Article(id=1149781961000051646, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1149781952959574654, articleNumber=null, orderNo=null, doi=10.12404/j.issn.1671-1815.2402441, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=research-article, receivedDate=1712332800000, receivedDateStr=2024-04-06, revisedDate=1733414400000, revisedDateStr=2024-12-06, acceptedDate=null, acceptedDateStr=null, onlineDate=1752058981417, onlineDateStr=2025-07-09, pubDate=1743091200000, pubDateStr=2025-03-28, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1752058981417, onlineIssueDateStr=2025-07-09, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1752058981417, creator=13701087609, updateTime=1752058981417, updator=13701087609, issue=Issue{id=1149781952959574654, tenantId=1146029695717560320, journalId=1146123166801305609, year='2025', volume='25', issue='9', pageStart='3529', pageEnd='3967', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=0, createTime=1752058979501, creator=13701087609, updateTime=1776333392421, updator=13701087609, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1251596220226027613, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1149781952959574654, language=EN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1251596220226027614, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1149781952959574654, language=CN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=3778, endPage=3787, ext={EN=ArticleExt(id=1149781961205572543, articleId=1149781961000051646, tenantId=1146029695717560320, journalId=1146123166801305609, language=EN, title=Small and Medium-sized Obstacle Detection Methods in Intelligent Driving Scenarios, columnId=1156262729162810294, journalTitle=Science Technology and Engineering, columnName=Papers·Automation and Computational Technology, runingTitle=null, highlight=null, articleAbstract=

Aiming at the problems such as small and medium-sized obstacles on the road are prone to miss detection, small target obstacles are difficult to detect, and the number of model parameters is large in smart driving scenarios, the obstacle target detection algorithm with improved YOLOv8n was proposed. Distribution shifting convolution (DSConv) was used in the backbone network to replace floating point operation with integer operation, reducing the amount of redundant computation, and maintaining the accuracy by imitating the original convolution layer by quantization and distribution shifting. By adding small target detection layer, the feature information of small target can be captured better and the scale characteristics of small target can be adapted. Combined with SimAM parameterless attention mechanism, SPPF-SimAM module was introduced to improve the quality and diversity of feature representation, and the detection accuracy was improved without increasing the number of parameters. By combining ghost-shuffle convolution (GSConv) and VoV-GSCSP modules, the neck feature fusion network was lightweight, reducing the number of parameters and calculation of the model. The experimental results show that the accuracy, recall, and mean average precision of the improved model are improved by 1.6%, 8.0%, and 6.2%, respectively. The number of parameters is reduced by 6.7% compared with the original model, and the proposed algorithm effectively improves the detection accuracy of small and medium-sized obstacles in smart driving scenarios, and achieves a better balance between the detection performance and the model lightweighting.

, correspAuthors=Xin-yuan NAN, 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=Xiao-yu LONG, Xin-yuan NAN), CN=ArticleExt(id=1149781993971475440, articleId=1149781961000051646, tenantId=1146029695717560320, journalId=1146123166801305609, language=CN, title=智能驾驶场景下的中小型障碍物检测方法, 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=

针对智能驾驶场景下路面中小型障碍物易发生漏检、小目标障碍物难检测、模型参数量大等问题,提出了改进YOLOv8n的障碍物目标检测算法。在主干网络中融入分布移位卷积(distribution shifting convolution,DSConv),将浮点运算替换为整数运算,减少了冗余计算量,通过量化和分布移位的方式模仿原始卷积层,维持了准确率;通过添加小目标检测层,更好地捕捉小目标的特征信息,适配小目标的尺度特征;结合SimAM无参数注意力机制,引入SPPF-SimAM模块,提高特征表示的质量与多样性,在不增加参数量的情况下实现了检测精度的提升;通过组合鬼影混洗卷积(ghost-shuffle convolution,GSConv)和VoV-GSCSP模块的方式轻量化颈部特征融合网络,降低了模型的参数量和计算量。实验结果表明,改进后模型的准确率、召回率、平均精度均值相较于原始模型分别提升了1.6%、8.0%、6.2%,参数量降低了6.7%,所提算法有效提升了智能驾驶场景下中小型障碍物的检测精度,并且在检测性能与模型轻量化之间达到较好的平衡。

, correspAuthors=南新元, authorNote=null, correspAuthorsNote=
* 南新元(1967—),男,汉族,新疆乌鲁木齐人,硕士,教授。研究方向:智能控制系统开发、机器视觉。E-mail:
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=Nxp6AX12snMDpbxKN79g8w==, magXml=jhE60zYt3Tdt5NFKdt7YOw==, pdfUrl=null, pdf=MXtah3yFi03XwM1eI0wBLA==, pdfFileSize=8212545, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=/K8xPS9Kj3ruaXqii8vtFw==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=Isu36LHNjQutlcULPixHzA==, mapNumber=null, authorCompany=null, fund=null, authors=

