 2008., articleTitle=合成孔径雷达图像理解与应用, refAbstract=null), Reference(id=1156986842215174171, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2008, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[1], rfOrder=1, authorNames=Song Jianshe, Zheng Yong'an, Yuan Lihai, journalName=null, refType=null, unstructuredReference=Song Jianshe, Zheng Yong'an, Yuan Lihai. Synthetic aperture radar image understanding and applications[M]. Beijing: Science Press, 2008., articleTitle=Synthetic aperture radar image understanding and applications, refAbstract=null), Reference(id=1156986842261311517, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=324, pageEnd=331, url=null, language=null, rfNumber=[2], rfOrder=2, authorNames=Liu Z, Yuan L, Weng L, journalName=null, refType=null, unstructuredReference=Liu Z, Yuan L, Weng L, et al. A high resolution opticalsatellite image dataset for ship recognition and some new baselines[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Shanghai: IEEE, 2017. 324-331., articleTitle=A high resolution opticalsatellite image dataset for ship recognition and some new baselines, refAbstract=null), Reference(id=1156986842328420383, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=3886, pageEnd=3890, url=null, language=null, rfNumber=[3], rfOrder=3, authorNames=Li C, Xu C, Cui Z, journalName=null, refType=null, unstructuredReference=Li C, Xu C, Cui Z, et al. Feature-attentioned object detection in remote sensing imagery[C]// IEEE International Conference on Image Processing (ICIP). New York: IEEE, 2019. 3886-3890., articleTitle=Feature-attentioned object detection in remote sensing imagery, refAbstract=null), Reference(id=1156986842395529249, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=3974, pageEnd=3983, url=null, language=null, rfNumber=[4], rfOrder=4, authorNames=Xia GS, Bai X, Ding J, journalName=null, refType=null, unstructuredReference=Xia GS, Bai X, Ding J, et al. DOTA: A large-scale dataset for object detection in aerial images[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018. 3974-3983., articleTitle=DOTA: A large-scale dataset for object detection in aerial images, refAbstract=null), Reference(id=1156986842450055203, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2015, volume=74, issue=null, pageStart=5263, pageEnd=5287, url=null, language=null, rfNumber=[5], rfOrder=5, authorNames=Kawano Y, Yanai K, journalName=Multimedia Tools and Applications, refType=null, unstructuredReference=Kawano Y, Yanai K. Foodcam: a real-time food recognition system on a smartphone[J]. Multimedia Tools and Applications, 2015. 74. 5263-5287., articleTitle=Foodcam: a real-time food recognition system on a smartphone, refAbstract=null), Reference(id=1156986842491998245, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2013, volume=35, issue=7, pageStart=1606, pageEnd=1621, url=null, language=null, rfNumber=[6], rfOrder=6, authorNames=Mumtaz A, Coviello E, Lanckriet GRG, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=Mumtaz A, Coviello E, Lanckriet GRG, et al. Clustering dynamic textures with the hierarchical EM algorithm for modeling video[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013. 35(7): 1606-1621., articleTitle=Clustering dynamic textures with the hierarchical EM algorithm for modeling video, refAbstract=null), Reference(id=1156986842563301415, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=39, pageEnd=54, url=null, language=null, rfNumber=[7], rfOrder=7, authorNames=Lewis B, Scarnati T, Sudkamp E, journalName=null, refType=null, unstructuredReference=Lewis B, Scarnati T, Sudkamp E, et al. A SAR dataset for ATR development: the synthetic and measured paired labeled experiment (SAMPLE)[C]// Algorithms for Synthetic Aperture Radar Imagery. New York: IEEE, 2019. 39-54., articleTitle=A SAR dataset for ATR development: the synthetic and measured paired labeled experiment (SAMPLE), refAbstract=null), Reference(id=1156986842609438761, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2024, volume=39, issue=2, pageStart=322, pageEnd=331, url=null, language=null, rfNumber=[8], rfOrder=8, authorNames=Shan JZ, Qu XJ, Chao K, journalName=Chinese Journal of Radio Science, refType=null, unstructuredReference=Shan JZ, Qu XJ, Chao K, et al. Efficient SAR image simulation for target on rough ground surface using the improved SBR technique[J]. Chinese Journal of Radio Science, 2024. 39(2): 322-331., articleTitle=Efficient SAR image simulation for target on rough ground surface using the improved SBR technique, refAbstract=null), Reference(id=1156986842663964716, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2021, volume=21, issue=17, pageStart=7159, pageEnd=7167, url=null, language=null, rfNumber=[9], rfOrder=9, authorNames=付松, 黄秀琼, 周龙建, journalName=科学技术与工程, refType=null, unstructuredReference=付松, 黄秀琼, 周龙建, 等. 粗糙地面上复杂目标的电磁散射特性[J]. 科学技术与工程, 2021. 21(17): 7159-7167., articleTitle=粗糙地面上复杂目标的电磁散射特性, refAbstract=null), Reference(id=1156986842714296366, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2021, volume=21, issue=17, pageStart=7159, pageEnd=7167, url=null, language=null, rfNumber=[9], rfOrder=10, authorNames=Fu Song, Huang Xiuqiong, Zhou Longjian, journalName=Science Technology and Engineering, refType=null, unstructuredReference=Fu Song, Huang Xiuqiong, Zhou Longjian, et al. Electromagnetic scattering characteristics of complex targets on rough ground[J]. Science Technology and Engineering, 2021. 21(17): 7159-7167., articleTitle=Electromagnetic scattering characteristics of complex targets on rough ground, refAbstract=null), Reference(id=1156986842768822321, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2022, volume=2022, issue=null, pageStart=7537732, pageEnd=null, url=null, language=null, rfNumber=[10], rfOrder=11, authorNames=Hu LP, Yao CM, Huang J, journalName=International Journal of Antennas and Propagation, refType=null, unstructuredReference=Hu LP, Yao CM, Huang J, et al. Highlyrobust synthetic aperture radar target recognition method based on simulation data training[J]. International Journal of Antennas and Propagation, 2022. 2022. 7537732., articleTitle=Highlyrobust synthetic aperture radar target recognition method based on simulation data training, refAbstract=null), Reference(id=1156986842827542579, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2019, volume=2019, issue=21, pageStart=8093, pageEnd=8097, url=null, language=null, rfNumber=[11], rfOrder=12, authorNames=Wang J, Li J, Sun B, journalName=The Journal of Engineering, refType=null, unstructuredReference=Wang J, Li J, Sun B, et al. SAR image synthesis based on conditional generative adversarial networks[J]. The Journal of Engineering, 2019. 2019(21): 8093-8097., articleTitle=SAR image synthesis based on conditional generative adversarial networks, refAbstract=null), Reference(id=1156986842898845749, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2017, volume=14, issue=9, pageStart=1484, pageEnd=1488, url=null, language=null, rfNumber=[12], rfOrder=13, authorNames=Malmgren-Hansen D, Kusk A, Dall J, journalName=IEEE Geoscience and remote sensing Letters, refType=null, unstructuredReference=Malmgren-Hansen D, Kusk A, Dall J, et al. Improving SAR automatic target recognition models with transfer learning from simulated data[J]. IEEE Geoscience and remote sensing Letters, 2017. 14(9): 1484-1488., articleTitle=Improving SAR automatic target recognition models with transfer learning from simulated data, refAbstract=null), Reference(id=1156986842957566007, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2022, volume=11, issue=6, pageStart=921, pageEnd=942, url=null, language=null, rfNumber=[13], rfOrder=14, authorNames=Xing MD, Xie YY, Gao YX, journalName=Journal of Radars, refType=null, unstructuredReference=Xing MD, Xie YY, Gao YX, et al. Electromagnetic scattering characteristic extraction and imaging recognition algorithm: a review[J]. Journal of Radars, 2022. 11(6): 921-942., articleTitle=Electromagnetic scattering characteristic extraction and imaging recognition algorithm: a review, refAbstract=null), Reference(id=1156986843024674874, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2023, volume=15, issue=14, pageStart=36, pageEnd=37, url=null, language=null, rfNumber=[14], rfOrder=15, authorNames=Wu K, Jin G W, Xiong X, journalName=Remote Sensing, refType=null, unstructuredReference=Wu K, Jin G W, Xiong X, et al. Fast SAR image simulation based on echo matrix cell algorithm including multiple scattering[J]. Remote Sensing, 2023. 15(14): 36-37., articleTitle=Fast SAR image simulation based on echo matrix cell algorithm including multiple scattering, refAbstract=null), Reference(id=1156986843091783739, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=2223, pageEnd=2232, url=null, language=null, rfNumber=[15], rfOrder=16, authorNames=Zhu JY, Park T, Isola P, journalName=null, refType=null, unstructuredReference=Zhu JY, Park T, Isola P, et al. Unpaired image-to-image translation using cycle-consistent adversarial networks[C]// Proceedings of the IEEE International Conference on Computer Vision. New York: IEEE, 2017. 2223-2232., articleTitle=Unpaired image-to-image translation using cycle-consistent adversarial networks, refAbstract=null), Reference(id=1156986843179864125, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2012, volume=7, issue=null, pageStart=e536, pageEnd=null, url=null, language=null, rfNumber=[16], rfOrder=17, authorNames=Iqbal N, Mumtaz R, Shafi U, journalName=PeerJ Computer Science, refType=null, unstructuredReference=Iqbal N, Mumtaz R, Shafi U, et al. Gray level co-occurrence matrix (GLCM) texture based crop classification using low altitude remote sensing platforms[J]. PeerJ Computer Science, 2012. 7. e536., articleTitle=Gray level co-occurrence matrix (GLCM) texture based crop classification using low altitude remote sensing platforms, refAbstract=null), Reference(id=1156986843242778687, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2000, volume=16, issue=2, pageStart=156, pageEnd=162, url=null, language=null, rfNumber=[17], rfOrder=18, authorNames=张静远, 张冰, 蒋兴舟, journalName=信号处理, refType=null, unstructuredReference=张静远, 张冰, 蒋兴舟. 基于小波变换的特征提取方法分析[J]. 信号处理, 2000. 