Article(id=1149774733274739464, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1149774724923880044, articleNumber=null, orderNo=null, doi=10.12404/j.issn.1671-1815.2403743, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1716220800000, receivedDateStr=2024-05-21, revisedDate=1737475200000, revisedDateStr=2025-01-22, acceptedDate=null, acceptedDateStr=null, onlineDate=1752057258194, onlineDateStr=2025-07-09, pubDate=1745769600000, pubDateStr=2025-04-28, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1752057258194, onlineIssueDateStr=2025-07-09, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1752057258194, creator=13701087609, updateTime=1752057258194, updator=13701087609, issue=Issue{id=1149774724923880044, tenantId=1146029695717560320, journalId=1146123166801305609, year='2025', volume='25', issue='12', pageStart='4827', pageEnd='5272', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=0, createTime=1752057256203, creator=13701087609, updateTime=1768456746933, updator=13701087609, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1218559174552764785, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1149774724923880044, language=EN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1218559174552764786, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1149774724923880044, language=CN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=5093, endPage=5102, ext={EN=ArticleExt(id=1149774733547369230, articleId=1149774733274739464, tenantId=1146029695717560320, journalId=1146123166801305609, language=EN, title=Ship Target Detection Algorithm Based on Improved YOLOv8, columnId=1156262729162810294, journalTitle=Science Technology and Engineering, columnName=Papers·Automation and Computational Technology, runingTitle=null, highlight=null, articleAbstract=

An improved DGA-YOLOv8 offshore ship target detection algorithm was proposed to tackle the issues of low accuracy and single ship detection categories that are present in traditional ship target detection algorithms. Firstly, the network was adapted to include deformable convolution, which expanded the model's receptive field. Learnable offsets were introduced, allowing the model to adaptively adjust the size and shape of the receptive field in response to the actual shape of the object, ensuring that the convolution area can precisely cover the contour of the ship object. Secondly, the incorporation of a GAM(global attention mechanism) attention mechanism enabled the network to effectively emphasize the key features of ship targets, thereby enhancing the target recognition capability. The experimental results demonstrate that the improved algorithm achieves accuracy and average accuracy mean (mAP) of 96.4% and 92.2%, respectively. An frames per second(FPS) of 43.55 is recorded, indicating not only an enhancement in accuracy but also the maintenance of a certain detection speed, thus fulfilling the requirements for real-time detection. When compared with other mainstream algorithms, such as faster region-based convolutional neural network(Faster R-CNN) and YOLOv5s, YOLOv10. The results show that the proposed algorithm exhibits higher average accuracy and significant superior classification performance.

, correspAuthors=Qing-hua LIU, 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=Cong-xin DONG, Qing-hua LIU), CN=ArticleExt(id=1149774768183931054, articleId=1149774733274739464, tenantId=1146029695717560320, journalId=1146123166801305609, language=CN, title=基于改进YOLOv8的船只目标检测算法, columnId=1156262729783567290, journalTitle=科学技术与工程, columnName=论文·自动化技术、计算机技术, runingTitle=null, highlight=null, articleAbstract=

针对传统船只目标检测算法的精度较低、船只检测类别单一等问题,提出一种改进的DGA-YOLOv8海上船只目标检测算法。首先,网络采用可变形卷积扩大模型的感受野,通过引入可学习偏移量,使模型能够根据物体实际形状自适应调整感受野大小和形状,确保卷积区域能够精确覆盖船只物体的轮廓。其次,引入GAM(global attention mechanism)注意力机制,使网络能够有效突出船只目标的关键特征,提高目标识别能力。实验结果表明:改进后算法的精确度和平均精度均值(mAP)达到96.4%和92.2%,FPS(frame per second)为43.55,在提升精度的同时也保证了一定的检测速度,满足了实时性检测的需求。同时与其他主流算法对比,其中包括Faster R-CNN(faster region-based convolutional neural network)、YOLOv5s和YOLOv10等。结果表明:所提算法具有更高的平均精度和更显著的分类效果。

, correspAuthors=刘庆华, authorNote=null, correspAuthorsNote=
* 刘庆华(1974—),女,汉族,四川南江人,博士,教授,博士研究生导师。研究方向:阵列信号处理。E-mail:
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=SVAb4IYpLpw3/xDlQg2b9g==, magXml=/rcdrASDYDpHfiQoeBbPYg==, pdfUrl=null, pdf=PrkhN6WQloPR4d4HLVkORA==, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=4EB6j+wDuf1nCXC3/FKhpA==, mapNumber=null, authorCompany=null, fund=null, authors=

董从鑫(2001—),男,汉族,广西贺州人,硕士研究生。研究方向:目标检测。E-mail:

, authorsList=董从鑫, 刘庆华)}, authors=[Author(id=1180089380308926665, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=2440701788@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1180089380376035531, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, authorId=1180089380308926665, language=EN, stringName=Cong-xin DONG, firstName=Cong-xin, middleName=null, lastName=DONG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=Information and Communication College, Guilin University of Electronic and Technology, Guilin 541010, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1180089380430561484, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, authorId=1180089380308926665, language=CN, stringName=董从鑫, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=桂林电子科技大学信息与通信学院, 桂林 541010, bio={"content":"

董从鑫(2001—),男,汉族,广西贺州人,硕士研究生。研究方向:目标检测。E-mail:

"}, bioImg=null, bioContent=

董从鑫(2001—),男,汉族,广西贺州人,硕士研究生。研究方向:目标检测。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1180089380233429189, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, xref=null, ext=[AuthorCompanyExt(id=1180089380241817798, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, companyId=1180089380233429189, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=Information and Communication College, Guilin University of Electronic and Technology, Guilin 541010, China), AuthorCompanyExt(id=1180089380250206407, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, companyId=1180089380233429189, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=桂林电子科技大学信息与通信学院, 桂林 541010)])]), Author(id=1180089380493476046, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=qhliu@guet.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1180089380560584912, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, authorId=1180089380493476046, language=EN, stringName=Qing-hua LIU, firstName=Qing-hua, middleName=null, lastName=LIU, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=Information and Communication College, Guilin University of Electronic and Technology, Guilin 541010, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1180089380627693777, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, authorId=1180089380493476046, language=CN, stringName=刘庆华, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=桂林电子科技大学信息与通信学院, 桂林 541010, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1180089380233429189, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, xref=null, ext=[AuthorCompanyExt(id=1180089380241817798, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, companyId=1180089380233429189, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=Information and Communication College, Guilin University of Electronic and Technology, Guilin 541010, China), AuthorCompanyExt(id=1180089380250206407, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, companyId=1180089380233429189, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=桂林电子科技大学信息与通信学院, 桂林 541010)])])], keywords=[Keyword(id=1180089380761911506, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, orderNo=1, keyword=YOLOv8), Keyword(id=1180089380820631763, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, orderNo=2, keyword=ship target detection), Keyword(id=1180089380879352020, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, orderNo=3, keyword=deformable convolution), Keyword(id=1180089380938072277, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, orderNo=4, keyword=attention mechanism), Keyword(id=1180089380992598230, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, orderNo=1, keyword=YOLOv8), Keyword(id=1180089381047124183, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, orderNo=2, keyword=船只目标检测), Keyword(id=1180089381114233048, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, orderNo=3, keyword=可变形卷积), Keyword(id=1180089381177147609, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, orderNo=4, keyword=注意力机制)], refs=[Reference(id=1180089383735673085, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2022, volume=22, issue=25, pageStart=11241, pageEnd=11249, url=null, language=null, rfNumber=[1], rfOrder=0, authorNames=徐慧智, 宋爱秋, 武笑宇, journalName=科学技术与工程, refType=null, unstructuredReference=徐慧智, 宋爱秋, 武笑宇. 基于均匀设计的船舶目标检测深度学习模型训练方法研究[J]. 科学技术与工程, 2022, 22(25): 11241-11249., articleTitle=基于均匀设计的船舶目标检测深度学习模型训练方法研究, refAbstract=null), Reference(id=1180089383798587646, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2022, volume=22, issue=25, pageStart=11241, pageEnd=11249, url=null, language=null, rfNumber=[1], rfOrder=1, authorNames=Xu Huizhi, Song Aiqiu, Wu Xiaoyu, journalName=Science Technology and Engineering, refType=null, unstructuredReference=Xu Huizhi, Song Aiqiu, Wu Xiaoyu. Research on training method of deep learning to ship target detection based on uniform design[J]. Science Technology and Engineering, 2022, 22(25): 11241-11249., articleTitle=Research on training method of deep learning to ship target detection based on uniform design, refAbstract=null), Reference(id=1180089383853113599, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2019, volume=49, issue=S2, pageStart=185, pageEnd=191, url=null, language=null, rfNumber=[2], rfOrder=2, authorNames=石洪基, 郎海涛, 宋棋, journalName=中国海洋大学学报(自然科学版), refType=null, unstructuredReference=石洪基, 郎海涛, 宋棋, 等. 基于改进区域卷积神经网络的SAR图像船只检测方法[J]. 中国海洋大学学报(自然科学版), 2019, 49(S2): 185-191., articleTitle=基于改进区域卷积神经网络的SAR图像船只检测方法, refAbstract=null), Reference(id=1180089383916028160, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2019, volume=49, issue=S2, pageStart=185, pageEnd=191, url=null, language=null, rfNumber=[2], rfOrder=3, authorNames=Shi Hongji, Lang Haitao, Song Qi, journalName=Journal of Ocean University of China(Natural Science Edition), refType=null, unstructuredReference=Shi Hongji, Lang Haitao, Song Qi, et al. Ship detection method in SAR images based on improved regional convolutional neural network[J]. Journal of Ocean University of China(Natural Science Edition), 2019, 49(S2): 185-191., articleTitle=Ship detection method in SAR images based on improved regional convolutional neural network, refAbstract=null), Reference(id=1180089383962165505, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2022, volume=46, issue=4, pageStart=22, pageEnd=27, url=null, language=null, rfNumber=[3], rfOrder=4, authorNames=王程丽, 赵静, 杨攀攀, journalName=移动通信, refType=null, unstructuredReference=王程丽, 赵静, 杨攀攀, 等. 基于迁移学习的小样本SAR船只目标识别[J]. 移动通信, 2022, 46(4): 22-27., articleTitle=基于迁移学习的小样本SAR船只目标识别, refAbstract=null), Reference(id=1180089384020885762, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2022, volume=46, issue=4, pageStart=22, pageEnd=27, url=null, language=null, rfNumber=[3], rfOrder=5, authorNames=Wang Chengli, Zhao Jing, Yang Panpan, journalName=Mobile Communications, refType=null, unstructuredReference=Wang Chengli, Zhao Jing, Yang Panpan, et al. Smallsample SAR ship target recognition based on transfer learning[J]. Mobile Communications, 2022, 46(4): 22-27., articleTitle=Smallsample SAR ship target recognition based on transfer learning, refAbstract=null), Reference(id=1180089384087994627, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2021, volume=10, issue=6, pageStart=885, pageEnd=894, url=null, language=null, rfNumber=[4], rfOrder=6, authorNames=刘方坚, 李媛, journalName=雷达学报, refType=null, unstructuredReference=刘方坚, 李媛. 基于视觉显著性的 SAR 遥感图像 NanoDet 舰船检测方法[J]. 雷达学报, 2021, 10(6): 885-894., articleTitle=基于视觉显著性的 SAR 遥感图像 NanoDet 舰船检测方法, refAbstract=null), Reference(id=1180089384146714884, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2021, volume=10, issue=6, pageStart=885, pageEnd=894, url=null, language=null, rfNumber=[4], rfOrder=7, authorNames=Liu Fangjian, Li Yuan, journalName=Radar Journal, refType=null, unstructuredReference=Liu Fangjian, Li Yuan. NanoDet ship detection method for SAR remote sensing images based on visual saliency[J]. Radar Journal, 2021, 10(6): 885-894., articleTitle=NanoDet ship detection method for SAR remote sensing images based on visual saliency, refAbstract=null), Reference(id=1180089384205435141, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2024, volume=24, issue=17, pageStart=7205, pageEnd=7213, url=null, language=null, rfNumber=[5], rfOrder=8, authorNames=宣以国, 余成波, 蒋启超, journalName=科学技术与工程, refType=null, unstructuredReference=宣以国, 余成波, 蒋启超, 等. 基于改进YOLOv7的道路裂缝和坑洞检测算法[J]. 科学技术与工程, 2024, 24(17): 7205-7213., articleTitle=基于改进YOLOv7的道路裂缝和坑洞检测算法, refAbstract=null), Reference(id=1180089384268349702, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2024, volume=24, issue=17, pageStart=7205, pageEnd=7213, url=null, language=null, rfNumber=[5], rfOrder=9, authorNames=Xuan Yiguo, Yu Chengbo, Jiang Qichao, journalName=Science Technology and Engineering, refType=null, unstructuredReference=Xuan Yiguo, Yu Chengbo, Jiang Qichao, et al. Improved YOLOv7 road crack and pothole detection algorithm[J]. Science Technology and Engineering, 2024, 24(17): 7205-7213., articleTitle=Improved YOLOv7 road crack and pothole detection algorithm, refAbstract=null), Reference(id=1180089384339652871, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2014, volume=null, issue=null, pageStart=580, pageEnd=587, url=null, language=null, rfNumber=[6], rfOrder=10, authorNames=Girshick R, Donahue J, Darrell T, journalName=Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, refType=null, unstructuredReference=Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2014: 580-587., articleTitle=Rich feature hierarchies for accurate object detection and semantic segmentation, refAbstract=null), Reference(id=1180089384402567432, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2015, volume=null, issue=null, pageStart=1440, pageEnd=1448, url=null, language=null, rfNumber=[7], rfOrder=11, authorNames=Girshick R, journalName=Proceedings of the IEEE International Conference on Computer Vision, refType=null, unstructuredReference=Girshick R. Fast R-CNN[C]// Proceedings of the IEEE International Conference on Computer Vision. New York: IEEE, 2015: 1440-1448., articleTitle=Fast R-CNN, refAbstract=null), Reference(id=1180089384461287689, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2016, volume=39, issue=6, pageStart=1137, pageEnd=1149, url=null, language=null, rfNumber=[8], rfOrder=12, authorNames=Ren S, He K, Girshick R, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=Ren S, He K, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 39(6): 1137-1149., articleTitle=Faster R-CNN: towards real-time object detection with region proposal networks, refAbstract=null), Reference(id=1180089384520007946, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=779, pageEnd=788, url=null, language=null, rfNumber=[9], rfOrder=13, authorNames=Redmon J, Divvala S, Girshick R, journalName=Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, refType=null, unstructuredReference=Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2016: 779-788., articleTitle=You only look once: unified, real-time object detection, refAbstract=null), Reference(id=1180089384582922507, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=7263, pageEnd=7271, url=null, language=null, rfNumber=[10], rfOrder=14, authorNames=Redmon J, Farhadi A, journalName=Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, refType=null, unstructuredReference=Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2017: 7263-7271., articleTitle=YOLO9000: better, faster, stronger, refAbstract=null), Reference(id=1180089384641642764, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[11], rfOrder=15, authorNames=Redmon J, Farhadi A, journalName=arXiv Preprint, refType=null, unstructuredReference=Redmon J, Farhadi A. YOLOv3: an incremental improvement[J]. arXiv Preprint, 2018: arXiv: 1804.02767., articleTitle=YOLOv3: an incremental improvement, refAbstract=null), Reference(id=1180089384700363021, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[12], rfOrder=16, authorNames=Bochkovskiy A, Wang C Y, Liao H Y M, journalName=arXiv Preprint, refType=null, unstructuredReference=Bochkovskiy A, Wang C Y, Liao H Y M. YOLOv4: optimal speed and accuracy of object detection[J]. arXiv Preprint, 2020: arXiv: 2004.10934., articleTitle=YOLOv4: optimal speed and accuracy of object detection, refAbstract=null), Reference(id=1180089384767471886, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=7464, pageEnd=7475, url=null, language=null, rfNumber=[13], rfOrder=17, authorNames=Wang C Y, Bochkovskiy A, Liao H Y M, journalName=Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, refType=null, unstructuredReference=Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 7464-7475., articleTitle=YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors, refAbstract=null), Reference(id=1180089384830386447, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2023, volume=38, issue=1, pageStart=37, pageEnd=43, url=null, language=null, rfNumber=[14], rfOrder=18, authorNames=李静, 鲜林, 王海江, journalName=成都信息工程大学学报, refType=null, unstructuredReference=李静, 鲜林, 王海江. 基于YOLOv3的船只检测算法研究[J]. 成都信息工程大学学报, 2023, 38(1): 37-43., articleTitle=基于YOLOv3的船只检测算法研究, refAbstract=null), Reference(id=1180089384889106704, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2023, volume=38, issue=1, pageStart=37, pageEnd=43, url=null, language=null, rfNumber=[14], rfOrder=19, authorNames=Li Jing, Xian Lin, Wang Haijiang, journalName=Journal of Chengdu University of Information Technology, refType=null, unstructuredReference=Li Jing, Xian Lin, Wang Haijiang. Research on ship detection algorithm based on YOLOv3[J]. Journal of Chengdu University of Information Technology, 2023, 38(1): 37-43., articleTitle=Research on ship detection algorithm based on YOLOv3, refAbstract=null), Reference(id=1180089384952021265, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=483, pageEnd=487, url=null, language=null, rfNumber=[15], rfOrder=20, authorNames=Liu T, Zhou B, Zhao Y, journalName=2021 6th International Conference on Automation, Control and Robotics Engineering(CACRE), refType=null, unstructuredReference=Liu T, Zhou B, Zhao Y, et al. Ship detection algorithm based on improved YOLOv5[C]// 2021 6th International Conference on Automation, Control and Robotics Engineering(CACRE). New York: IEEE, 2021: 483-487., articleTitle=Ship detection algorithm based on improved YOLOv5, refAbstract=null), Reference(id=1180089385010741522, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=222, pageEnd=227, url=null, language=null, rfNumber=[16], rfOrder=21, authorNames=Wang B, Han B, Yang L, journalName=2021 6th International Conference on Transportation Information and Safety(ICTIS), refType=null, unstructuredReference=Wang B, Han B, Yang L. Accurate real-time ship target detection using YOLOv4[C]// 2021 6th International Conference on Transportation Information and Safety(ICTIS). NewYork: IEEE, 2021: 222-227., articleTitle=Accurate real-time ship target detection using YOLOv4, refAbstract=null), Reference(id=1180089385086238995, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2024, volume=null, issue=null, pageStart=659, pageEnd=665, url=null, language=null, rfNumber=[17], rfOrder=22, authorNames=Zhang J, journalName=International Conference on Mechatronic Engineering and Artificial Intelligence, refType=null, unstructuredReference=Zhang J. Ship target detection based on CBAM-YOLOv8[C]// International Conference on Mechatronic Engineering and Artificial Intelligence. Shanghai: SPIE, 2024: 659-665., articleTitle=Ship target detection based on CBAM-YOLOv8, refAbstract=null), Reference(id=1180089385149153556, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=764, pageEnd=773, url=null, language=null, rfNumber=[18], rfOrder=23, authorNames=Dai J, Qi H, Xiong Y, journalName=Proceedings of the IEEE International Conference on Computer Vision, refType=null, unstructuredReference=Dai J, Qi H, Xiong Y, et al. Deformable convolutional networks[C]// Proceedings of the IEEE International Conference on Computer Vision. New York: IEEE, 2017: 764-773., articleTitle=Deformable convolutional networks, refAbstract=null), Reference(id=1180089385207873813, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[19], rfOrder=24, authorNames=Liu Y, Shao Z, Hoffmann N, journalName=arXiv Preprint, refType=null, unstructuredReference=Liu Y, Shao Z, Hoffmann N. Global attention mechanism: retain information to enhance channel-spatial interactions[J]. arXiv Preprint, 2021: arXiv: 2112.05561., articleTitle=Global attention mechanism: retain information to enhance channel-spatial interactions, refAbstract=null), Reference(id=1180089385258205462, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=558, pageEnd=573, url=null, language=null, rfNumber=[20], rfOrder=25, authorNames=Bolya D, Foley S, Hays J, journalName=Computer Vision-ECCV 2020: 16th European Conference, refType=null, unstructuredReference=Bolya D, Foley S, Hays J, et al. Tide: a general toolbox for identifying object detection errors[C]// Computer Vision-ECCV 2020: 16th European Conference. Glasgow: Springer International Publishing, 2020: 558-573., articleTitle=Tide: a general toolbox for identifying object detection errors, refAbstract=null), Reference(id=1180089385312731415, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=7132, pageEnd=7141, url=null, language=null, rfNumber=[21], rfOrder=26, authorNames=Hu J, Shen L, Sun G, journalName=Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, refType=null, unstructuredReference=Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018: 7132-7141., articleTitle=Squeeze-and-excitation networks, refAbstract=null), Reference(id=1180089385375645976, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=11534, pageEnd=11542, url=null, language=null, rfNumber=[22], rfOrder=27, authorNames=Wang Q, Wu B, Zhu P, journalName=Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, refType=null, unstructuredReference=Wang Q, Wu B, Zhu P, et al. ECA-Net: efficient channel attention for deep convolutional neural networks[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 11534-11542., articleTitle=ECA-Net: efficient channel attention for deep convolutional neural networks, refAbstract=null), Reference(id=1180089385434366233, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=3, pageEnd=19, url=null, language=null, rfNumber=[23], rfOrder=28, authorNames=Woo S, Park J, Lee J Y, journalName=Proceedings of the European Conference on Computer Vision(ECCV), refType=null, unstructuredReference=Woo S, Park J, Lee J Y, et al. CBAM: convolutional block attention module[C]// Proceedings of the European Conference on Computer Vision(ECCV). New York: IEEE, 2018: 3-19., articleTitle=CBAM: convolutional block attention module, refAbstract=null)], funds=[Fund(id=1180089383601455356, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, awardId=桂科AA21077008, language=CN, fundingSource=广西创新驱动发展专项(桂科AA21077008), fundOrder=null, country=null)], companyList=[AuthorCompany(id=1180089380233429189, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, xref=null, ext=[AuthorCompanyExt(id=1180089380241817798, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, companyId=1180089380233429189, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=Information and Communication College, Guilin University of Electronic and Technology, Guilin 541010, China), AuthorCompanyExt(id=1180089380250206407, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, companyId=1180089380233429189, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=桂林电子科技大学信息与通信学院, 桂林 541010)])], figs=[ArticleFig(id=1180089381344919770, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.1, caption=Network structure of DGA-YOLOv8, figureFileSmall=QC5KRQIjzsk0cpEq3KtaNA==, figureFileBig=JXVOkKkN/p+vCh4MkbKW5Q==, tableContent=null), ArticleFig(id=1180089381407834331, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图1, caption=DGA-YOLOv8网络结构