龙小羽(2000—),女,汉族,新疆和静人,硕士研究生。研究方向:深度学习、目标检测。E-mail:

, authorsList=龙小羽, 南新元)}, authors=[Author(id=1251249353340436784, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=874480489@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1251249353428517175, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, authorId=1251249353340436784, language=EN, stringName=Xiao-yu LONG, firstName=Xiao-yu, middleName=null, lastName=LONG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Electrical Engineering, Xinjiang University, Urumqi 830017, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1251249353508208957, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, authorId=1251249353340436784, language=CN, stringName=龙小羽, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=新疆大学电气工程学院, 乌鲁木齐 830017, bio={"content":"

龙小羽(2000—),女,汉族,新疆和静人,硕士研究生。研究方向:深度学习、目标检测。E-mail:

"}, bioImg=null, bioContent=

龙小羽(2000—),女,汉族,新疆和静人,硕士研究生。研究方向:深度学习、目标检测。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1251249353264939303, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, xref=null, ext=[AuthorCompanyExt(id=1251249353269133608, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, companyId=1251249353264939303, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Electrical Engineering, Xinjiang University, Urumqi 830017, China), AuthorCompanyExt(id=1251249353277522217, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, companyId=1251249353264939303, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=新疆大学电气工程学院, 乌鲁木齐 830017)])]), Author(id=1251249353613066564, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=xynan@xju.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1251249353764061514, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, authorId=1251249353613066564, language=EN, stringName=Xin-yuan NAN, firstName=Xin-yuan, middleName=null, lastName=NAN, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=School of Electrical Engineering, Xinjiang University, Urumqi 830017, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1251249353956999508, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, authorId=1251249353613066564, language=CN, stringName=南新元, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=新疆大学电气工程学院, 乌鲁木齐 830017, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1251249353264939303, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, xref=null, ext=[AuthorCompanyExt(id=1251249353269133608, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, companyId=1251249353264939303, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Electrical Engineering, Xinjiang University, Urumqi 830017, China), AuthorCompanyExt(id=1251249353277522217, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, companyId=1251249353264939303, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=新疆大学电气工程学院, 乌鲁木齐 830017)])])], keywords=[Keyword(id=1251249355697635687, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, orderNo=1, keyword=obstacle detection), Keyword(id=1251249355861213551, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, orderNo=2, keyword=YOLOv8n), Keyword(id=1251249355953488248, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, orderNo=3, keyword=intelligent driving), Keyword(id=1251249356041568640, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, orderNo=4, keyword=small target detection), Keyword(id=1251249356154814858, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, orderNo=5, keyword=attention mechanism), Keyword(id=1251249356255478165, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, orderNo=1, keyword=障碍物检测), Keyword(id=1251249356343558558, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, orderNo=2, keyword=YOLOv8n), Keyword(id=1251249356448416168, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, orderNo=3, keyword=智能驾驶), Keyword(id=1251249356637159861, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, orderNo=4, keyword=小目标检测), Keyword(id=1251249356754600388, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, orderNo=5, keyword=注意力机制)], refs=[Reference(id=1251249363008308200, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2022, volume=58, issue=15, pageStart=68, pageEnd=77, url=null, language=null, rfNumber=[1], rfOrder=0, authorNames=茅智慧, 朱佳利, 吴鑫, journalName=计算机工程与应用, refType=null, unstructuredReference=茅智慧, 朱佳利, 吴鑫, . 基于YOLO的自动驾驶目标检测研究综述[J]. 