16(2): 156-162., articleTitle=基于小波变换的特征提取方法分析, refAbstract=null), Reference(id=1156986843301498945, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2000, volume=16, issue=2, pageStart=156, pageEnd=162, url=null, language=null, rfNumber=[17], rfOrder=19, authorNames=Zhang Jingyuan, Zhang Bing, Jiang Xingzhou, journalName=Signal Processing, refType=null, unstructuredReference=Zhang Jingyuan, Zhang Bing, Jiang Xingzhou. Analysis of wavelet transform based feature extraction methods[J]. Signal Processing, 2000. 16(2): 156-162., articleTitle=Analysis of wavelet transform based feature extraction methods, refAbstract=null), Reference(id=1156986843368607810, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2012, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[18], rfOrder=20, authorNames=Feichtinger HG, Thomas S, journalName=null, refType=null, unstructuredReference=Feichtinger HG, Thomas S. Gabor analysis and algorithms: theory and applications[M]. Berlin: Springer International Publishing, 2012., articleTitle=Gabor analysis and algorithms: theory and applications, refAbstract=null), Reference(id=1156986843431522371, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2011, volume=21, issue=4, pageStart=1488, pageEnd=1499, url=null, language=null, rfNumber=[19], rfOrder=21, authorNames=Brunet D, Edward RV, Zhou W, journalName=IEEE Transactions on Image Processing, refType=null, unstructuredReference=Brunet D, Edward RV, Zhou W. On the mathematical properties of the structural similarity index[J]. IEEE Transactions on Image Processing, 2011. 21(4): 1488-1499., articleTitle=On the mathematical properties of the structural similarity index, refAbstract=null), Reference(id=1156986843481854021, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, doi=null, pmid=null, pmcid=null, year=2014, volume=arXiv, issue=null, pageStart=1409, pageEnd=1556, url=null, language=null, rfNumber=[20], rfOrder=22, authorNames=Simonyan K, Zisserman A, journalName=arXiv Preprint, refType=null, unstructuredReference=Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. arXiv Preprint, 2014. arXiv. 1409-1556., articleTitle=Very deep convolutional networks for large-scale image recognition, refAbstract=null)], funds=[Fund(id=1156986842059984918, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, awardId=62271116, language=CN, fundingSource=国家自然科学基金(62271116), fundOrder=null, country=null)], companyList=[AuthorCompany(id=1156986832723464927, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, xref=1, ext=[AuthorCompanyExt(id=1156986832731853536, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986832723464927, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Engineering Chengdu College of University of Electronic Science and Technology of China Chengdu 611731 China), AuthorCompanyExt(id=1156986834921280311, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986832723464927, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 电子科技大学成都学院 工学院 成都 611731)]), AuthorCompany(id=1156986834988389179, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, xref=2, ext=[AuthorCompanyExt(id=1156986834992583483, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986834988389179, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 School of Information and Communication Engineering University of Electronic Science and Technology of China Chengdu 611731 China), AuthorCompanyExt(id=1156986835000972092, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, companyId=1156986834988389179, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 电子科技大学 信息与通信工程学院 成都 611731)])], figs=[ArticleFig(id=1156986839610512332, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 1, caption=CycleGAN network architecture, figureFileSmall=sTnG4zgtVzgDsgk8v2KqzA==, figureFileBig=xI/8fU0KPQKk6tVCMvKNrw==, tableContent=null), ArticleFig(id=1156986839686009807, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图1, caption=CycleGAN 网络结构图, figureFileSmall=sTnG4zgtVzgDsgk8v2KqzA==, figureFileBig=xI/8fU0KPQKk6tVCMvKNrw==, tableContent=null), ArticleFig(id=1156986839757312978, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 2, caption=Generator network architecture, figureFileSmall=vncvrOFOIoUKJ8QnE9ArPA==, figureFileBig=EJWf5l8vnTsP32z1dri+ig==, tableContent=null), ArticleFig(id=1156986839820227540, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图2, caption=生成器网络架构, figureFileSmall=vncvrOFOIoUKJ8QnE9ArPA==, figureFileBig=EJWf5l8vnTsP32z1dri+ig==, tableContent=null), ArticleFig(id=1156986839870559190, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 3, caption=Residual module structure, figureFileSmall=FKQZiD9PGth0fKuNznl2Fw==, figureFileBig=g6b1uCI01KsfISisIp9M7g==, tableContent=null), ArticleFig(id=1156986839941862360, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图3, caption=残差模块结构图, figureFileSmall=FKQZiD9PGth0fKuNznl2Fw==, figureFileBig=g6b1uCI01KsfISisIp9M7g==, tableContent=null), ArticleFig(id=1156986840013165530, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 4, caption=Discriminator network architectures, figureFileSmall=qPWzasEK98LCWhN640K46Q==, figureFileBig=FAb2YoA+vgTZO0p+YINx4Q==, tableContent=null), ArticleFig(id=1156986840076080092, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图4, caption=判别器网络架构, figureFileSmall=qPWzasEK98LCWhN640K46Q==, figureFileBig=FAb2YoA+vgTZO0p+YINx4Q==, tableContent=null), ArticleFig(id=1156986840134800350, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 5, caption=Schematic diagram of the CycleGANdouble-loop process, figureFileSmall=pOpb0oJlHcXv6D4J/uzrmA==, figureFileBig=51EgB8DtGjUr+cD22fxCGw==, tableContent=null), ArticleFig(id=1156986840189326304, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图5, caption=CycleGAN 双循环过程示意图, figureFileSmall=pOpb0oJlHcXv6D4J/uzrmA==, figureFileBig=51EgB8DtGjUr+cD22fxCGw==, tableContent=null), ArticleFig(id=1156986840294183908, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 6, caption=Structure of the measurement system for structure similarity index measure, figureFileSmall=KgLSYCpfkYIQgqDWuQXOMA==, figureFileBig=GDDyveQ22oO6B7+YkYdBiQ==, tableContent=null), ArticleFig(id=1156986840357098470, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图6, caption=结构相似度衡量指标的测量系统结构, figureFileSmall=KgLSYCpfkYIQgqDWuQXOMA==, figureFileBig=GDDyveQ22oO6B7+YkYdBiQ==, tableContent=null), ArticleFig(id=1156986840420013032, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 7, caption=Optical morphology of ten MSTAR target classes and their corresponding SAR imaging morphology, figureFileSmall=C549FkgLvZXVWqJreA2dPw==, figureFileBig=7meeUmlfEI2MoEblSaVDsg==, tableContent=null), ArticleFig(id=1156986840474538986, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图7, caption=MSTAR 10 类目标的光学形态及其对应的 SAR 成像形态, figureFileSmall=C549FkgLvZXVWqJreA2dPw==, figureFileBig=7meeUmlfEI2MoEblSaVDsg==, tableContent=null), ArticleFig(id=1156986840533259244, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 8, caption=Optical morphology of ten MSTAR target classes and their corresponding SAR imaging morphology, figureFileSmall=uKdo8j9DRSHuaWXANkA88w==, figureFileBig=B2x9c5DybEGXPeQr5/Xk1Q==, tableContent=null), ArticleFig(id=1156986840591979502, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图8, caption=MSTAR 10 类目标的光学形态及其对应的 SAR 成像形态, figureFileSmall=uKdo8j9DRSHuaWXANkA88w==, figureFileBig=B2x9c5DybEGXPeQr5/Xk1Q==, tableContent=null), ArticleFig(id=1156986840642311152, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 9, caption=Comparison of results from measured SAR images, simulated images, CycleGAN optimization results, and the proposed method optimization results, figureFileSmall=C4U/rWZ4L7jJ5NRdiCzAlw==, figureFileBig=FNu5Ukp66q9VxrvFg5wARQ==, tableContent=null), ArticleFig(id=1156986840726197233, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图9, caption=实测、仿真、CycleGAN 优化以及本文方法结果对比图, figureFileSmall=C4U/rWZ4L7jJ5NRdiCzAlw==, figureFileBig=FNu5Ukp66q9VxrvFg5wARQ==, tableContent=null), ArticleFig(id=1156986840797500402, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 10, caption=Comparison of gray scale histograms of measured samples and generated samples for the same class targets, figureFileSmall=n9BQ0PkJ9J85lc+Nlinxlw==, figureFileBig=C2+n8IIHN/5kmr5hxdgHiA==, tableContent=null), ArticleFig(id=1156986840864609268, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图10, caption=同类目标实测样本与生成样本灰度直方图对比, figureFileSmall=n9BQ0PkJ9J85lc+Nlinxlw==, figureFileBig=C2+n8IIHN/5kmr5hxdgHiA==, tableContent=null), ArticleFig(id=1156986840914940918, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 11, caption=Histogram similarity index statistical results, figureFileSmall=BO3kq8koE7RqbM+krl106w==, figureFileBig=XEiM6jKbmOrhRLlo83ko3g==, tableContent=null), ArticleFig(id=1156986840982049784, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图11, caption=直方图相似度指数统计结果, figureFileSmall=BO3kq8koE7RqbM+krl106w==, figureFileBig=XEiM6jKbmOrhRLlo83ko3g==, tableContent=null), ArticleFig(id=1156986841070130170, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 12, caption=Recognition performance test results for three types of small sample datasets in table 2, figureFileSmall=Q4/QNlzbnFcsLlBdtJx9Xg==, figureFileBig=g4HpAcNmvomTLN1HC3nZqA==, tableContent=null), ArticleFig(id=1156986841141433341, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图12, caption=表2 中完全小样本数据集识别性能测试结果, figureFileSmall=Q4/QNlzbnFcsLlBdtJx9Xg==, figureFileBig=g4HpAcNmvomTLN1HC3nZqA==, tableContent=null), ArticleFig(id=1156986841200153599, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Fig. 13, caption=Results of recognition performance tests on selected small sample datasets in table 3, figureFileSmall=gAMNooAFfgUfD39u4ht91w==, figureFileBig=UwRusyoL7K1PFFKWhtH51g==, tableContent=null), ArticleFig(id=1156986841267262465, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=图13, caption=表3 中部分小样本数据集识别性能测试结果, figureFileSmall=gAMNooAFfgUfD39u4ht91w==, figureFileBig=UwRusyoL7K1PFFKWhtH51g==, tableContent=null), ArticleFig(id=1156986841351147523, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Table 1, caption=Image optimization training set composition, figureFileSmall=null, figureFileBig=null, tableContent=
训练子集 类别 数量
Trainset A BMP2 仿真 108
BTR70 仿真 96
T72 仿真 110
合计 314
Trainset B BMP2 10
BTR70 10
T72 10
其他七类 420
合计 450
), ArticleFig(id=1156986841422450693, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=表1, caption=图像优化训练集构成, figureFileSmall=null, figureFileBig=null, tableContent=
训练子集 类别 数量
Trainset A BMP2 仿真 108
BTR70 仿真 96
T72 仿真 110
合计 314