Backbone为骨干网络;Conv为卷积;C2f为跨阶段双分支特征融合模块;C2f_DCN为本文设计模块,其中DCN为可变形卷积网络;GAM为全局注意力机制;SPPF为空间金字塔池化快速模块;Neck为颈部网络;Concat为拼接;Upsample为上采样;Head为头部网络;Detect为检测模块;Conv2d为二维卷积;Box_Loss为锚框损失;Cls_Loss为类别损失;BatchNorm2d为2维批量归一化;SiLU为激活函数;Split为分割;DCN_Bottleneck为本文设计模块,其中Bottleneck为瓶颈模块;Deformable Conv为可变形卷积;shortcut=True为使用残差连接;MaxPool2d为2维最大值池化

, figureFileSmall=QC5KRQIjzsk0cpEq3KtaNA==, figureFileBig=JXVOkKkN/p+vCh4MkbKW5Q==, tableContent=null), ArticleFig(id=1180089381529469148, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.2, caption=Traditional convolution and deformable convolution sampling processes, figureFileSmall=5TD0Ypwgo2++JnMvIb6DAQ==, figureFileBig=A3lxrymauerzc6+p54U4jw==, tableContent=null), ArticleFig(id=1180089381592383709, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图2, caption=传统卷积与可变形卷积采样过程, figureFileSmall=5TD0Ypwgo2++JnMvIb6DAQ==, figureFileBig=A3lxrymauerzc6+p54U4jw==, tableContent=null), ArticleFig(id=1180089381680464094, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.3, caption=C2f_DCN module, figureFileSmall=91lGwksPDj1WYH7VvC12Qw==, figureFileBig=OK4ikRLv3LnSrqj0qg1lfA==, tableContent=null), ArticleFig(id=1180089381743378655, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图3, caption=C2f_DCN模块, figureFileSmall=91lGwksPDj1WYH7VvC12Qw==, figureFileBig=OK4ikRLv3LnSrqj0qg1lfA==, tableContent=null), ArticleFig(id=1180089381797904608, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.4, caption=DCN_Bottleneck structure, figureFileSmall=+FIvzloMbi8OLXxKfDUP2g==, figureFileBig=9JOWdBZAO+dpgd1SEAw/oQ==, tableContent=null), ArticleFig(id=1180089381865013473, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图4, caption=DCN_Bottleneck结构, figureFileSmall=+FIvzloMbi8OLXxKfDUP2g==, figureFileBig=9JOWdBZAO+dpgd1SEAw/oQ==, tableContent=null), ArticleFig(id=1180089381932122338, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.5, caption=GAM attention module, figureFileSmall=kUq8OIu6XNmzhzzCYu7rlw==, figureFileBig=WP4VBSyb96zqyA748GQ3fw==, tableContent=null), ArticleFig(id=1180089381990842595, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图5, caption=GAM注意力模块

Input features为输入特征;Channel Attention为通道注意力;Spatial Attention为空间注意力;Output features为输出特征

, figureFileSmall=kUq8OIu6XNmzhzzCYu7rlw==, figureFileBig=WP4VBSyb96zqyA748GQ3fw==, tableContent=null), ArticleFig(id=1180089382066340068, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.6, caption=Channel attention module, figureFileSmall=iQGPLvgHPVT/sE3cokB+9g==, figureFileBig=TfF/7zLf9YFAH3is9YKwrA==, tableContent=null), ArticleFig(id=1180089382129254629, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图6, caption=通道注意力模块

permutation为维度转换;MLP为多层感知机;reverse permutation为维度反转换;Sigmoid为激活函数

, figureFileSmall=iQGPLvgHPVT/sE3cokB+9g==, figureFileBig=TfF/7zLf9YFAH3is9YKwrA==, tableContent=null), ArticleFig(id=1180089382187974886, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.7, caption=Spatial attention module, figureFileSmall=9fHQUV6dCFGNBkvDEVeMMg==, figureFileBig=vcNrr1RJPOQc1BHnfA5ttg==, tableContent=null), ArticleFig(id=1180089382259278055, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图7, caption=空间注意力模块

Conv为卷积

, figureFileSmall=9fHQUV6dCFGNBkvDEVeMMg==, figureFileBig=vcNrr1RJPOQc1BHnfA5ttg==, tableContent=null), ArticleFig(id=1180089382309609704, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.8, caption=TIDE error type definition, figureFileSmall=46KvQj8H3vNshBSaKHPbFg==, figureFileBig=M23FEwemhm/xoXhanxKSmQ==, tableContent=null), ArticleFig(id=1180089382368329961, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图8, caption=TIDE错误类型定义

分类错误(Cls)表示定位正确,但分类错误;定位错误(Loc)表示分类正确,但定位错误;分类和定位错误(Cls+Loc)表示分类和定位均错误;重复检测错误(Duplicate)表示多次检测到目标物体;背景错误(Bkgd)表示将背景检测为目标物体;未检测错误(Missed)表示未检测到目标物体

, figureFileSmall=46KvQj8H3vNshBSaKHPbFg==, figureFileBig=M23FEwemhm/xoXhanxKSmQ==, tableContent=null), ArticleFig(id=1180089382431244522, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.9, caption=Training process analysis, figureFileSmall=Q+lWd7P7s62mZ3GafHavDA==, figureFileBig=e4xQtyGEJdifGgF69NXglw==, tableContent=null), ArticleFig(id=1180089382485770475, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图9, caption=训练过程分析, figureFileSmall=Q+lWd7P7s62mZ3GafHavDA==, figureFileBig=e4xQtyGEJdifGgF69NXglw==, tableContent=null), ArticleFig(id=1180089382536102124, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.10, caption=Comparison of detection result, figureFileSmall=AaAMrwgEg+Czqgrv4r6XAg==, figureFileBig=igNJU+6xCMdK2yJyqI5G8Q==, tableContent=null), ArticleFig(id=1180089382586433773, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图10, caption=检测结果对比

绿色框为TP;蓝色框为FP;红色框为FN

, figureFileSmall=AaAMrwgEg+Czqgrv4r6XAg==, figureFileBig=igNJU+6xCMdK2yJyqI5G8Q==, tableContent=null), ArticleFig(id=1180089382640959726, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Fig.11, caption=Comparison of heat map effects, figureFileSmall=lQSEnts/HQajtEer4gBi8Q==, figureFileBig=nHdeAkhV8wJiTUtzO1HIsg==, tableContent=null), ArticleFig(id=1180089382699679983, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=图11, caption=热力图效果对比, figureFileSmall=lQSEnts/HQajtEer4gBi8Q==, figureFileBig=nHdeAkhV8wJiTUtzO1HIsg==, tableContent=null), ArticleFig(id=1180089382754205936, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Table 1, caption=

Experimental environment

, figureFileSmall=null, figureFileBig=null, tableContent=
参数 配置
图像大小 640×640
迭代次数 200
批量大小 16
优化器 SGD
初始学习率 0.01
学习率衰减因子 0.01
权重衰减 0.000 5
动量 0.937
), ArticleFig(id=1180089382821314801, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=表1, caption=

实验环境

, figureFileSmall=null, figureFileBig=null, tableContent=
参数 配置
图像大小 640×640
迭代次数 200
批量大小 16
优化器 SGD
初始学习率 0.01
学习率衰减因子 0.01
权重衰减 0.000 5
动量 0.937
), ArticleFig(id=1180089382880035058, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Table 2, caption=

Dataset distribution

, figureFileSmall=null, figureFileBig=null, tableContent=
船只类型 训练集/张 验证集/张 总计/张
矿砂船 1 508 168 1 676
散装货船 1 325 148 1 473
普通货船 1 188 133 1 321
集装箱货船 1 052 117 1 169
渔船 1 154 129 1 283
客运船 378 43 421
邮轮 795 89 884
大型货轮 1 513 169 1 682
帆船 202 23 225
其他类型 858 89 947
总计 9 973 1 108 11 081
), ArticleFig(id=1180089382942949619, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=表2, caption=

数据集分布

, figureFileSmall=null, figureFileBig=null, tableContent=
船只类型 训练集/张 验证集/张 总计/张
矿砂船 1 508 168 1 676
散装货船 1 325 148 1 473
普通货船 1 188 133 1 321
集装箱货船 1 052 117 1 169
渔船 1 154 129 1 283
客运船 378 43 421
邮轮 795 89 884
大型货轮 1 513 169 1 682
帆船 202 23 225
其他类型 858 89 947
总计 9 973 1 108 11 081
), ArticleFig(id=1180089383010058484, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Table 3, caption=

Module ablation experiment

, figureFileSmall=null, figureFileBig=null, tableContent=
YOLOv8n C2f_DCN GAM 准确
率/%
召回
率/%
mAP@0.5/
%
mAP@0.5:
0.95/%
93.1 86.1 91.8 73.1
94.7 86.3 91.4 73.1
95.4 87.4 91.4 73.5
96.4 87.3 92.2 74.1
), ArticleFig(id=1180089383093944565, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=表3, caption=

模块消融实验

, figureFileSmall=null, figureFileBig=null, tableContent=
YOLOv8n C2f_DCN GAM 准确
率/%
召回
率/%
mAP@0.5/
%
mAP@0.5:
0.95/%
93.1 86.1 91.8 73.1
94.7 86.3 91.4 73.1
95.4 87.4 91.4 73.5
96.4 87.3 92.2 74.1
), ArticleFig(id=1180089383165247734, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Table 4, caption=