计算机工程与应用, 2022, 58(15): 68-77., articleTitle=基于YOLO的自动驾驶目标检测研究综述, refAbstract=null), Reference(id=1251249363121554420, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2022, volume=58, issue=15, pageStart=68, pageEnd=77, url=null, language=null, rfNumber=[1], rfOrder=1, authorNames=Mao Zhihui, Zhu Jiali, Wu Xin, journalName=Computer Engineering and Applications, refType=null, unstructuredReference=Mao Zhihui, Zhu Jiali, Wu Xin, et al. Review of YOLO based target detection for autonomous driving[J]. Computer Engineering and Applications, 2022, 58(15): 68-77., articleTitle=Review of YOLO based target detection for autonomous driving, refAbstract=null), Reference(id=1251249364677641217, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=2102, pageEnd=null, url=null, language=null, rfNumber=[2], rfOrder=2, authorNames=Breitenstein J, Termöhlen J A, Lipinski D, journalName=ArXiv Preprint, refType=null, unstructuredReference=Breitenstein J, Termöhlen J A, Lipinski D, et al. Corner cases for visual perception in automated driving: some guidance on detection approaches[J]. ArXiv Preprint, 2021: 2102.05897., articleTitle=Corner cases for visual perception in automated driving: some guidance on detection approaches, refAbstract=null), Reference(id=1251249364849606670, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[3], rfOrder=3, authorNames=Gupta K, Javed S A, Gandhi V, journalName=ArXiv Preprint, refType=null, unstructuredReference=Gupta K, Javed S A, Gandhi V, et al. Mergenet: a deep net architecture for small obstacle discovery[J]. ArXiv Preprint, 2018: 1803.06508., articleTitle=Mergenet: a deep net architecture for small obstacle discovery, refAbstract=null), Reference(id=1251249365021573144, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2019, volume=47, issue=S1, pageStart=213, pageEnd=216, url=null, language=null, rfNumber=[4], rfOrder=4, authorNames=俞骏威, 张黎明, 陈凯, journalName=同济大学学报(自然科学版), refType=null, unstructuredReference=俞骏威, 张黎明, 陈凯, . 基于道路消失点的远距离路面微小障碍物检测[J]. 同济大学学报(自然科学版), 2019, 47(S1): 213-216., articleTitle=基于道路消失点的远距离路面微小障碍物检测, refAbstract=null), Reference(id=1251249365155790882, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2019, volume=47, issue=S1, pageStart=213, pageEnd=216, url=null, language=null, rfNumber=[4], rfOrder=5, authorNames=Yu Junwei, Zhang Liming, Chen Kai, journalName=Journal of Tongji University(Natural Science), refType=null, unstructuredReference=Yu Junwei, Zhang Liming, Chen Kai, et al. Long-distance small road obstacles detection based on road vanishing point[J]. Journal of Tongji University(Natural Science), 2019, 47(S1): 213-216., articleTitle=Long-distance small road obstacles detection based on road vanishing point, refAbstract=null), Reference(id=1251249365277425712, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2022, volume=23, issue=9, pageStart=16318, pageEnd=16329, url=null, language=null, rfNumber=[5], rfOrder=6, authorNames=Song T J, Jeong J, Kim J H, journalName=IEEE Transactions on Intelligent Transportation Systems, refType=null, unstructuredReference=Song T J, Jeong J, Kim J H. End-to-end real-time obstacle detection network for safe self-driving via multi-task learning[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(9): 16318-16329., articleTitle=End-to-end real-time obstacle detection network for safe self-driving via multi-task learning, refAbstract=null), Reference(id=1251249365436809277, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2023, volume=23, issue=5, pageStart=2051, pageEnd=2058, url=null, language=null, rfNumber=[6], rfOrder=7, authorNames=吴晨曦, 应保胜, 许小伟, journalName=科学技术与工程, refType=null, unstructuredReference=吴晨曦, 应保胜, 许小伟, . 基于改进单步多框目标检测的道路小目标检测算法[J]. 科学技术与工程, 2023, 23(5): 2051-2058., articleTitle=基于改进单步多框目标检测的道路小目标检测算法, refAbstract=null), Reference(id=1251249365571027016, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2023, volume=23, issue=5, pageStart=2051, pageEnd=2058, url=null, language=null, rfNumber=[6], rfOrder=8, authorNames=Wu Chenxi, Ying Baosheng, Xu Xiaowei, journalName=Science Technology and Engineering, refType=null, unstructuredReference=Wu Chenxi, Ying Baosheng, Xu Xiaowei, et al. Road small target detection algorithm based on improved single shot multibox detector[J]. Science Technology and Engineering, 2023, 23(5): 2051-2058., articleTitle=Road small target detection algorithm based on improved single shot multibox detector, refAbstract=null), Reference(id=1251249365742993498, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2024, volume=60, issue=5, pageStart=336, pageEnd=346, url=null, language=null, rfNumber=[7], rfOrder=9, authorNames=李旭, 宋焕生, 史勤, journalName=计算机工程与应用, refType=null, unstructuredReference=李旭, 宋焕生, 史勤, . CIEFRNet: 面向高速公路的抛洒物检测算法[J]. 