Trainset B BMP2 10
BTR70 10
T72 10
其他七类 420
合计 450
), ArticleFig(id=1156986841481170952, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Table 2, caption=Experimental dataset based on a completely small sample of scenarios, figureFileSmall=null, figureFileBig=null, tableContent=
目标类别 目标型号 训练样本数 测试样本数 俯仰角/(°)
训练 测试
BMP2 9563 10 196 17 15
BTR70 c71 10 196 17 15
T72 con-132 10 196 17 15
), ArticleFig(id=1156986841569251337, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=表2, caption=基于完全小样本场景的试验数据集, figureFileSmall=null, figureFileBig=null, tableContent=
目标类别 目标型号 训练样本数 测试样本数 俯仰角/(°)
训练 测试
BMP2 9563 10 196 17 15
BTR70 c71 10 196 17 15
T72 con-132 10 196 17 15
), ArticleFig(id=1156986841657331724, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Table 3, caption=Experimental dataset based on selected small sample scenarios, figureFileSmall=null, figureFileBig=null, tableContent=
目标类别 目标序列号 训练样 本数 测试样 本数 目标俯仰角度/(${}^{\circ }$)
训练 测试
2S1 bol 100 274 17 15
BMP2 9563 10 195 17 15
BRDM2 E71 100 274 17 15
BTR60 K10yt7532 100 195 17 15
BTR70 c71 10 196 17 15
D7 92v13015 100 274 17 15
T62 A51 100 273 17 15
T72 con-132 10 196 17 15
ZIL131 E12 100 274 17 15
ZSU23-4 d08 100 274 17 15
), ArticleFig(id=1156986841732829198, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=表3, caption=基于部分小样本场景的试验数据集, figureFileSmall=null, figureFileBig=null, tableContent=
目标类别 目标序列号 训练样 本数 测试样 本数 目标俯仰角度/(${}^{\circ }$)
训练 测试
2S1 bol 100 274 17 15
BMP2 9563 10 195 17 15
BRDM2 E71 100 274 17 15
BTR60 K10yt7532 100 195 17 15
BTR70 c71 10 196 17 15
D7 92v13015 100 274 17 15
T62 A51 100 273 17 15
T72 con-132 10 196 17 15
ZIL131 E12 100 274 17 15
ZSU23-4 d08 100 274 17 15
), ArticleFig(id=1156986841820909585, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=EN, label=Table 4, caption=Main hardware configurations and software versions of the testenvironment, figureFileSmall=null, figureFileBig=null, tableContent=
硬件/软件 型号/版本
GPU 型号 RTX3060
系统环境 Windows10 64 位
Visual Studio 版本 VS2016
Python 版本 Python3.6
CUDA 版本 11.1
Pytorch 版本 1.80
), ArticleFig(id=1156986841875435538, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1156963928124253027, language=CN, label=表4, caption=试验环境的主要硬件配置和软件版本, figureFileSmall=null, figureFileBig=null, tableContent=
硬件/软件 型号/版本
GPU 型号 RTX3060
系统环境 Windows10 64 位
Visual Studio 版本 VS2016
Python 版本 Python3.6
CUDA 版本 11.1
Pytorch 版本 1.80
)], attaches=null, journal=Journal(id=1146119176004939786, delFlag=0, nameCn=科学技术与工程, nameEn=Science Technology and Engineering, nameHistory1=null, nameHistory2=null, issn=1671-1815, eissn=, cn=11-4688/T, coden=null, periodic=4, language=CN, oaType=是, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=UKU/O7GSka5polgCTkbIIw==, journalPrice=null, startedYear=null, abbrevIsoEn=Sci Technol Eng, journalRemark=null, publicationField=null, createdTime=null, updatedTime=1754445529766, createdBy=null, updatedBy=13701087609, firstLetterCn=S, firstLetterEn=S, subjectCode=Natural Sciences, subjectName=自然科学, subjectCodeEn=Natural Sciences, subjectNameEn=null, picCn=UKU/O7GSka5polgCTkbIIw==, picEn=5hwlULoNwcbj3xUmVi9MAQ==, jcr=null, cjcr=null, exts=[JournalExt(id=1159791870395564357, language=CN, name=科学技术与工程, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=http://www.stae.com.cn/jsygc/home, createdTime=1754445529793, updatedTime=1754445529793, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=http://www.stae.com.cn/jsygc/site/menus/20090429150146001, submissionAuthorUrl=http://www.stae.com.cn/jsygc/author/login, submissionEditorUrl=http://www.stae.com.cn/jsygc/editor/login, submissionReviewUrl=http://www.stae.com.cn/jsygc/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1159791870441701702, language=EN, name=Science Technology and Engineering, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=http://www.stae.com.cn/jsygc/home, createdTime=1754445529804, updatedTime=1754445529804, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.stae.com.cn/jsygc/author/login, submissionEditorUrl=http://www.stae.com.cn/jsygc/editor/login, submissionReviewUrl=http://www.stae.com.cn/jsygc/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1146123166801305609, websiteList=[Website(id=1148243202391400884, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146123166801305609, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/kxjsygc/CN, language=CN, createTime=1751692112777, createBy=18614031015, updateTime=1753520965431, updateBy=18614031015, name=科学技术与工程-中文站点, tplId=1146099689490845704, title=科学技术与工程, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1148622798802673703, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=articleTextType, value=kx, createTime=1751782615614, updateTime=1751782615614, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798781702180, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=banner, value=null, createTime=1751782615609, updateTime=1751782615609, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798769119267, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=j86gbwi+p0Idkyl5SzIlmQ==, createTime=1751782615606, updateTime=1751782615606, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798794285094, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1751782615612, updateTime=1751782615612, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798790090789, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1751782615611, updateTime=1751782615611, creator=18614031015, updator=18614031015)]), Website(id=1155914124811976731, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146123166801305609, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/kxjsygc/EN, language=EN, createTime=1753521003206, createBy=18614031015, updateTime=1753521003206, updateBy=18614031015, name=科学技术与工程-英文站点, tplId=1146101810881728533, title=Science Technology and Engineering, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1155914371227308235, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=articleTextType, value=kx, createTime=1753521061952, updateTime=1753521061952, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371210531016, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=banner, value=null, createTime=1753521061947, updateTime=1753521061947, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371202142407, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=j86gbwi+p0Idkyl5SzIlmQ==, createTime=1753521061945, updateTime=1753521061945, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371223113930, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1753521061950, updateTime=1753521061950, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371218919625, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1753521061949, updateTime=1753521061949, creator=18614031015, updator=18614031015)])], journalTitle=科学技术与工程, weixinUrl=null, journalUrl=null, iacademicId=null, status=0, seqNo=null, journalTitleEn=Science Technology and Engineering, journalPhotoCn=UKU/O7GSka5polgCTkbIIw==, journalPhotoEn=5hwlULoNwcbj3xUmVi9MAQ==, journalFirstLetter=S, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=null, provinceCode=null, provinceName=null, collectFlag=false), detailUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2403736, detailUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/10.12404/j.issn.1671-1815.2403736, pdfUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/PDF/10.12404/j.issn.1671-1815.2403736, pdfUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/PDF/10.12404/j.issn.1671-1815.2403736, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
基于目标特性约束的SAR 仿真数据优化
收藏切换
PDF下载
张慧 1 , 牟立强 2 , 李沂蔚 2 , 崔宗勇 2, *
科学技术与工程 | 电子技术、通信技术 2025,25(8): 3268-3279
收起
收藏切换
科学技术与工程 | 电子技术、通信技术 2025, 25(8): 3268-3279
基于目标特性约束的SAR 仿真数据优化
全屏
张慧1 , 牟立强2, 李沂蔚2, 崔宗勇2, *
作者信息
  • 1 电子科技大学成都学院 工学院 成都 611731
  • 2 电子科技大学 信息与通信工程学院 成都 611731
  • 张慧(1984-),女,汉族,山东菏泽人,硕士,副教授。研究方向:模式识别、人工智能。E-mail: zh21y@163.com。

通讯作者:

*崔宗勇(1984—),男,汉族,山东菏泽人,博士,副教授。研究方向:SAR图像解译。E-mail: zycui@uestc.edu.cn。
SAR Simulation Data Optimization Based on Target Characteristic Constraints
Hui ZHANG1 , Li-qiang MOU2, Yi-wei LI2, Zong-yong CUI2, *
Affiliations
  • 1 School of Engineering Chengdu College of University of Electronic Science and Technology of China Chengdu 611731 China
  • 2 School of Information and Communication Engineering University of Electronic Science and Technology of China Chengdu 611731 China
出版时间: 2025-03-18 doi: 10.12404/j.issn.1671-1815.2403736
文章导航
收藏切换

基于深度网络的合成孔径雷达(synthetic aperture radar, SAR)目标识别方法,需要大量训练数据,而实际应用中 SAR 成像系统获取数量充足、分布均匀的目标数据难度极高。解决SAR目标识别小样本问题的途径之一,就是采用电磁仿真技术生成大量 SAR仿真数据。然而仿真图像与实测 SAR图像仍存在较大差异,直接使用仿真数据并不能带来目标识别性能的显著提升。针对上述问题,提出一种基于SAR目标特性约束的仿真数据优化方法。该方法在分析 SAR目标特性的基础上,构建基于纹理结构-循环一致性的生成对抗网络(texture structure cycle-consistent generative adversarial network, TS-CycleGAN),采用结构相似度衡量指标约束 CycleGAN的生成过程,降低仿真数据与实测数据之间的差异,提高仿真数据的可用性。针对SAR SAMPLE 数据集的试验结果表明,所提出的方法相较于其他仿真数据优化方法,在图像质量评估和分类性能方面,都取得了明显的提升。

合成孔径雷达(SAR)  /  仿真数据  /  目标特性  /  生成对抗网络(GAN)

Synthetic aperture radar (SAR) target recognition method based on deep networks requires a large amount of training data, and in practical applications, it is extremely difficult for SAR imaging systems to obtain sufficient and evenly distributed target data. One way to solve the small sample problem in SAR target recognition, is to use electromagnetic simulation technology to generate a large amount of SAR simulation data. However, there are still significant differences between simulated images and measured SAR images, so using simulation data directly cannot bring significant performance improvement for target recognition. A simulation data optimization method based on SAR target characteristic constraints was proposed to address the above issues. On the basis of analyzing the characteristics of SAR targets, a texture structure cycle-consistent generative adversarial network (TS-CycleGAN) based on texture structure and cycle consistency was constructed, in which the structural similarity measure was used to constrain the generation process of CycleGAN. This method can reduce the difference between simulation data and measured data, and can improve the usability of simulation data. The experimental results on the SAR SAMPLE dataset show that, compared to other simulation data optimization methods, the proposed method achieves significant improvements in image quality evaluation and classification performance.