Comparison of TIDE results

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 Cls Loc Both Duplicate Bkg Missed FP FN
YOLOv8 1.57 1.39 0.12 0.17 1.15 0.58 5.56 2.28
DGA-YOLOv8 1.60 1.19 0.14 0.18 1.10 0.65 5.53 2.08
提升 +0.03 -0.20 +0.02 +0.01 -0.05 +0.07 -0.03 -0.20
), ArticleFig(id=1180089383232356599, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=表4, caption=

TIDE结果对比

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 Cls Loc Both Duplicate Bkg Missed FP FN
YOLOv8 1.57 1.39 0.12 0.17 1.15 0.58 5.56 2.28
DGA-YOLOv8 1.60 1.19 0.14 0.18 1.10 0.65 5.53 2.08
提升 +0.03 -0.20 +0.02 +0.01 -0.05 +0.07 -0.03 -0.20
), ArticleFig(id=1180089383291076856, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Table 5, caption=

Comparison of different attention mechanisms

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 准确率/
%
召回率/
%
mAP@0.5/
%
mAP@0.5:0.95/
%
YOLOv8n 93.1 86.1 91.8 73.1
+SE 95.5 86.7 90.9 71.7
+ECA 93.9 84.9 91.1 72.5
+CBAM 92.9 87.2 91.4 72.6
+GAM 95.4 87.4 91.4 73.5
), ArticleFig(id=1180089383345602809, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=表5, caption=

不同注意力机制比较

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 准确率/
%
召回率/
%
mAP@0.5/
%
mAP@0.5:0.95/
%
YOLOv8n 93.1 86.1 91.8 73.1
+SE 95.5 86.7 90.9 71.7
+ECA 93.9 84.9 91.1 72.5
+CBAM 92.9 87.2 91.4 72.6
+GAM 95.4 87.4 91.4 73.5
), ArticleFig(id=1180089383408517370, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=EN, label=Table 6, caption=

Comparison experiment of different algorithms

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 准确率/
%
召回率/
%
mAP@0.5/
%
FPS 体积/
MB
Faster R-CNN 86.5 77.7 86.9 8.00 108.00
YOLOv3 93.1 86.4 90.4 6.33 235.00
YOLOv5s 92.4 89.7 90.2 27.77 13.80
YOLOv7-tiny 92.1 85.3 91.0 38.81 11.70
YOLOv8n 93.1 86.1 91.8 43.11 5.98
CBAM-YOLOv8[17] 91.6 87.8 91.1 52.18 7.00
YOLOv9s 95.3 87.6 91.1 24.13 19.30
YOLOv10n 95.5 85.9 90.6 65.79 5.53
DGA-YOLOv8 96.4 87.3 92.2 43.55 10.60
), ArticleFig(id=1180089383467237627, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149774733274739464, language=CN, label=表6, caption=

不同算法对比实验

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 准确率/
%
召回率/
%
mAP@0.5/
%
FPS 体积/
MB
Faster R-CNN 86.5 77.7 86.9 8.00 108.00
YOLOv3 93.1 86.4 90.4 6.33 235.00
YOLOv5s 92.4 89.7 90.2 27.77 13.80
YOLOv7-tiny 92.1 85.3 91.0 38.81 11.70
YOLOv8n 93.1 86.1 91.8 43.11 5.98
CBAM-YOLOv8[17] 91.6 87.8 91.1 52.18 7.00
YOLOv9s 95.3 87.6 91.1 24.13 19.30
YOLOv10n 95.5 85.9 90.6 65.79 5.53
DGA-YOLOv8 96.4 87.3 92.2 43.55 10.60
)], 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.1671-1815.2403743, detailUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/10.12404/j.issn.1671-1815.2403743, pdfUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/PDF/10.12404/j.issn.1671-1815.2403743, pdfUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/PDF/10.12404/j.issn.1671-1815.2403743, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
基于改进YOLOv8的船只目标检测算法
收藏切换
PDF下载
董从鑫 , 刘庆华 *
科学技术与工程 | 论文·自动化技术、计算机技术 2025,25(12): 5093-5102
收起
收藏切换
科学技术与工程 | 论文·自动化技术、计算机技术 2025, 25(12): 5093-5102
基于改进YOLOv8的船只目标检测算法
全屏
董从鑫 , 刘庆华*
作者信息
  • 桂林电子科技大学信息与通信学院, 桂林 541010
  • 董从鑫(2001—),男,汉族,广西贺州人,硕士研究生。研究方向:目标检测。E-mail:

通讯作者:

* 刘庆华(1974—),女,汉族,四川南江人,博士,教授,博士研究生导师。研究方向:阵列信号处理。E-mail:
Ship Target Detection Algorithm Based on Improved YOLOv8
Cong-xin DONG , Qing-hua LIU*
Affiliations
  • Information and Communication College, Guilin University of Electronic and Technology, Guilin 541010, China
出版时间: 2025-04-28 doi: 10.12404/j.issn.1671-1815.2403743
文章导航
收藏切换

针对传统船只目标检测算法的精度较低、船只检测类别单一等问题,提出一种改进的DGA-YOLOv8海上船只目标检测算法。首先,网络采用可变形卷积扩大模型的感受野,通过引入可学习偏移量,使模型能够根据物体实际形状自适应调整感受野大小和形状,确保卷积区域能够精确覆盖船只物体的轮廓。其次,引入GAM(global attention mechanism)注意力机制,使网络能够有效突出船只目标的关键特征,提高目标识别能力。实验结果表明:改进后算法的精确度和平均精度均值(mAP)达到96.4%和92.2%,FPS(frame per second)为43.55,在提升精度的同时也保证了一定的检测速度,满足了实时性检测的需求。同时与其他主流算法对比,其中包括Faster R-CNN(faster region-based convolutional neural network)、YOLOv5s和YOLOv10等。结果表明:所提算法具有更高的平均精度和更显著的分类效果。

YOLOv8  /  船只目标检测  /  可变形卷积  /  注意力机制

An improved DGA-YOLOv8 offshore ship target detection algorithm was proposed to tackle the issues of low accuracy and single ship detection categories that are present in traditional ship target detection algorithms. Firstly, the network was adapted to include deformable convolution, which expanded the model's receptive field. Learnable offsets were introduced, allowing the model to adaptively adjust the size and shape of the receptive field in response to the actual shape of the object, ensuring that the convolution area can precisely cover the contour of the ship object. Secondly, the incorporation of a GAM(global attention mechanism) attention mechanism enabled the network to effectively emphasize the key features of ship targets, thereby enhancing the target recognition capability. The experimental results demonstrate that the improved algorithm achieves accuracy and average accuracy mean (mAP) of 96.4% and 92.2%, respectively. An frames per second(FPS) of 43.55 is recorded, indicating not only an enhancement in accuracy but also the maintenance of a certain detection speed, thus fulfilling the requirements for real-time detection. When compared with other mainstream algorithms, such as faster region-based convolutional neural network(Faster R-CNN) and YOLOv5s, YOLOv10. The results show that the proposed algorithm exhibits higher average accuracy and significant superior classification performance.