计算机工程与应用, 2024, 60(5): 336-346., articleTitle=CIEFRNet: 面向高速公路的抛洒物检测算法, refAbstract=null), Reference(id=1251249365881405540, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2024, volume=60, issue=5, pageStart=336, pageEnd=346, url=null, language=null, rfNumber=[7], rfOrder=10, authorNames=Li Xu, Song Huansheng, Shi Qin, journalName=Computer Engineering and Applications, refType=null, unstructuredReference=Li Xu, Song Huansheng, Shi Qin, et al. CIEFRNet: abandoned objects detection algorithm for highway[J]. Computer Engineering and Applications, 2024, 60(5): 336-346., articleTitle=CIEFRNet: abandoned objects detection algorithm for highway, refAbstract=null), Reference(id=1251249365990457452, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=5148, pageEnd=5157, url=null, language=null, rfNumber=[8], rfOrder=11, authorNames=Nascimento M G, Fawcett R, Prisacariu V A, journalName=Proceedings of the IEEE/CVF International Conference on Computer Vision, refType=null, unstructuredReference=Nascimento M G, Fawcett R, Prisacariu V A. DSConv: efficient convolution operator[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 5148-5157., articleTitle=DSConv: efficient convolution operator, refAbstract=null), Reference(id=1251249366116286583, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=11863, pageEnd=11874, url=null, language=null, rfNumber=[9], rfOrder=12, authorNames=Yang L, Zhang R Y, Li L, journalName=International Conference on Machine Learning, refType=null, unstructuredReference=Yang L, Zhang R Y, Li L, et al. SimAM:a simple, parameter-free attention module for convolutional neural networks[C]// International Conference on Machine Learning. Online: PMLR, 2021: 11863-11874., articleTitle=SimAM:a simple, parameter-free attention module for convolutional neural networks, refAbstract=null), Reference(id=1251249366237921409, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[10], rfOrder=13, authorNames=Li H, Li J, Wei H, journalName=ArXiv Preprint, refType=null, unstructuredReference=Li H, Li J, Wei H, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[J]. ArXiv Preprint, 2022: 2206.02424., articleTitle=Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles, refAbstract=null), Reference(id=1251249366401499275, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=1099, pageEnd=1106, url=null, language=null, rfNumber=[11], rfOrder=14, authorNames=Pinggera P, Ramos S, Gehrig S, journalName=null, refType=null, unstructuredReference=Pinggera P, Ramos S, Gehrig S, et al. Lost and found:detecting small road hazards for self-driving vehicles[C]//2016 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS). Daejeon: IEEE, 2016: 1099-1106., articleTitle=Lost and found:detecting small road hazards for self-driving vehicles, refAbstract=null), Reference(id=1251249366510551187, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=8513, pageEnd=8520, url=null, language=null, rfNumber=[12], rfOrder=15, authorNames=Singh A, Kamireddypalli A, Gandhi V, journalName=null, refType=null, unstructuredReference=Singh A, Kamireddypalli A, Gandhi V, et al. Lidar guided small obstacle segmentation[C]//2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Las Vegas: IEEE, 2020: 8513-8520., articleTitle=Lidar guided small obstacle segmentation, refAbstract=null), Reference(id=1251249366636380325, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=3776, pageEnd=3794, url=null, language=null, rfNumber=[13], rfOrder=16, authorNames=Maag K, Chan R, Uhlemeyer S, journalName=Proceedings of the Asian Conference on Computer Vision, refType=null, unstructuredReference=Maag K, Chan R, Uhlemeyer S, et al. Two video data sets for tracking and retrieval of out of distribution objects[C]// Proceedings of the Asian Conference on Computer Vision. Macau: Springer, 2022: 3776-3794., articleTitle=Two video data sets for tracking and retrieval of out of distribution objects, refAbstract=null), Reference(id=1251249366762209458, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=406, pageEnd=423, url=null, language=null, rfNumber=[14], rfOrder=17, authorNames=Li K, Chen K, Wang H, journalName=European Conference on Computer Vision, refType=null, unstructuredReference=Li K, Chen K, Wang H, et al. Coda: a real-world road corner case dataset for object detection in autonomous driving[C]// European Conference on Computer Vision. Tel Aviv: Springer, 2022: 406-423., articleTitle=Coda: a real-world road corner case dataset for object detection in autonomous driving, refAbstract=null), Reference(id=1251249366984507590, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=2209, pageEnd=null, url=null, language=null, rfNumber=[15], rfOrder=18, authorNames=Li C, Zhou A, Yao A, journalName=ArXiv Preprint, refType=null, unstructuredReference=Li