synthetic aperture radar (SAR)  /  simulation data  /  target characteristic  /  generative adversarial network(GAN)
张慧, 牟立强, 李沂蔚, 崔宗勇. 基于目标特性约束的SAR 仿真数据优化. 科学技术与工程, 2025 , 25 (8) : 3268 -3279 . DOI: 10.12404/j.issn.1671-1815.2403736
Hui ZHANG, Li-qiang MOU, Yi-wei LI, Zong-yong CUI. SAR Simulation Data Optimization Based on Target Characteristic Constraints[J]. Science Technology and Engineering, 2025 , 25 (8) : 3268 -3279 . DOI: 10.12404/j.issn.1671-1815.2403736
合成孔径雷达 (synthetic aperture radar, SAR) 目标识别作为能够从雷达图像中提取目标位置、类别、型号等信息的关键技术, 是提升 SAR 系统信息感知能力的核心关键。深度网络可以通过训练数据自动学习到 SAR 图像中目标的特征和规律并对目标进行识别, 在 SAR 目标识别领域有着广泛应用并逐渐成为主流方向[1]。但是深度学习方法严重依赖于训练样本数据库的完备性。当训练样本数量不充足的情况下, 现有 SAR 目标识别技术无法深入挖掘目标的深层次特征,容易出现过拟合现象。然而, 现实中战场形势错综复杂、环境复杂多变, 军事目标往往极难被大量探测到, 形成完备的样本数据集十分困难, 因此 SAR 目标识别可以看作一个小样本识别问题。
近年来, 许多研究者尝试通过他源信息辅助的方式来提升系统的识别性能和可靠性, 解决 SAR 目标识别的小样本问题。他源信息的来源广泛, 光学系统成像图像[2-4] 、红外成像数据[5-6] 、电磁建模仿真图像数据[7-10] 等方法。但是由于军事目标本身难以被探测的原因,其光学成像和红外成像图像同样存在数量缺少的问题。
电磁仿真技术能够在不受外界条件影响的情况下, 人工生成大量优质仿真数据, 并且能够对仿真过程的俯仰角、目标角度等参数进行精确控制和调整。然而电磁建模仿真的过程与实际成像过程虽然类似, 但电磁波在实际传播过程中会与背景中的物体 (如草地、灌木、石块等) 相互作用,此过程难以通过简单的模型进行仿真模拟; 同时现实成像过程中的大气条件、天线状态等不确定性极高的偶然因素也会对回波数据的噪声产生影响, 从而限制背景仿真结果的真实性。因此, 电磁仿真图像与实测 SAR 图像仍存在较大差异,导致其难以直接应用于识别的训练过程, 需要将其优化为与实测数据分布更加接近的数据, 才能够用于网络训练以提高识别性能。
针对 MSTAR 数据集中的实测图像和 SAMPLE 数据集中的电磁仿真图像,文献 [7] 采用 t-SNE 对二者的特征空间进行绘图, 发现它们之间存在巨大差异。Wang 等[11] 将电磁模拟与条件生成对抗网络 (conditional generative adversarial networks, cGAN) 相结合,使用电磁仿真建模得到的图像作为 cGANs 的输入, 然后利用生成器生成包含标签信息的逼真的 SAR 图像,以此来提升目标识别率。Malmgren-Hansen 等[12] 研究了仿真数据与实测 SAR 图像的可转移性, 通过在聚焦之前将目标雷达反射率添加到单个点散射体的地形模型中来获取仿真数据, 发现采用仿真数据来预先训练卷积神经网络比只在实测数据上训练,收敛更快且在 MSTAR 上进行基准测试时精度更高。
以上方法虽然可以在一定程度上减小 SAR 仿真数据与实测数据之间的差异, 但是其关注点主要集中于目标本身, 未充分考虑背景的作用以及目标与背景的关联性,导致识别性能提升效果有${\mathbb{R}}^{\left\lbrack {13}- {14}\right\rbrack }$。鉴于此,提出基于目标特性约束的 SAR 仿真数据优化方法, 能够在保持目标特性的前提下, 减小电磁仿真 SAR 图像的背景纹理结构特性与实测 SAR 图像之间的差异, 从而实现 SAR 仿真数据在目标识别任务中的有效利用。
在 CycleGAN 网络基本架构与原理的基础上, 分析基于结构相似性度量的 SAR 目标纹理特性描述,构建纹理结构损失函数,约束 CycleGAN 的生成过程, 实现 SAR 仿真数据的优化。
CycleGAN[15] 是一种基于生成对抗网络的深度学习网络模型, 本质上由两个镜像对称的生成对抗网络组成, 如图1 所示。每个生成对抗网络都由生成器和判别器两个部分组成, 二者在网络迭代过程中不断优化, 在相互博弈中促进提升。
CycleGAN 的生成器包含:编码器、转换器和解码器,结构如图2 所示。
首先,输入图像被传递到编码器,用于提取输入图像的特征。编码器利用卷积神经网络从输入图像中提取特征。然后, 将经过编码器激活函数后的输出传递给转换器, 其中包含 6 层残差模块, 如图3 所示。此结构通过组合图像的不相近特征, 将图像在$X$ 域中的特征向量转换为$Y$ 域中的特征向量。最后,将转换器的输出输入解码器,进行特征还原。解码器利用两个反卷积层从特征向量中还原出低级特征, 最后将图像扩大到原始大小, 最后得到生成图像。
C7s1-64 为卷积核为$7 * 7$、步长为 1 、输出通道数为 64 的卷积层;$\mathrm{d}{128}$$\mathrm{d}{256}$ 分别为输出通道数为 128 和 256、步幅为 2 的 3 * 3 卷积层;R256 中包含两个 3 * 3 卷积层,它们的滤波器数量相同, 均为 256 转换器; u128 和 u64 分别为输出通道数为 128 和 64 的反卷积层; C7 sl -3 为卷积核为$7 * 7$、步长为 1 、输出通道数为 3 的卷积层
判别器的输入为生成器生成的图片, 它将生成图片与实测图片进行比较并将两类图片进行分类, 在 CycleGAN 网络模型中使用马尔可夫判别器 (PatchGAN)。判别器网络结构如图4 所示。
$\mathrm{C}k$ 表示卷积核大小为$4 * 4$,输出通道数为$k$ 个且步长为 2 的卷积层
CycleGAN 两个生成器为$G$$F$,它们的映射函数可表示为
$\left\{\begin{array}{l} G : X \rightarrow Y \\ F : Y \rightarrow X \end{array}\right.$
式 (1) 中:$X$ 为输入图像分布;$Y$ 为期望的输出图像分布。
CycleGAN 的损失函数包含两方面:用于将生成图像的分布与目标域中的数据分布匹配的对抗性损失 (adversarial loss); 防止学习到的映射$G$$F$ 相互矛盾的循环一致性损失(cycle consistency loss)。
(1)对抗性损失${L}_{\mathrm{{GAN}}}$。将对抗性损失应用于两种映射。对于映射函数$G : X \rightarrow Y$ 和它的判别器${D}_{Y}$, 损失函数为
${L}_{G}\left({G,{D}_{Y}, X, Y}\right)= {E}_{y \sim {p}_{\text{data }}\left(y\right)}\left\lbrack {\lg {D}_{Y}\left(y\right)}\right\rbrack +\\{E}_{x \sim {p}_{\text{data }}\left(x\right)}\left({\lg \left\{{1 -{D}_{Y}\left\lbrack {G\left(x\right)}\right\rbrack }\right\}}\right)$
式 (2) 中:${E}_{y \sim {p}_{\text{data }}\left(y\right)}$ 为目标域$Y$ 中样本$y$ 服从实测 SAR 图像分布${p}_{\text{data }}\left(y\right)$ 的期望;${E}_{x \sim {p}_{\text{data }}\left(x\right)}$ 为源域$X$ 中样本$x$ 服从电磁仿真图像分布${p}_{\text{data }}\left(x\right)$ 的期望;$G$ 生成与$Y$ 域中图像相似的图像$G\left(x\right)$,而${D}_{Y}$ 则用来区分生成样本$G\left(x\right)$ 和实测样本$y \circ G$ 的目标是最小化损失函数如式 (2) 所示,而${D}_{Y}$ 则需要最大化${L}_{G}$,即
$\mathop{\min }\limits_{G}\mathop{\max }\limits_{{D}_{Y}}{L}_{G}\left({G,{D}_{Y}, X, Y}\right)$
同理,对映射函数$F : Y \rightarrow X$ 及其判别器${D}_{X}$ 也引入类似对抗损失, 可表示为
${L}_{F}\left({F,{D}_{X}, X, Y}\right)= {E}_{x \sim {p}_{\text{data }}\left(x\right)}\left\lbrack {\lg {D}_{X}\left(x\right)}\right\rbrack +\\{E}_{y \sim {p}_{\text{data }}\left(y\right)}\left({\lg \left\{{1 -{D}_{X}\left\lbrack {F\left(y\right)}\right\rbrack }\right\}}\right)$
最终的目的为
$\mathop{\min }\limits_{F}\mathop{\max }\limits_{{D}_{X}}{L}_{F}\left({F,{D}_{X}, Y, X}\right)$
综上所述, 对抗性损失为
${L}_{\mathrm{{GAN}}}= {L}_{G}\left({G,{D}_{Y}, X, Y}\right)+ {L}_{F}\left({F,{D}_{X}, X, Y}\right)\\= {E}_{y \sim {p}_{\text{data }}\left(y\right)}\left\lbrack {\lg {D}_{Y}\left(y\right)}\right\rbrack +\\{E}_{x \sim {p}_{\text{data }}\left(x\right)}\left({\lg \left\{{1 -{D}_{Y}\left\lbrack {G\left(x\right)}\right\rbrack }\right\}}\right)+ \\{E}_{x \sim {p}_{\text{data }}\left(x\right)}\left\lbrack {\lg {D}_{X}\left(x\right)}\right\rbrack +\\{E}_{y \sim {p}_{\text{data }}\left(y\right)}\left({\lg \left\{{1 -{D}_{X}\left\lbrack {F\left(y\right)}\right\rbrack }\right\}}\right)$