YOLOv8  /  ship target detection  /  deformable convolution  /  attention mechanism
董从鑫, 刘庆华. 基于改进YOLOv8的船只目标检测算法. 科学技术与工程, 2025 , 25 (12) : 5093 -5102 . DOI: 10.12404/j.issn.1671-1815.2403743
Cong-xin DONG, Qing-hua LIU. Ship Target Detection Algorithm Based on Improved YOLOv8[J]. Science Technology and Engineering, 2025 , 25 (12) : 5093 -5102 . DOI: 10.12404/j.issn.1671-1815.2403743
随着海上运输行业的不断发展,交通环境日渐变得复杂,船只失事等交通事故日趋增多。此外,由于海洋处于国家边境地带,常有船只走私、偷渡等危害国家安全利益的事态发生。因此,做好海上安防管理、船只监督与检测工作以及开展相关研究具有尤为重要的意义。
目前,常用海上船只目标检测的方法主要分为传统船只检测方法和基于深度学习的船只目标检测方法。传统船只检测方法主要包括使用合成孔径雷达(synthetic aperture radar,SAR)和通过卫星遥感图像等方式[1]。石洪基等[2]优化了SAR图像中的船只目标,改进区域卷积神经网络输入和目标候选框提取,实现了多分辨率SAR图像中密集分布、沿岸分布等场景下的船只检测。王程丽等[3]针对SAR船只目标样本量少的问题,提出了一种基于VGG16迁移学习的识别方法。对模型进行参数微调,解决过拟合和局部最优问题,能够有效地提高SAR船只目标的识别准确率。刘方坚等[4]提出的基于视觉显著性的NanoDet舰船检测方法,通过自动聚类划分场景、差异化显著性检测、优化模型学习显著性图特征实现SAR舰船图像的快速和高精确度检测效果。传统的方法虽然可以实现船只检测,但也有一定的缺陷,SAR的图像容易产生各种噪点,且易受到海岸线背景等因素影响;卫星遥感方式虽然具有相对直观的光学图像,但是易受天气条件影响,且无法准确识别出目标外形的特点。
随着深度学习技术的发展,深度学习模型具有更高的检测精度和实时性能,能够准确和高效地完成船只检测任务。在深度学习中,目标检测主要分为一阶段算法和二阶段算法[5]。二阶段算法需要先产生候选区域,再进行目标检测任务,如R-CNN(region-based convolutional neural network)[6]、Fast R-CNN(fast region-based convolutional neural network)[7]和Faster R-CNN(faster region-based convolutional neural network)[8]等。
一阶段算法可以同时完成分类和定位任务,YOLO系列以其优越的性能被广泛应用于目标检测领域。2016年,Redmon等[9-10]提出了YOLOv1、YOLOv2算法,在此基础上,后续研究人员们提出了YOLOv3[11]、YOLOv4[12]、YOLOv5、YOLOv7[13]等算法。李静等[14]优化了YOLOv3模型中的K-means聚类算法,使聚类中心之间的距离最大化,实现全局最优结果,显著提升了船只目标检测的准确率,整体mAP(平均精度均值)提高了9.31%。Liu等[15]基于YOLOv5算法,结合GhostbottleNet特征提取技术,优化特征提取过程,实现更高精度和实时性的船舶检测。Wang等[16]提出了一种基于YOLOv4的船舶目标检测算法,通过优化锚框尺寸、调整模型结构和应用Mixup数据增强技术,提高了检测的准确性和实时性。Zhang[17]提出了一种基于CBAM(convolutional block attention module)-YOLOv8的船舰目标检测算法,使用空间深度卷积提高特征利用率,加入CBAM注意力融合特征信息,并采用MPDIOU损失函数,实现了较高的精度和检测速度。
上述研究中,R-CNN系列算法虽然检测精度较高,但检测速度较低,不适用于实时检测任务,同时模型体积偏大,不利于在硬件上部署;使用YOLO系列算法的研究方案在应用于遥感图像处理时,无法准确识别到目标的外观形状等特点,同时对于船只较小或者图像分辨率较低时,难以准确检测,对船只的检测类别单一。因此,设计一种精度更高、分类效果更加显著的海上船只目标检测算法具有重要意义。
鉴于此,提出一种基于改进YOLOv8的船只目标检测算法DGA-YOLOv8,网络中使用可变形卷积DCNv2(deformable ConvNets v2)[18],通过引入可学习的偏移量,使感受野不再是固定的矩形形状,能根据物体形状进行调整,调整后的卷积区域能够更好地覆盖物体形状周围,实现感受野的扩大。同时引入GAM(global attention mechanism)[19]注意力机制,在特征提取后进行注意力的调整,丰富特征信息,提高检测精度。
由于检测需要同时兼顾精度与速度,以YOLOv8系列中最轻量级的YOLOv8n模型为基线模型,并在此基础上加以改进,改进后的网络结构如图1所示。
将可变形卷积DCN和GAM注意力机制加入网络中,设计DGA-YOLOv8算法,对比原始的YOLOv8网络,DGA-YOLOv8进行以下改进。
(1)引入可变形卷积替换Bottleneck结构当中的传统卷积方式,设计C2f_DCN模块,将原始结构当中的Backbone和Neck部分结构的C2f模块改为C2f_DCN模块,使网络在应对不同尺度、形状的目标时能展现出更好的应对能力。将C2f_DCN模块放置于不同位置时,进行相关实验,如单独更改Backbone部分和单独更改Neck部分的C2f模块,最终可以产生最优的效果,如图1所示。
(2)在Backbone结构中融入注意力机制,将GAM注意力模块添加到SPPF结构前,增强主干网络的特征提取的能力;同时在Neck与Head的连接部分中,分别在3个Detect检测头前加入GAM注意力,增强了特征融合的能力,提高了细节信息的利用率。
在YOLOv8中加入可变形卷积DCN和GAM注意力机制后,网络的参数量和计算量分别增加2.3 M、1.2GFLOPs(每秒十亿次的浮点运算数),但其网络宽度和深度基本保持不变,同时增加量在网络中的占比有限,目的是通过引入少量的额外参数和计算量来实现模型的精度与性能提升。
可变形卷积通过在传统卷积的基础上引入可学习的偏移量来增加网络对船只物体形状的变化适应性,偏移量通过传统卷积计算得出,使得卷积核的采样位置可以根据特征图的内容进行动态调整。相比于传统卷积方式,可以更灵活地适应输入特征的变化,如图2所示。通过引入可变形卷积,网络模型可以更好地适应不同尺度、形状和角度的船只物体,有利于提高网络模型在船只检测中的性能和精度。
设输入特征图为x, w为卷积核,大小为3×3,y为输出特征图,可变形卷积的计算公式为
y(p0)= p n R 'w(pn)x(p0+pn+Δpn)
式(1)中:p0为特征图上的每个像素点的位置;pn为卷积核窗口中其他像素点与中心像素点的相对位置;Δpn为位置偏移量;R'定义了卷积核大小以及卷积核周围像素点的偏移,可表示为
R'={(-1,-1),(-1,0),…,(0,0),(1,1)}
在YOLOv8网络结构中,C2f模块中使用的传统卷积方式只能对规则的矩形区域进行特征提取,虽然在处理简单或规则的数据时效果良好,但针对海上复杂目标时无法完全拟合船只目标的形状和姿态。因此,在处理尺度不同、形状复杂的目标时,原本的网络具有一定的局限性,当出现噪声或者背景干扰时,漏检或错检的可能性会提升。
为提高模型对不同船只物体的特征提取能力和对复杂目标的适应能力,利用可变形卷积设计C2f_DCN模块,将原本C2f模块中的卷积方式用可变形卷积替换,其结构如图3图4所示。首先对输入的特征图进行卷积操作改变通道数,再通过Split操作切分特征,接着通过堆叠多个可变形卷积模块来扩大网络的感受野。使用可变形卷积可以自适应地调整采样点的位置,从而不断调节感受野的尺寸和大小,使网络能够更好地适应输入特征的几何变化,捕获到更多的对象信息。
在网络中引入注意力机制来增强网络对船只的特征提取能力。注意力机制可以有效帮助模型聚焦于输入图像中最有信息量的部分,并忽略掉无关的背景信息,使模型能获取到更多有效的特征。在网络模型中加入GAM注意力机制,有助于模型更好地捕捉全局信息,提高特征的表达能力。其模块结构如图5所示,其结合了通道注意力机制和空间注意力机制两部分,不仅能够关注通道信息,同时还重视空间信息,能够在减少信息弥散的情况下,同时放大全局维度交互特征[19]。输入特征图为F1,先后经过通道注意力模块和空间注意力模块计算生成通道权重和空间权重,将通道权重和空间权重应用到原始特征图上,得到输出特征图F3
通道注意力子模块结构如图6所示,对于给定的输入特征图F1,将特征图从原本的C×W×H(其中,C为通道数,W为宽度,H为高度)进行维度转换为W×H×C,将维度转换后的特征图经过多层感知机MLP,再转换回原来的维度,之后经过Sigmoid激活函数输出。
空间注意力子模块结构如图7所示,将经过通道注意力子模块的输出特征图F2依次经过两次7×7的卷积进行特征信息融合,最后再通过Sigmoid输出。
整个过程的定义为
$F_{2}=M_{\mathrm{c}}\left(F_{1}\right) \otimes F_{1}$
$F_{3}=M_{\mathrm{s}}\left(F_{2}\right) \otimes F_{2}$
式中:Mc为通道注意力模块;Ms为空间注意力模块;$\otimes$表示逐元素相乘。
本次实验在Windows10操作系统下进行,GPU使用NVIDIA TITAN RTX,显存为24 GB,采用Pytorch深度学习框架进行模型训练,运行环境为:CUDA 11.1、Python 3.8.18、Pytorch 1.9.0。