C, Zhou A, Yao A. Omni-dimensional dynamic convolution[J]. ArXiv Preprint, 2022: 2209.07947., articleTitle=Omni-dimensional dynamic convolution, refAbstract=null), Reference(id=1251249367118725330, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=1251, pageEnd=1258, url=null, language=null, rfNumber=[16], rfOrder=19, authorNames=Chollet F, journalName=Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, refType=null, unstructuredReference=Chollet F. Xception: deep learning with depthwise separable convolutions[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 1251-1258., articleTitle=Xception: deep learning with depthwise separable convolutions, refAbstract=null), Reference(id=1251249367261331677, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=1580, pageEnd=1589, url=null, language=null, rfNumber=[17], rfOrder=20, authorNames=Han K, Wang Y, Tian Q, journalName=Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, refType=null, unstructuredReference=Han K, Wang Y, Tian Q, et al. Ghostnet: more features from cheap operations[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. Seattle: IEEE, 2020: 1580-1589., articleTitle=Ghostnet: more features from cheap operations, refAbstract=null), Reference(id=1251249367412326635, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=3139, pageEnd=3148, url=null, language=null, rfNumber=[18], rfOrder=21, authorNames=Misra D, Nalamada T, Arasanipalai A, journalName=IEEE/CVF Winter Conference on Applications of Computer Vision, refType=null, unstructuredReference=Misra D, Nalamada T, Arasanipalai A, et al. Rotate to attend: convolutional triplet attention module[C]// IEEE/CVF Winter Conference on Applications of Computer Vision. Waikoloa: IEEE, 2021: 3139-3148., articleTitle=Rotate to attend: convolutional triplet attention module, refAbstract=null), Reference(id=1251249367609458940, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=2235, pageEnd=9, url=null, language=null, rfNumber=[19], rfOrder=22, authorNames=Zhang Q L, Yang Y B, journalName=ICASSP-IEEE International Conference on Acoustics, refType=null, unstructuredReference=Zhang Q L, Yang Y B. SA-Net: shuffle attention for deep convolutional neural networks[C]// ICASSP-IEEE International Conference on Acoustics. Toronto: IEEE, 2021: 2235-9., articleTitle=SA-Net: shuffle attention for deep convolutional neural networks, refAbstract=null), Reference(id=1251249369215877382, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2024, volume=236, issue=null, pageStart=121352, pageEnd=null, url=null, language=null, rfNumber=[20], rfOrder=23, authorNames=Lau K W, Po L M, Rehman Y A U, journalName=Expert Systems with Applications, refType=null, unstructuredReference=Lau K W, Po L M, Rehman Y A U. Large separable kernel attention: rethinking the large kernel attention design in CNN[J]. Expert Systems with Applications, 2024, 236: 121352., articleTitle=Large separable kernel attention: rethinking the large kernel attention design in CNN, refAbstract=null), Reference(id=1251249369371066638, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2023, volume=45, issue=2, pageStart=1489, pageEnd=1500, url=null, language=null, rfNumber=[21], rfOrder=24, authorNames=Li Y H, Yao T, Pan Y W, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=Li Y H, Yao T, Pan Y W, et al. Contextual transformer networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(2): 1489-1500., articleTitle=Contextual transformer networks for visual recognition, refAbstract=null), Reference(id=1251249369555616027, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=21, pageEnd=37, url=null, language=null, rfNumber=[22], rfOrder=25, authorNames=Liu W, Anguelov D, Erhan D, journalName=European Conference on Computer Vision, refType=null, unstructuredReference=Liu W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector[C]// European Conference on Computer Vision. Amsterdam: Springer, 2016: 21-37., articleTitle=SSD: single shot multibox detector, refAbstract=null), Reference(id=1251249369689833765, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[23], rfOrder=26, authorNames=Zhou X, Wang D, Krähenbühl P, journalName=ArXiv Preprint, refType=null, unstructuredReference=Zhou X, Wang D, Krähenbühl P. Objects as points[J]. ArXiv Preprint, 2019: 1904.07850., articleTitle=Objects as points, refAbstract=null), Reference(id=1251249369811468586, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[24], rfOrder=27, authorNames=Redmon J, Farhadi A, journalName=ArXiv Preprint, refType=null, unstructuredReference=Redmon J, Farhadi A. YOLOv3: an incremental improvement[J]. ArXiv Preprint, 2018: 