(2)循环一致性损失${L}_{\text{cyc }}$。对抗性损失函数是针对生成器$G$$F$ 分别进行约束的,为了促使二者彼此保持一致, CycleGAN 提出了循环一致性损失使其网络结构成为一个整体, 其主要包含两个有向循环, 如图5 所示。
$X$ 为电磁仿真图像;$Y$$X$ 经过生成器$G$ 映射后校正为的 SAR 真实图像;$X$$Y$ 经过生成器$F$ 映射被转换为的电磁仿真图像;$\widehat{x}$ 为生成器$F$ 映射过程;$\widehat{y}$ 为生成器$G$ 映射过程
两种循环分别被称为前向循环一致性 Forward 和后向循环一致性 Backward, 可表示为
$\left\{\begin{array}{l}\text{ Forward:}x \rightarrow G\left(x\right)\rightarrow F\left\lbrack {G\left(x\right)}\right\rbrack \approx x \\\text{ Backward:}y \rightarrow F\left(y\right)\rightarrow G\left\lbrack {F\left(y\right)}\right\rbrack \approx y \end{array}\right.$
由式 (7) 可知, 前向循环一致性意图使$F\left\lbrack {G\left(x\right)}\right\rbrack$$x$ 保持相似,后向循环一致性则使$G\left\lbrack {F\left(y\right)}\right\rbrack$$y$ 相似。因此,循环一致性损失函数可表示为
${L}_{\text{cyc }}\left({G, F}\right)= {E}_{x \sim {p}_{\text{data }}\left(x\right)}\left\{{\parallel F\left\lbrack {G\left(x\right)}\right\rbrack - x{\parallel }_{1}}\right\}+ \\{E}_{y \sim {p}_{\text{data }}\left(y\right)}\left\{{\parallel G\left\lbrack {F\left(y\right)}\right\rbrack - y{\parallel }_{1}}\right\}$
(3)同一性损失${L}_{\text{identity }}$。CycleGAN 引入一个额外的同一性损失, 它的作用是保留输入和输出图像色调组成的一致性, 可表示为
${L}_{\text{identity }}\left({G, F}\right)= {E}_{y \sim {p}_{\text{data }}\left(y\right)}\left\lbrack {\parallel G\left(y\right)- y{\parallel }_{1}}\right\rbrack +\\{E}_{x \sim {p}_{{\mathrm{{data}}}^{\left(x\right)}}}\left\lbrack {\parallel F\left(x\right)- x{\parallel }_{1}}\right\rbrack $
因此,结合式 (6)、式 (8) 和式 (9), 最终的损失函数为
$ L\left({G, F,{D}_{X},{D}_{Y}}\right)= {L}_{\text{GAN }}+ {L}_{\text{cycle }}+ {L}_{\text{identity }}\\= {L}_{\mathrm{G}}\left({G,{D}_{Y}, X, Y}\right)+ \\{L}_{F}\left({F,{D}_{X}, Y, X}\right)+ {\lambda }_{1}{L}_{\text{cyc }}\left({G, F}\right)+ \\{\lambda }_{2}{L}_{\text{identity }}\left({G, F}\right)$
式 (10) 中:${\lambda }_{1}$${\lambda }_{2}$ 分别为${L}_{\text{cyc }}$${L}_{\text{identity }}$ 的权重参数, 该参数数值越小, 对应损失函数对网络的约束越弱, 当处于 0 时, 此损失函数不发挥作用。
CycleGAN 的目标函数可表示为
${G}^{* },{F}^{* }= \arg \mathop{\min }\limits_{{G, F}}\mathop{\max }\limits_{{{D}_{X},{D}_{Y}}}L\left({G, F,{D}_{X},{D}_{Y}}\right)$
式 (11) 中:${G}^{* }\text{、}{F}^{* }$ 为通过网络训练得到的最优映射。
图像的纹理结构特征与对比度、亮度、结构等多方面因素相关, 这些因素可以通过许多质量评估方法判断,灰度共生矩阵[16] 、小波变换[17] 、Gabor 滤波[18] 等,但他们多用于对单一图像进行评判,过程中不会涉及到两幅图片的相关性。但是, 结构相似度衡量指标[19] (structure similarity index measure, SSIM)能够将两幅图片基于亮度 (luminance)、对比度 (contrast) 和结构 (structure) 进行比较, 这些因素可以反映出图像纹理结构特征。
SAR 图像中,目标物及背景的亮度信息与其自身的结构和散射系数相关, 不同材质的亮度表现不同, 因而导致其对应位置的纹理结构特征不同。对比度是图像中灰度值的变化程度, 对比度越大则灰度差越明显,图片表现出的纹理结构也更加清晰。结构是指图像中出亮度和对比度之外的高频信息, 包括局部灰度变化频率、线条、边缘特征等。亮度、 对比度、结构三者均与图像纹理结构特征相关, 共同组成结构相似度衡量指标。因此, 本文方法利用结构相似度衡量指标来描述 CycleGAN 网络训练过程中的生成图像与实测图像的纹理结构相似度, 并将此作为改进网络模型的约束方法。结构相似度衡量指标的测量系统结构如图6 所示。
对于图像由离散信号组成的数据,亮度指标${\mu }_{x}$ 用图像平均灰度值来衡量, 通过平均所有像素的灰度值得到, 即
${\mu }_{x}= \frac{1}{N}\mathop{\sum }\limits_{{i = 1}}^{N}{x}_{i}$
式 (12) 中:$N$ 为图像中像素点的数量;${x}_{i}$ 为第$i$ 个点的灰度值。
亮度对比函数$l\left({x, y}\right)$ 为关于两幅图像亮度指标${\mu }_{x}$${\mu }_{y}$ 的函数,可表示为
$ l\left({x, y}\right)= \frac{2{\mu }_{x}{\mu }_{y}+ {C}_{1}}{{\mu }_{x}^{2}+ {\mu }_{y}^{2}+ {C}_{1}}$
式 (13) 中:${C}_{1}$ 为常数,设置此常数的目的是为了避免当${\mu }_{x}^{2}+ {\mu }_{y}^{2}$ 接近 0 时会造成系统不稳定的现象发生。
对比度指标$\sigma$ 通过灰度标准差来衡量。其计算方式首先通过将平均灰度值从图像中去除, 得到${x}_{i}- {\mu }_{x}$,再使用标准差作为估量值,即
${\sigma }_{x}= {\left\lbrack \frac{1}{N - 1}\mathop{\sum }\limits_{{i = 1}}^{N}{\left({x}_{i}- {\mu }_{x}\right)}^{2}\right\rbrack }^{\frac{1}{2}}$
对比度对比函数$c\left({x, y}\right)$ 是关于两幅图片对比度指标${\sigma }_{x}$${\sigma }_{y}$ 的函数,可表示为
$ c\left({x, y}\right)= \frac{2{\sigma }_{x}{\sigma }_{y}+ {C}_{2}}{{\sigma }_{x}^{2}+ {\sigma }_{y}^{2}+ {C}_{2}}$
式(15)中:${C}_{2}$ 为常数,其目的与${C}_{1}$ 相同,为避免${\sigma }_{x}^{2}+ {\sigma }_{y}^{2}$ 接近 0 时造成系统不稳定的情况出现。
结构对比函数$s\left({x, y}\right)$ 可以用相关性系数衡量, 它比较的是经过归一化后的$\frac{{x}_{i}- {\mu }_{x}}{{\sigma }_{x}}$$\frac{{y}_{i}- {\mu }_{y}}{{\sigma }_{y}}$,即
$ s\left({x, y}\right)= \frac{{\sigma }_{xy}+ {C}_{3}}{{\sigma }_{x}{\sigma }_{y}+ {C}_{3}}$
式(16)中:${C}_{3}$ 为常数,其目的与${C}_{1}\text{、}{C}_{2}$ 相同。
${\sigma }_{xy}= \frac{1}{N - 1}\mathop{\sum }\limits_{{i = 1}}^{N}\left({{x}_{i}- {\mu }_{x}}\right)\left({{y}_{i}- {\mu }_{y}}\right)$
最后, 3 个对比函数组合成一个完整的结构相似度衡量指标函数$\operatorname{SSIM}\left({x, y}\right)$,可表示为
$ f\left\lbrack {l\left({x, y}\right), c\left({x, y}\right), s\left({x, y}\right)}\right\rbrack $
它在性质上应当满足 3 个条件:① 对称性,$\operatorname{SSIM}\left({x, y}\right)= \operatorname{SSIM}\left({y, x}\right)$ ; ② 有界性,$0 \leq \operatorname{SSIM}(x$,$y)\leq 1$ ; ③最大值唯一性,当且仅当$x = y$ 时,$\operatorname{SSIM}\left({x, y}\right)= 1$