基于对数据集特性、计算资源和硬件环境的综合考虑,将训练过程中的具体参数设置为如表1所示。为了满足算法结构并兼顾计算效率,将图像大小设置为640×640。训练200次足以使模型达到稳定的性能,避免了欠拟合或过拟合。批量大小选择为16,避免内存溢出;SGD在深度学习中有着良好的性能,因此选择SGD作为优化算法。初始学习率和学习率衰减因子设为0.01以加快初期训练并防止出现局部最优解。权重衰减值设为0.000 5以简化模型;动量值设为0.937可促进模型快速收敛。
本次实验所使用的数据集为自制的海上船只数据集,其主要包括公开的SeaShips海上船只数据集7 000张、网络搜集图片、航道实拍船只图片等共计11 081张图片,具体数据分布如表2所示。
为了验证网络模型的性能,本次实验主要采用准确率(precision,记为P)、召回率(recall,记为R)和平均精度均值mAP(mean average precision)对模型检测性能进行评估。其中准确率是指正确预测的样本(TP)在全部预测为正的样本结果中(TP+FP)所占的比例,其计算公式为
P= T P T P + F P
召回率则是指正确预测的样本数量(TP)在实际为正的样本数量(TP+FN)中所占的比例,其计算公式为
R= T P T P + F N
对于每个类别来说,P-R曲线与坐标轴所围成的面积即为该类别的平均精度值(AP),将各类别的AP取平均便得到mAP,即代表整个数据集中所有类别AP的均值,两者的计算公式分别为
AP= 0 1 P(R)dR
mAP= 1 N i = 1 NAPi
式中:N为数据集中样本的类别数;APi为第i个类别的平均精度值。
mAP值作为模型性能的核心指标,综合考虑了精确度和召回率,更能够提供一个全面且直观的评估。mAP@0.5为在IoU(intersection over union)阈值设为0.5时计算得到的模型的平均精度均值,mAP@0.5:0.95为在IoU阈值由0.5~0.95的区间(步长为0.05)内变化时,计算得到的模型的平均精度均值。
为了更全面地评估算法性能,使用TIDE(toolkit for identifying detection and segmentation errors)[20],一个用于分析目标检测和实例分割算法中错误来源的框架和工具包。在TIDE中定义6种错误类型,如图8所示。
主要进行消融实验和算法对比实验,在消融实验中,针对算法的不同模块进行系统性能测试与分析,观察到不同模块对最终性能产生的具体影响。将本文算法与现有的其他主流算法进行对比,通过在同一数据集上进行实验,比较不同算法在各关键指标上的表现。
为了分析不同改进模块的有效性,进行消融实验,对比各个模块对检测模型性能的影响。表3为以YOLOv8n为基线模型,依次改进可变形卷积、全局注意力机制的方式进行消融实验的结果。
表3可知,通过改进C2f_DCN,模型的精确度有了良好的提升,说明网络使用可变形卷积后,面对复杂的几何变换时能够保持稳定的性能,对目标特征的提取能力有了较好的提升。在加入GAM注意力模块之后,对模型的精确度有了进一步的提升,准确率和召回率分别提高了2.3%和1.3%,说明加入注意力能降低模型对目标的错检概率,能够更好地区分背景和船只目标,提升了模型的抗干扰能力。将C2f_DCN模块和GAM注意力机制同时加入网络时,形成DGA-YOLOv8方法,则准确率、召回率、mAP@0.5和mAP@0.5:0.95分别提升了3.3、1.2、0.4和1.0个百分点,实验结果表明,当C2f_DCN和GAM两者一起作用于网络时,取得了最优的效果。
图9为本文模型DGA-YOLOv8和YOLOv8模型在训练过程中的性能指标变化曲线,包括锚框损失、类别损失、mAP@0.5和mAP@0.5:0.95随着训练轮数的增加而变化的曲线。从图9中可以看出,DGA-YOLOv8与YOLOv8模型相比,其训练时的损失下降得更快,模型所需要的收敛时间更短。mAP的变化同样也表明本文模型性能更优于YOLOv8模型,并进一步证明将可变形卷积和GAM注意力机制加入网络中对提高性能有重要作用。
使用TIDE工具包计算YOLOv8算法和DGA-YOLOv8算法的各项错误类别,结果如表4所示。
表4结果可以看出,改进后的模型降低了定位和背景检测的错误率。具体而言,使用可变形卷积能够帮助模型更好地定位目标物体,提高检测灵活性和准确性;注意力的引入使模型在检测过程中关注到更多重要信息,对目标和背景的区分能力更强。此外,改进后的模型在改善错检和漏检方面也有更出色的表现,其加入的可变卷积和注意力机制能提升特征提取能力和抗干扰能力,不仅减少了对目标的误判,提升识别的准确性,还能更全面地检测出目标,降低了错误率。
为了验证GAM注意力机制的有效性,在训练条件相同的前提下,与其他注意力机制进行了对比实验,结果如表5所示。
表5可知,当网络中使用GAM注意力机制时,整体取得了最优的效果,具体而言,准确率、召回率和mAP@0.5:0.95分别提高了2.3、1.3和0.4个百分点,虽然mAP@0.5略有下降,但模型整体性能和效果上都得到了一定的优化。相比之下,加入SE(squeeze-and-excitation)[21]、ECA(efficient channel attention)[22]和CBAM(convolutional block attention module)[23]注意力机制后,mAP@0.5和mAP@0.5:0.95均出现下降趋势,表明模型的性能有所降低。
为了进一步验证本文算法的优势,将本文算法与其他目前较为热门的算法在同一数据集上、相同的训练参数下进行对比实验,结果如表6所示。
表6可知,Faster R-CNN和YOLOv3算法的FPS值都偏低,并且模型体积都较大。YOLOv5s算法在精度和mAP上略低于YOLOv3,但其FPS值提高到了27.77,同时在模型大小上,比YOLOv3减少了221.2 MB,与前两者对比模型更加轻量化。YOLOv7-tiny在精确度和mAP上与YOLOv5并没有显著的差异,FPS上则有了较大的提升。YOLOv8n在召回率上虽比YOLOv5s略低,但其精确度和mAP值都有了一定的提升,同时模型的体积也得到缩减,检测速度显著得到了提高。文献[17]提出的CBAM-YOLOv8算法虽然参数量和计算量得到了一定的降低,模型得到了轻量化,但是精度却低于原始模型。YOLOv9s算法在准确率、mAP方面均不及本文算法,同时检测速度较低;YOLOv10n虽然在检测速度上达到最高,但准确率和mAP均略低于本文算法。所提出的DGA-YOLOv8模型精确度和mAP值提升到了96.4%和92.2%,同时FPS也较高,模型大小在YOLOv8n的基础上仅增加4.62 MB,与其他模型相比,本文模型在准确率和速度方面都有着良好的表现。
图10为检测过程中的TP(绿色框)、FP(蓝色框)和FN(红色框)的可视化结果。通过可视化结果,可以对模型检测性能的各个方面进行评估和分析。
图10可知,原始的YOLOv8模型出现了较多的检错与漏检的现象,相比之下,所提出的DGA-YOLOv8算法在检错和漏检方面有了一定的改善。对于图中存在多数小目标船只的情况,本文算法对远处的小目标船只仍能实现正确的检测,提高了对小目标的识别准确性。其次,在漏检方面,改进后的算法能够检测出区域内的更多船只,检测的覆盖率得到了提高并降低了漏检率,同时对船只类别的分类效果更为准确。此外,从图10中可以看出,改进后的模型可以有效地从周围的环境干扰中将船只准确识别出来,对海上周围环境杂物有着更好的分辨能力。由此可见,所采取的改进方式不仅有效提升了模型的检测准确性,降低了误检和漏检的风险,同时还增强了模型对背景的抑制能力。
为了更直观地展示改进后的模型对特征提取能力的增强,使用Grad-CAM技术生成热力图进行展示。热力图可以直观地反映出模型所关注于特征图的哪些区域,通过这种可视化的方式,可以更好地了解模型用于预测的图像区域,图11为对比改进后模型与原始模型的特征提取过程。
图11可知,YOLOv8模型对图像中的船只目标,其所关注的只有少数部分船只,且只关注到了整体目标中的局部位置,并没有对其余小目标的船只给予更多的关注。相较于基线模型,在涉及小目标船只的场景中,本文方法能够更加聚焦于所有存在目标的区域,展现出更高的关注度。对于尺寸更小的船只目标,本文方法的检测效果也更为出色,从而实现了更为精准的目标识别与定位。针对尺寸较大的船只,本文模型能够更精准地聚焦于船体部分,确保对目标区域的有效识别。此外,从图11中可以看出,YOLOv8模型在特征提取过程中受到部分背景区域的干扰。而本文方法则对背景干扰有着更好地抑制能力,能够更有效地将注意力集中在船只目标上。
基于YOLOv8提出一种改进的海上船只目标检测算法。该算法使用可变形卷积并设计C2f_DCN模块替换原始模型中的C2f结构,通过引入可变形卷积,扩大了模型的感受野,使模型在特征提取时能够自适应地调整感受野大小和形状,以更好地捕捉输入数据的关键特征。此外,通过嵌入GAM注意力机制,网络对重要特征的提取能力得到了强化,同时能够抑制不相关的背景信息的干扰,增强对关键信息的关注能力。实验结果表明,改进后模型的精确度和mAP分别提高至96.4%和92.2%,在原始模型的基础上都有了良好的提升;在一定程度上兼顾了准确度和速度,同时模型大小为10.6 MB,满足硬件平台的部署和实时检测的任务需求。然而,模型的泛化能力仍需进一步地验证,在面对不同场景的船只时,网络性能可能会受到影响,同时其检测精度还有继续提升的空间。
在未来研究中,针对模型在小目标和遮挡物检测方面中的不足,可进一步优化网络,提升这些方面的检测能力;同时探索更轻量化的模型,在减少参数量的同时提升检测速度和效率。
  • 广西创新驱动发展专项(桂科AA21077008)
参考文献 引证文献
排序方式:
[1]
徐慧智, 宋爱秋, 武笑宇. 基于均匀设计的船舶目标检测深度学习模型训练方法研究[J]. 科学技术与工程, 2022, 22(25): 11241-11249.