1804.02767., articleTitle=YOLOv3: an incremental improvement, refAbstract=null), Reference(id=1251249369949880628, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=7464, pageEnd=7475, url=null, language=null, rfNumber=[25], rfOrder=28, 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. Vancouver: 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=1251249370067321144, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[26], rfOrder=29, authorNames=Ge Z, Liu S, Wang F, journalName=ArXiv Preprint, refType=null, unstructuredReference=Ge Z, Liu S, Wang F, et al. YOLOX: exceeding YOLO series in 2021[J]. ArXiv Preprint, 2021: 2107.08430., articleTitle=YOLOX: exceeding YOLO series in 2021, refAbstract=null), Reference(id=1251249370188955970, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2024, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[27], rfOrder=30, authorNames=Wang C Y, Yeh I H, Liao H Y M, journalName=ArXiv Preprint, refType=null, unstructuredReference=Wang C Y, Yeh I H, Liao H Y M. YOLOv9: learning what you want to learn using programmable gradient information[J]. ArXiv Preprint, 2024: 2402.13616., articleTitle=YOLOv9: learning what you want to learn using programmable gradient information, refAbstract=null), Reference(id=1251249370348339533, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2015, volume=111, issue=null, pageStart=98, pageEnd=136, url=null, language=null, rfNumber=[28], rfOrder=31, authorNames=Everingham M, Eslami S A, Van Gool L, journalName=International Journal of Computer Vision, refType=null, unstructuredReference=Everingham M, Eslami S A, Van Gool L, et al. The pascal visual object classes challenge: a retrospective[J]. International Journal of Computer Vision, 2015, 111: 98-136., articleTitle=The pascal visual object classes challenge: a retrospective, refAbstract=null), Reference(id=1251249370465780059, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, doi=null, pmid=null, pmcid=null, year=2013, volume=32, issue=11, pageStart=1231, pageEnd=1237, url=null, language=null, rfNumber=[29], rfOrder=32, authorNames=Geiger A, Lenz P, Stiller C, journalName=The International Journal of Robotics Research, refType=null, unstructuredReference=Geiger A, Lenz P, Stiller C, et al. Vision meets robotics: the KITTI dataset[J]. The International Journal of Robotics Research, 2013, 32(11): 1231-1237., articleTitle=Vision meets robotics: the KITTI dataset, refAbstract=null)], funds=[Fund(id=1251249362622432186, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, awardId=62303394, language=CN, fundingSource=国家自然科学基金(62303394), fundOrder=null, country=null), Fund(id=1251249362748261318, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, awardId=2022D01C694, language=CN, fundingSource=新疆维吾尔自治区自然科学基金(2022D01C694), fundOrder=null, country=null)], companyList=[AuthorCompany(id=1251249353264939303, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, xref=null, ext=[AuthorCompanyExt(id=1251249353269133608, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, companyId=1251249353264939303, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Electrical Engineering, Xinjiang University, Urumqi 830017, China), AuthorCompanyExt(id=1251249353277522217, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, companyId=1251249353264939303, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=新疆大学电气工程学院, 乌鲁木齐 830017)])], figs=[ArticleFig(id=1251249357153059295, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Fig.1, caption=Improvement of YOLOv8n network structure, figureFileSmall=MziJWHR5uuEwXUxa/enH9w==, figureFileBig=bVlF1Y6dWBcB9UWCVCmGPA==, tableContent=null), ArticleFig(id=1251249357337608680, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=图1, caption=改进的YOLOv8n网络结构, figureFileSmall=MziJWHR5uuEwXUxa/enH9w==, figureFileBig=bVlF1Y6dWBcB9UWCVCmGPA==, tableContent=null), ArticleFig(id=1251249357505380858, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Fig.2, caption=DSConv module structure, figureFileSmall=FcOB10GMAAPeWF3G4Hsw2w==, figureFileBig=/96cffKEgrVUg9y3afd94w==, tableContent=null), ArticleFig(id=1251249357627015686, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=图2, caption=DSConv模块结构, figureFileSmall=FcOB10GMAAPeWF3G4Hsw2w==, figureFileBig=/96cffKEgrVUg9y3afd94w==, tableContent=null), ArticleFig(id=1251249357727678997, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Fig.3, caption=SimAM module structure, figureFileSmall=Bpoc90CLTkyyzWILQdUY+w==, figureFileBig=IIPRnq1lmrcfudqaJeBg4w==, tableContent=null), ArticleFig(id=1251249357912228389, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=图3, caption=SimAM模块结构