为满足上述条件,设${C}_{1}= {\left({K}_{1}L\right)}^{2},{C}_{2}= {\left({K}_{2}L\right)}^{2}$,${C}_{3}= \frac{{C}_{2}}{2}$,取${K}_{1}= {0.01},{K}_{2}= {0.03}\left({{K}_{1}\text{、}{K}_{2}}\right.$ 为对应于${C}_{1}$${C}_{2}$ 常数参数),图像灰度级数$L ={2}^{\text{bits per pixel }}-1$。例如,对于 8-bit 灰度图像,$L ={255}$
SSIM 指数的函数可以进一步表示为
$\operatorname{SSIM}\left({x, y}\right)= {\left\lbrack l\left(x, y\right)\right\rbrack }^{\alpha }{\left\lbrack c\left(x, y\right)\right\rbrack }^{\beta }{\left\lbrack s\left(x, y\right)\right\rbrack }^{\gamma }$
式 (19) 中:$\alpha ,\beta ,\gamma > 0$,分别代表各自不同特征在 SSIM 衡量指标中的占比。
为了得到简化形式,令$\alpha =\beta =\gamma = 1$,结合式(13)、式(15)和式(16),得
$\operatorname{SSIM}\left({x, y}\right)= \frac{\left({2{\mu }_{x}{\mu }_{y}+ {C}_{1}}\right)\left({2{\sigma }_{xy}+ {C}_{2}}\right)}{\left({{\mu }_{x}^{2}+ {\mu }_{y}^{2}+ {C}_{1}}\right)\left({{\sigma }_{x}^{2}+ {\sigma }_{y}^{2}+ {C}_{2}}\right)} $
利用结构相似度衡量指标的算法原理,对 CycleGAN的损失函数进行优化, 将图像间的纹理结构相似程度作为约束条件进一步改进网络模型, 提出纹理结构-循环一致性的生成对抗网络 (texture structure cycle-consistent generative adversarial network, TS-CycleGAN)。
对于结构相似度衡量指标而言, 在图像中对它进行局部求解更有利于反映出纹理和结构信息。主要原因有: ①全局求解只能实现对图像的整体感知, 无法捕捉到图像中不同区域间的纹理结构特征变化, 而局部求解对纹理变化更加敏感; ② 部分图像可能出现失真的问题, 将会对全局结构相似度衡量指标的计算产生不良影响, 但局部计算可以减轻此现象危害, 提升算法鲁棒性; ③局部求解相较于全局求解的计算速度更快,有利于提高算法综合效率。
因此, 设计算法时采用局部求解结构相似度衡量指标的方式,对式 (20) 中的${\mu }_{x}\text{、}{\sigma }_{x}$${\sigma }_{xy}$ 都进行加窗操作,通过加权窗口对整幅图片进行逐像素遍历。此遍历过程的每一步均根据窗口内的像素灰度值进行计算,最终得到一个$\operatorname{SSIM}\left({x, y}\right)$ 的矩阵。为了防止均匀的简单加权窗口 (如$n \times n$ 固定大小的方形窗)导致 SSIM 矩阵出现 “分块效应”,从而影响对局部纹理结构信息的评估, 采用标准差为 1.5 、 窗口尺寸为${11}\times {11}$ 的高斯加权窗口,可表示为
$ W =\{ w\left({i, j}\right)\mid i, j = 1,2,\cdots ,{11}\}$
$ w\left({i, j}\right)= \frac{1}{{2\pi }{\sigma }^{2}}\exp \left\lbrack {-\frac{{\left(i - 6\right)}^{2}+ {\left(j - 6\right)}^{2}}{2{\sigma }^{2}}}\right\rbrack $
式 (22) 中:$i$$j$ 分别为窗口中的行和列索引;$w\left({i, j}\right)$ 为窗口中每个像素的权重;$\sigma$ 为标准差。
因此,${\mu }_{x}\text{、}{\sigma }_{x}\text{、}{\sigma }_{xy}$ 可分别表示为
$\left\{\begin{array}{l}{\mathbf{\mu }}_{x}= \mathop{\sum }\limits_{{i = 1}}^{N}{w}_{i}{x}_{i}\\{\mathbf{\sigma }}_{x}= {\left\lbrack \mathop{\sum }\limits_{{i = 1}}^{N}{w}_{i}{\left({x}_{i}- {\mu }_{x}\right)}^{2}\right\rbrack }^{\frac{1}{2}}\\{\mathbf{\sigma }}_{xy}= \mathop{\sum }\limits_{{i = 1}}^{N}{w}_{i}\left({{x}_{i}- {\mu }_{x}}\right)\left({{y}_{i}- {\mu }_{y}}\right)\end{array}\right.$
式 (23) 中:${w}_{i}$ 为第$i$ 个像素的权重。
最后, 将各个通过加权窗口得到的结构相似度衡量指标$\operatorname{SSIM}\left({{x}_{i},{y}_{j}}\right)$ 通过平均的方法计算,得到全局图像整体的结构相似度评估值 MSSIM, 可表示为
$\operatorname{MSSIM}\left({x, y}\right)= \frac{1}{MN}\mathop{\sum }\limits_{{i = 1}}^{M}\mathop{\sum }\limits_{{i = 1}}^{N}\operatorname{SSIM}\left({{x}_{i},{y}_{i}}\right)$
式 (24) 中:$x\text{、}y$ 为图像;${x}_{i}$${y}_{i}$ 为局部 SSIM 指数在映射中的位置;$M$$N$ 为局部窗口的数量。
将 MSSIM 表达的纹理结构特征作为约束设计损失函数${L}_{\mathrm{{TS}}}$。结构相似度衡量指标的取值范围在$\left\lbrack {0,1}\right\rbrack$ 内,当值为 0 时,表明两张图片毫不相同,值为 1 时两张图片完全相同。在网络训练过程中, 损失函数的作用是衡量网络输出数据与实测数据之间的差异, 损失函数值越大代表二者之间的差异越大。反之代表二者差异越小, 网络的最终目标是使损失函数实现最小化。因此, 将纹理结构损失函数定义为
${L}_{\mathrm{{TS}}}= {L}_{G}\left\lbrack {G\left(X\right), Y}\right\rbrack +{L}_{F}\left\lbrack {F\left(Y\right), X}\right\rbrack \\= \{ 1 -\operatorname{MSSIM}\left\lbrack {G\left(X\right), Y}\right\rbrack \}+ \\\{ 1 -\operatorname{MSSIM}\left\lbrack {F\left(Y\right), X}\right\rbrack \}$
最后, 将基于结构纹理衡量指标的损失函数加入 CycleGAN 原本的损失函数式(10)中,得
$ L\left({G, F,{D}_{X},{D}_{Y}}\right)= {L}_{\mathrm{{GAN}}}+ {L}_{\text{cycle }}+ {L}_{\text{identity }}+ {L}_{\mathrm{{TS}}}\\= {L}_{\mathrm{G}}\left({G,{D}_{Y}, X, Y}\right)+ \\{L}_{F}\left({F,{D}_{X}, Y, X}\right)+ \\{\lambda }_{1}{L}_{\text{cyc }}\left({G, F}\right)+ \\{\lambda }_{2}{L}_{\text{identity }}\left({G, F}\right)+ \\{\lambda }_{3}\left\{{{L}_{G}\left\lbrack {G\left(X\right), Y}\right\rbrack +}\right.\\{L}_{F}\left\lbrack {F\left(Y\right), X}\right\rbrack $
式(26)中:${\lambda }_{3}$ 为纹理结构损失函数${L}_{\mathrm{{TS}}}$ 的权重参数,当其值设置为 0 时,${L}_{\mathrm{{SSIM}}}$ 不发挥作用。
在式 (26) 所示的损失函数的引导下, TS-Cy-cleGAN 网络训练流程如下。
步骤 1 构建两个数据集$A$$B$,其中,$A$ 为电磁仿真图像,$B$ 为实测 SAR 图像。
步骤 2 构建 TS-CycleGAN 网络, 包括两个生成器和两个判别器。其中,一个生成器将$A$ 映射到$B$,另一个生成器将$B$ 映射到$A$,两个判别器分别对$A$$B$ 进行判别。
步骤 3 定义损失函数如式 (26) 所示,${L}_{\mathrm{{GAN}}}$ 对网络中两个生成器$G$$F$ 进行约束,使它们能够生成与实测样本更相似的图片,${L}_{\mathrm{{cyc}}}$ 保证网络训练出的图片在改变图像背景特征的同时保持目标内容不变,${L}_{\text{identity }}$ 使网络输入图像和输出图像的颜色组成保持一致,${L}_{\mathrm{{SSIM}}}$ 则关注图像的纹理结构特征,使输出的优化图像纹理结构特征与输入实测图像达到尽可能相近。
步骤 4 训练优化,首先固定生成器参数,将$B$ 数据集图像输入判别器中, 训练其达到一定识别率。其次,固定判别器参数,使生成器将数据集$A$ 中的图像转换为近似数据集$B$ 特征的图像${B}^{\prime }$,以此训练生成器,随后将生成图像${B}^{\prime }$ 输入判别器与实测$B$ 进行比较,输出对生成样本的识别结果并对判别器和生成器反向传播。最后,再将$A$ 数据集图像输入判别器中, 先训练判别器, 再训练生成器, 并重复上述训练过程。在训练生成器时, 同时优化损失函数, 以达到最优化的效果。
本试验的训练与验证部分分别使用 SAMPLE 数据集和 MSTAR 数据集。
MSTAR (the moving and stationary target acquisition and recognition) 数据集由多种类别的地面静止 SAR 目标图像组成, 由美国国防部高级研究计划局 (Defense Advanced Research Projects Agency, DARPA) 资助的航空目标识别数据集[9]