Xu Huizhi, Song Aiqiu, Wu Xiaoyu. Research on training method of deep learning to ship target detection based on uniform design[J]. Science Technology and Engineering, 2022, 22(25): 11241-11249.
[2]
石洪基, 郎海涛, 宋棋, 等. 基于改进区域卷积神经网络的SAR图像船只检测方法[J]. 中国海洋大学学报(自然科学版), 2019, 49(S2): 185-191.
Shi Hongji, Lang Haitao, Song Qi, et al. Ship detection method in SAR images based on improved regional convolutional neural network[J]. Journal of Ocean University of China(Natural Science Edition), 2019, 49(S2): 185-191.
[3]
王程丽, 赵静, 杨攀攀, 等. 基于迁移学习的小样本SAR船只目标识别[J]. 移动通信, 2022, 46(4): 22-27.
Wang Chengli, Zhao Jing, Yang Panpan, et al. Smallsample SAR ship target recognition based on transfer learning[J]. Mobile Communications, 2022, 46(4): 22-27.
[4]
刘方坚, 李媛. 基于视觉显著性的 SAR 遥感图像 NanoDet 舰船检测方法[J]. 雷达学报, 2021, 10(6): 885-894.
Liu Fangjian, Li Yuan. NanoDet ship detection method for SAR remote sensing images based on visual saliency[J]. Radar Journal, 2021, 10(6): 885-894.
[5]
宣以国, 余成波, 蒋启超, 等. 基于改进YOLOv7的道路裂缝和坑洞检测算法[J]. 科学技术与工程, 2024, 24(17): 7205-7213.
Xuan Yiguo, Yu Chengbo, Jiang Qichao, et al. Improved YOLOv7 road crack and pothole detection algorithm[J]. Science Technology and Engineering, 2024, 24(17): 7205-7213.
[6]
Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2014: 580-587.
[7]
Girshick R. Fast R-CNN[C]// Proceedings of the IEEE International Conference on Computer Vision. New York: IEEE, 2015: 1440-1448.
[8]
Ren S, He K, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 39(6): 1137-1149.
[9]
Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2016: 779-788.
[10]
Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2017: 7263-7271.
[11]
Redmon J, Farhadi A. YOLOv3: an incremental improvement[J]. arXiv Preprint, 2018: arXiv: 1804.02767.
[12]
Bochkovskiy A, Wang C Y, Liao H Y M. YOLOv4: optimal speed and accuracy of object detection[J]. arXiv Preprint, 2020: arXiv: 2004.10934.
[13]
Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 7464-7475.
[14]
李静, 鲜林, 王海江. 基于YOLOv3的船只检测算法研究[J]. 成都信息工程大学学报, 2023, 38(1): 37-43.
Li Jing, Xian Lin, Wang Haijiang. Research on ship detection algorithm based on YOLOv3[J]. Journal of Chengdu University of Information Technology, 2023, 38(1): 37-43.
[15]
Liu T, Zhou B, Zhao Y, et al. Ship detection algorithm based on improved YOLOv5[C]// 2021 6th International Conference on Automation, Control and Robotics Engineering(CACRE). New York: IEEE, 2021: 483-487.
[16]
Wang B, Han B, Yang L. Accurate real-time ship target detection using YOLOv4[C]// 2021 6th International Conference on Transportation Information and Safety(ICTIS). NewYork: IEEE, 2021: 222-227.
[17]
Zhang J. Ship target detection based on CBAM-YOLOv8[C]// International Conference on Mechatronic Engineering and Artificial Intelligence. Shanghai: SPIE, 2024: 659-665.
[18]
Dai J, Qi H, Xiong Y, et al. Deformable convolutional networks[C]// Proceedings of the IEEE International Conference on Computer Vision. New York: IEEE, 2017: 764-773.
[19]
Liu Y, Shao Z, Hoffmann N. Global attention mechanism: retain information to enhance channel-spatial interactions[J]. arXiv Preprint, 2021: arXiv: 2112.05561.
[20]
Bolya D, Foley S, Hays J, et al. Tide: a general toolbox for identifying object detection errors[C]// Computer Vision-ECCV 2020: 16th European Conference. Glasgow: Springer International Publishing, 2020: 558-573.
[21]
Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018: 7132-7141.
[22]
Wang Q, Wu B, Zhu P, et al. ECA-Net: efficient channel attention for deep convolutional neural networks[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 11534-11542.
[23]
Woo S, Park J, Lee J Y, et al. CBAM: convolutional block attention module[C]// Proceedings of the European Conference on Computer Vision(ECCV). New York: IEEE, 2018: 3-19.
2025年第25卷第12期
PDF下载
274
100
引用本文
BibTeX
文章信息
doi: 10.12404/j.issn.1671-1815.2403743
  • 接收时间:2024-05-21
  • 首发时间:2025-07-09
  • 出版时间:2025-04-28
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2024-05-21
  • 修回日期:2025-01-22
基金
广西创新驱动发展专项(桂科AA21077008)
作者信息
    桂林电子科技大学信息与通信学院, 桂林 541010

通讯作者:

* 刘庆华(1974—),女,汉族,四川南江人,博士,教授,博士研究生导师。研究方向:阵列信号处理。E-mail:
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2403743
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