X为特征图;CHW分别为特征图的通道数、高、宽

, figureFileSmall=Bpoc90CLTkyyzWILQdUY+w==, figureFileBig=IIPRnq1lmrcfudqaJeBg4w==, tableContent=null), ArticleFig(id=1251249358025474609, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Fig.4, caption=SPPF-SimAM module structure, figureFileSmall=+NCWMW1b68SB5voyJh8GwA==, figureFileBig=nEvffrjS+3WdsxYGjOpLyg==, tableContent=null), ArticleFig(id=1251249358109360699, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=图4, caption=SPPF-SimAM模块结构, figureFileSmall=+NCWMW1b68SB5voyJh8GwA==, figureFileBig=nEvffrjS+3WdsxYGjOpLyg==, tableContent=null), ArticleFig(id=1251249358226801224, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Fig.5, caption=GSConv module structure, figureFileSmall=dWQzYMTmNG7WLP/yW7sLDg==, figureFileBig=1cyJ0HNt9KCFAyyRVhpobQ==, tableContent=null), ArticleFig(id=1251249358323270230, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=图5, caption=GSConv模块结构

c1c2分别为输入和输出通道数

, figureFileSmall=dWQzYMTmNG7WLP/yW7sLDg==, figureFileBig=1cyJ0HNt9KCFAyyRVhpobQ==, tableContent=null), ArticleFig(id=1251249358432322150, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Fig.6, caption=GS bottleneck and VoV-GSCSP module structure, figureFileSmall=+SA/GEQ+meYXsUUfh9scVw==, figureFileBig=JiCxaJOilLyzYpSVVVcPhQ==, tableContent=null), ArticleFig(id=1251249358524596849, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=图6, caption=GS bottleneck和VoV-GSCSP模块结构, figureFileSmall=+SA/GEQ+meYXsUUfh9scVw==, figureFileBig=JiCxaJOilLyzYpSVVVcPhQ==, tableContent=null), ArticleFig(id=1251249358616871550, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Fig.7, caption=Comparison plot of the detection results, figureFileSmall=71Neu8k6I5B3ld83BVl+LQ==, figureFileBig=N5UNEo6NcaW1Y2+BEhGVkQ==, tableContent=null), ArticleFig(id=1251249360227484303, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=图7, caption=检测结果对比图, figureFileSmall=71Neu8k6I5B3ld83BVl+LQ==, figureFileBig=N5UNEo6NcaW1Y2+BEhGVkQ==, tableContent=null), ArticleFig(id=1251249360332341914, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Fig.8, caption=Mean average precision curve before and after improvement, figureFileSmall=vC2sVlIblmojdwe5JnGO9A==, figureFileBig=uoILOC9yLKQ//xHcDQ1HkA==, tableContent=null), ArticleFig(id=1251249360466559655, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=图8, caption=改进前后平均精度均值曲线图, figureFileSmall=vC2sVlIblmojdwe5JnGO9A==, figureFileBig=uoILOC9yLKQ//xHcDQ1HkA==, tableContent=null), ArticleFig(id=1251249360600777398, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Table 1, caption=

Statistics on dataset labels

, figureFileSmall=null, figureFileBig=null, tableContent=
障碍物类别 标签数量
训练集 验证集 测试集 合计
路锥 2 311 279 350 2 940
石墩 1 788 186 182 2 156
废弃轮胎 835 82 95 1 012
小动物 1 160 129 151 1 440
球类 489 57 54 600
箱式货物 2 173 277 292 2 742
水瓶 692 100 130 922
水马 1 157 112 105 1 374
施工标志牌 579 58 81 718
防撞桶 983 136 117 1 236
路桩 1 548 148 228 1 924
石块 1 483 169 160 1 812
塑料袋 565 48 63 676
三角警示牌 270 30 26 326
合计 16 033 1 811 2 034 19 878
), ArticleFig(id=1251249360701440707, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=表1, caption=

数据集标签统计

, figureFileSmall=null, figureFileBig=null, tableContent=
障碍物类别 标签数量
训练集 验证集 测试集 合计
路锥 2 311 279 350 2 940
石墩 1 788 186 182 2 156
废弃轮胎 835 82 95 1 012
小动物 1 160 129 151 1 440
球类 489 57 54 600
箱式货物 2 173 277 292 2 742
水瓶 692 100 130 922
水马 1 157 112 105 1 374
施工标志牌 579 58 81 718
防撞桶 983 136 117 1 236
路桩 1 548 148 228 1 924
石块 1 483 169 160 1 812
塑料袋 565 48 63 676
三角警示牌 270 30 26 326
合计 16 033 1 811 2 034 19 878
), ArticleFig(id=1251249360831464151, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Table 2, caption=

Comparison result of introducing different convolutions

, figureFileSmall=null, figureFileBig=null, tableContent=
卷积 mAP/% Parameters/106 GFLOPs FPS/(帧·s-1)
Conv 84.6 3.0 8.1 60
ODConv 83.7 3.0 7.1 60
DWConv 84.5 2.6 7.2 67
GhostConv 84.3 2.8 7.7 52
DSConv 84.8 3.0 7.1 73
), ArticleFig(id=1251249360948904680, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=表2, caption=