本试验选择其中 10 类军事目标对本文算法进行 验 证:2S1、BMP2、BRDM2、BTR60、BTR70、D7、 T62、T72、ZIL131、ZSU23-4。图7 分别展示了这 10 类目标的光学图像与其对应的 SAR 图像。
SAMPLE (synthetic and measured paired labeled experiment) 数据集[7] 由俄亥俄州空军研究所于 2019 年发布。SAMPLE 数据集由来自 MSTAR 据集的 SAR 图像和与之匹配的电磁仿真图像组成, MSTAR 中 10 类目标的实测 SAR 图像与其对应的仿真图像对如图8 所示。
为了构造出 “优化 SAR 电磁仿真图像以对小样本 SAR 数据进行补充” 的应用场景, 本试验从 SAMPLE数据集中选取类别为 BMP2、BTR70 和 T72 的电磁仿真图像作为训练子集 A(Trainset A), 并从充足类别样本中随机抽取部分实测 SAR 图像作为训练子集 B(Trainset B)。
训练子集 B 中有 3 类目标仅包含 10 个图像,属于小样本数据集。两部分图像作为两种具有不同领域特征的图像组成图像优化方法的样本训练集, 将被用于图像优化实验中 CycleGAN 和 TS-CycleGAN的网络训练,具体数量如表1 所示。
在对优化样本识别性能验证试验中使用 MSTAR 数据集,并选用了其中俯仰角为${15}^{\circ }$${17}^{\circ }$ 的 SAR 目标图像,将 17°俯仰角的 SAR 目标图像作为训练识别模型时使用的样本集,${15}^{\circ }$ 俯仰角的 SAR 目标图像作为测试时使用的样本集。以目标物中心作为中心点, 在试验进行前首先将原始图像统一裁剪为${128}\times {128}$ 的尺寸,包含军事目标和它的阴影部分作为图像内容, 并保留了适量面积提供图像背景的结构纹理特征信息。为了构造出本试验需要的小样本问题下的 SAR 目标识别条件, 将两种数据集的组合方式设置为完全由小样本类别组成的完全小样本场景和部分类别为小样本数据集的部分小样本场景。
从 10 类目标中选取 BMP2、BTR70 和 T72 军事目标作为小样本类别, 其他 7 类为充足样本类别。对于完全小样本场景而言, 此场景下的每一类目标均为存在数据不足的情况。因此, 从 MSTAR 数据集中,对各类样本的${17}^{\circ }$ 俯仰角数据随机抽选出 10 张SAR 目标图像作为 3 类小样本目标训练样本, 而测试集为 196 张 15${}^{\circ }$ 俯仰角的 SAR 图像,构造出表2 所示的 3 类小样本数据集。
现实情况下, 并非所有的类别都面临着小样本状况,时常会出现“部分类别样本缺乏,但其他类别样本充足”的部分类别样本缺乏场景。为了使得试验更具有真实性, 本试验从 7 类充足样本类别的每类样本集中随机抽选出 100 张 SAR 目标图像作为训练样本, 加入表2 所示的 3 类小样本数据集中, 构造出部分类别数据量仅为正常条件下 10% 的部分小样本 MSTAR 数据集, 这意味着现有的分类网络很难利用该数据集将识别率提升到可接受水平, 如表3 所示。
本试验基于 Pytorch 框架, 使用的硬件和软件配置如表4 所示。
本试验利用TS-CycleGAN算法将电磁仿真图像优化为 SAR 图像, 从图像灰度统计特性的角度验证优化后样本的接近实测样本的程度。
图9 所示, 第 1 列为实测 SAR 图像, 第 2 列为电磁仿真 SAR 图像, 第 3 列为 CycleGAN 的优化结果, 第 4 列为 TS-CycleGAN 的优化结果。可以看出, CycleGAN 和 TS-CycleGAN 的优化效果之间存在差距,在二者都能保持电磁仿真图像提供的目标特征的前提下, 所提出的 TS-CycleGAN 使优化图像的背景与实测图像的背景更加相似, 达到的优化效果更真实。并且, 在第 5 列和第 6 列的红色框标注区域中能够明显观察到 CycleGAN 优化后的图像出现色调差异,且背景的纹理过于平滑、均匀。
本试验通过计算图像的灰度直方图,量化分析生成样本与实测样本的相似程度。试验从 3 个小样本类别中随机抽取优化样本及其对应的实测样本, 计算并对比它们的灰度直方图, 比较结果如图10 所示,横坐标为灰度级 0~255,纵坐标为像素个数。其中,第$1 \sim 3$ 行分别代表BMP2、BTR70 和 T72,每一行从左到右分别为实测图像、CycleGAN 优化图像和 TS-CycleGAN 优化图像的灰度直方图。
图10 可以看出, TS-CycleGAN 优化样本的直方图相似程度明显优于普通 CycleGAN, 尤其是对于 BMP2 和 T72 两个类别 (第 1 行和第 3 行), 二者的灰度直方图存在明显差异。
最后,为了让结果更具有说服力,本试验采用了直方图相似度指数来进一步衡量直方图之间的相似度, 从而对比图像之间的相似性。首先计算两个图片的直方图数据,对各自图像直方图进行归一化后, 使用巴氏系数计算得出图像的相似度值, 可表示为
$\text{HSI}= \mathop{\sum }\limits_{{x = 0}}^{{255}}\sqrt{p\left(x\right) q\left(x\right)} $
式 (27) 中:$p\left(x\right)$$q\left(x\right)$ 分别为两幅图像直方图上横坐标为$x$ 的位置上的概率分布; HSI 为直方图相似度,其范围为范围在$\left\lbrack {0,1}\right\rbrack$,为 1 时表明两幅图完全相同,为 0 时表示二者之间没有任何相似之处。
在训练的过程中, 每当训练一个循环后使用最新的生成模型随机优化 20 个样本, 计算其与实测样本的直方图相似度值指数并记录。在训练 CycleGAN与 TS-CycleGAN 网络时都使用上述方式计算直方图相似度指数, 统计结果如图11 所示。可以看出, 相比于一般的 CycleGAN, 所提出的 TS-CycleGAN的整体性能更好, 所优化的样本与实测样本的相似度更高。
从 2.3 节试验结果可以看出, 所提出的 TS-CycleGAN网络优化出的样本具有较高的真实性。因此需要验证这些优化样本在 SAR 目标识别模型信息缺失的情况下是否能够有效补充样本信息,提升模型的识别能力。
除样本质量外, 样本数量同样对模型识别性能有较大的影响, 且当训练集样本数量已足够让识别模型得到充分训练的时候, 新样本对识别性能的提升将会变得微小。因此, 为了充分验证 SAR 目标优化样本的有效价值, 试验首先构造表2表3 所示样本数量严重不足的场景, 然后采用随机抽取的方式, 以 30 个 SAR 目标样本作为一批, 将电磁仿真样本和两种优化样本分批添加到训练集的小样本类别中, 对 SAR 目标识别模型不断地再训练并观察识别率的变化。
目标识别准确率是评估目标识别算法性能的重要指标, 定义为
${A}_{\mathrm{{cc}}}= \frac{{T}_{\mathrm{p}}+ {T}_{\mathrm{N}}}{{T}_{\mathrm{p}}+ {T}_{\mathrm{N}}+ {F}_{\mathrm{P}}+ {F}_{\mathrm{N}}}$
式 (28) 中:${T}_{\mathrm{p}}$ 为正确识别为目标类别的样本数;${T}_{\mathrm{N}}$ 为正确识别为非目标类别的样本数;${F}_{\mathrm{p}}$ 为错误识别为目标类别的样本数;${F}_{\mathrm{N}}$ 为错误识别为非目标类别的样本数。
准确率表示模型在所有样本中的正确识别数量与总样本数量的比值, 是衡量模型整体识别性能的重要指标。
本试验中使用的 SAR 目标识别方法为卷积神经网络 VGG11[20],相对而言属于浅层次的简单网络, 收敛速度较快, 对于大量实验而言耗时相对较少, 但同时又具有一定的目标识别性能。因此可以用来验证本文方法是否有效, 如果目标识别率在加入本文方法后得到大幅度提升,则说明本优化方法具有实际应用价值。
图12 可以看出, 在 3 类样本同时缺失的完全小样本场景下, 未补充样本数据时, 各类目标的识别率均为${33}\%$,表明此时的目标识别性能几乎没有发挥作用。随着电磁仿真数据和优化样本的输入, 数据集各类别的识别率和总识别率明显提升。电磁仿真图像的加入使识别率提升至${60}\%$,且效果不稳定, 在输入超过 60 张仿真图像后识别率明显呈下降趋势。基于 CycleGAN 方法优化图像的加入对识别率的影响与电磁仿真试验趋势相同, 但效果明显优于电磁仿真图像,最高能达到 69.10% 的目标识别准确率。基于 TS-CycleGAN 方法的优化图像对识别率的提升效果在每个阶段均高于电磁仿真图像与 CycleGAN 优化数据, 达到 71.73%, 且随着扩充样本数量的加入目标识别性能尽管有所下降, 但其稳定性也高于前二者。
图13 可以看出, 在 10 类目标中仅 3 类数据不完备的部分小样本场景下, 直接将电磁仿真样本添加到训练集中训练时, SAR 目标网络的识别性能呈现震荡的趋势, 当添加一定数量之后性能才有一定上升,不过识别率仍然较低。当加入 CycleGAN 优化数据时, 识别性能也呈现震荡趋势, 整体识别率虽然高于电磁仿真数据效果但明显低于添加了基于 TS-CycleGAN 的优化样本训练效果。因此, 基于 TS-CycleGAN 网络优化的 SAR 目标样本对识别率的提升更加明显且稳定。
综上所述, 上述两个试验证明电磁仿真样本和优化样本都能够起到补充判别信息的效果, 并且由于优化样本经过了从电磁仿真图像到实测图像的转换, 在保留电磁仿真技术对目标精确建模的同时与 SAR 图像背景特征更相近, 因此对网络识别性能的提升效果更加明显。由试验结果可知:从两个不同条件下的试验结果可以看出, 过量添加非真实样本对识别效果起负作用;由图13 可知,在部分小样本条件下, 当各个类别样本数量达到均衡时, 整体识别效果最好。
本试验验证了 TS-CycleGAN 网络相比于未改进的 CycleGAN 网络具有更好的性能, 经过其对图像背景的校正,基于电磁仿真图像的优化图像在真实性上更接近实测 SAR 图像,并能够显著的提升小样本条件下模型的识别性能。
(1)为了进一步突破电磁建模仿真数据的应用瓶颈, 减小电磁仿真 SAR 图像与实测 SAR 图像之间的差异, 提升基于 SAR 电磁仿真图像的小样本条件下 SAR 目标识别模型的识别性能, 基于图像优化方法思路, 提出了更加关注 SAR 图像纹理结构信息的 TS-CycleGAN 网络, 通过对原始 CycleGAN 网络的损失函数进行重新设计, 将结构相似度衡量指数改造为损失函数并作为约束加入网络模型训练过程中。
(2)从样本真实性验证试验结果可以看出, TS-CycleGAN网络优化的仿真图像在统计特性上与实测图像的相似度更高。进一步设计识别性能验证试验, 将电磁仿真样本、CycleGAN 优化样本和 TS-CycleGAN 优化样本分批次加入小样本数据集中, 并对比在不同样本数量下模型的识别性能。从试验结果可以明显的看出, TS-CycleGAN 网络优化样本对 SAR 目标识别模型的识别性能提升最大, 这说明所提出的 TS-CycleGAN 能在保持图像目标特征的前提下, 对图像的背景纹理结构产生良好的优化效果。
(3)在识别性能验证试验中发现,对 SAR 仿真数据进行优化后, 优化样本的挑选方式也会影响模型的识别性能。因此, 后续工作会进一步研究数据的质量评估与筛选方法。
  • 国家自然科学基金(62271116)
参考文献 引证文献
排序方式:
[1]
宋建社, 郑永安, 袁礼海. 合成孔径雷达图像理解与应用[M]. 北京: 科学出版社, 2008.
Song Jianshe, Zheng Yong'an, Yuan Lihai. Synthetic aperture radar image understanding and applications[M]. Beijing: Science Press, 2008.
[2]
Liu Z, Yuan L, Weng L, et al. A high resolution opticalsatellite image dataset for ship recognition and some new baselines[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Shanghai: IEEE, 2017. 324-331.
[3]
Li C, Xu C, Cui Z, et al. Feature-attentioned object detection in remote sensing imagery[C]// IEEE International Conference on Image Processing (ICIP). New York: IEEE, 2019. 3886-3890.
[4]
Xia GS, Bai X, Ding J, et al. DOTA: A large-scale dataset for object detection in aerial images[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018. 3974-3983.
[5]
Kawano Y, Yanai K. Foodcam: a real-time food recognition system on a smartphone[J]. Multimedia Tools and Applications, 2015. 74. 5263-5287.
[6]
Mumtaz A, Coviello E, Lanckriet GRG, et al. Clustering dynamic textures with the hierarchical EM algorithm for modeling video[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013. 35(7): 1606-1621.
[7]
Lewis B, Scarnati T, Sudkamp E, et al. A SAR dataset for ATR development: the synthetic and measured paired labeled experiment (SAMPLE)[C]// Algorithms for Synthetic Aperture Radar Imagery. New York: IEEE, 2019. 39-54.
[8]
Shan JZ, Qu XJ, Chao K, et al. Efficient SAR image simulation for target on rough ground surface using the improved SBR technique[J]. Chinese Journal of Radio Science, 2024. 39(2): 322-331.
[9]
付松, 黄秀琼, 周龙建, 等. 粗糙地面上复杂目标的电磁散射特性[J]. 科学技术与工程, 2021. 21(17): 7159-7167.
Fu Song, Huang Xiuqiong, Zhou Longjian, et al. Electromagnetic scattering characteristics of complex targets on rough ground[J]. Science Technology and Engineering, 2021. 21(17): 7159-7167.
[10]
Hu LP, Yao CM, Huang J, et al. Highlyrobust synthetic aperture radar target recognition method based on simulation data training[J]. International Journal of Antennas and Propagation, 2022. 2022. 7537732.
[11]
Wang J, Li J, Sun B, et al. SAR image synthesis based on conditional generative adversarial networks[J]. The Journal of Engineering, 2019. 2019(21): 8093-8097.
[12]
Malmgren-Hansen D, Kusk A, Dall J, et al. Improving SAR automatic target recognition models with transfer learning from simulated data[J]. IEEE Geoscience and remote sensing Letters, 2017. 14(9): 1484-1488.
[13]
Xing MD, Xie YY, Gao YX, et al. Electromagnetic scattering characteristic extraction and imaging recognition algorithm: a review[J]. Journal of Radars, 2022. 11(6): 921-942.
[14]
Wu K, Jin G W, Xiong X, et al. Fast SAR image simulation based on echo matrix cell algorithm including multiple scattering[J]. Remote Sensing, 2023. 15(14): 36-37.
[15]
Zhu JY, Park T, Isola P, et al. Unpaired image-to-image translation using cycle-consistent adversarial networks[C]// Proceedings of the IEEE International Conference on Computer Vision. New York: IEEE, 2017. 2223-2232.
[16]
Iqbal N, Mumtaz R, Shafi U, et al. Gray level co-occurrence matrix (GLCM) texture based crop classification using low altitude remote sensing platforms[J]. PeerJ Computer Science, 2012. 7. e536.
[17]
张静远, 张冰, 蒋兴舟. 基于小波变换的特征提取方法分析[J]. 信号处理, 2000. 16(2): 156-162.
Zhang Jingyuan, Zhang Bing, Jiang Xingzhou. Analysis of wavelet transform based feature extraction methods[J]. Signal Processing, 2000. 16(2): 156-162.
[18]
Feichtinger HG, Thomas S. Gabor analysis and algorithms: theory and applications[M]. Berlin: Springer International Publishing, 2012.
[19]
Brunet D, Edward RV, Zhou W. On the mathematical properties of the structural similarity index[J]. IEEE Transactions on Image Processing, 2011. 21(4): 1488-1499.
[20]
Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. arXiv Preprint, 2014. arXiv. 1409-1556.
2025年第25卷第8期
PDF下载
334
134
引用本文
BibTeX
文章信息
doi: 10.12404/j.issn.1671-1815.2403736
  • 接收时间:2024-05-20
  • 首发时间:2025-07-29
  • 出版时间:2025-03-18
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2024-05-20
  • 修回日期:2025-01-02
基金
国家自然科学基金(62271116)
作者信息
    1 电子科技大学成都学院 工学院 成都 611731
    2 电子科技大学 信息与通信工程学院 成都 611731

通讯作者:

*崔宗勇(1984—),男,汉族,山东菏泽人,博士,副教授。研究方向:SAR图像解译。E-mail: zycui@uestc.edu.cn。
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2403736
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数

Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)

Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