引入不同卷积的对比结果

, figureFileSmall=null, figureFileBig=null, tableContent=
卷积 mAP/% Parameters/106 GFLOPs FPS/(帧·s-1)
Conv 84.6 3.0 8.1 60
ODConv 83.7 3.0 7.1 60
DWConv 84.5 2.6 7.2 67
GhostConv 84.3 2.8 7.7 52
DSConv 84.8 3.0 7.1 73
), ArticleFig(id=1251249361062150902, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Table 3, caption=

Comparison result of SPPF combined with different attentional mechanisms

, figureFileSmall=null, figureFileBig=null, tableContent=
算法 mAP/% Parameters/
106
GFLOPs FPS/
(帧·s-1)
SPPF 84.6 3.0 8.1 60
SPPF-Triplet
Attention
85.1 3.0 8.1 64
SPPF-Shuffle
Attention
84.6 3.0 8.1 67
SPPF-LSKA 84.6 3.3 8.3 67
SPPF-CoT
Attention
84.9 5.3 9.9 55
SPPF-SimAM 85.3 3.0 8.1 66
), ArticleFig(id=1251249361183785739, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=表3, caption=

SPPF与不同注意力机制结合的对比结果

, figureFileSmall=null, figureFileBig=null, tableContent=
算法 mAP/% Parameters/
106
GFLOPs FPS/
(帧·s-1)
SPPF 84.6 3.0 8.1 60
SPPF-Triplet
Attention
85.1 3.0 8.1 64
SPPF-Shuffle
Attention
84.6 3.0 8.1 67
SPPF-LSKA 84.6 3.3 8.3 67
SPPF-CoT
Attention
84.9 5.3 9.9 55
SPPF-SimAM 85.3 3.0 8.1 66
), ArticleFig(id=1251249361292837663, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Table 4, caption=

Comparison result of mAP of different obstacle classes

, figureFileSmall=null, figureFileBig=null, tableContent=
障碍物类别 准确率/% 召回率/% mAP/%
YOLOv8n 本文算法 YOLOv8n 本文算法 YOLOv8n 本文算法
路锥 88.4 92.5 74.0 77.4 81.0 87.0
石墩 82.5 92.8 70.9 78.2 79.1 88.1
废弃轮胎 88.2 95.0 88.4 99.2 94.8 98.2
小动物 95.7 79.8 90.7 84.6 94.7 84.3
球类 97.1 98.1 72.2 92.6 91.8 97.7
箱式货物 97.8 98.2 92.4 95.4 96.9 98.6
水瓶 79.9 77.8 63.1 75.4 74.0 83.9
水马 81.1 81.4 75.2 74.3 80.5 85.9
施工标志牌 90.6 88.7 77.8 82.7 82.7 88.1
防撞桶 86.8 91.0 72.8 77.9 82.9 89.1
路桩 83.1 89.2 55.3 69.3 71.8 84.9
石块 88.1 93.0 70.0 94.4 79.2 97.7
塑料袋 88.8 91.7 77.8 81.0 85.3 90.4
三角警示牌 95.6 79.8 84.0 84.6 89.0 84.3
平均值 88.8 90.4 76.0 84.0 84.6 90.8
), ArticleFig(id=1251249361439638321, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=表4, caption=

不同障碍物类别平均精度均值对比结果

, figureFileSmall=null, figureFileBig=null, tableContent=
障碍物类别 准确率/% 召回率/% mAP/%
YOLOv8n 本文算法 YOLOv8n 本文算法 YOLOv8n 本文算法
路锥 88.4 92.5 74.0 77.4 81.0 87.0
石墩 82.5 92.8 70.9 78.2 79.1 88.1
废弃轮胎 88.2 95.0 88.4 99.2 94.8 98.2
小动物 95.7 79.8 90.7 84.6 94.7 84.3
球类 97.1 98.1 72.2 92.6 91.8 97.7
箱式货物 97.8 98.2 92.4 95.4 96.9 98.6
水瓶 79.9 77.8 63.1 75.4 74.0 83.9
水马 81.1 81.4 75.2 74.3 80.5 85.9
施工标志牌 90.6 88.7 77.8 82.7 82.7 88.1
防撞桶 86.8 91.0 72.8 77.9 82.9 89.1
路桩 83.1 89.2 55.3 69.3 71.8 84.9
石块 88.1 93.0 70.0 94.4 79.2 97.7
塑料袋 88.8 91.7 77.8 81.0 85.3 90.4
三角警示牌 95.6 79.8 84.0 84.6 89.0 84.3
平均值 88.8 90.4 76.0 84.0 84.6 90.8
), ArticleFig(id=1251249361552884546, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Table 5, caption=

Results of ablation experiments

, figureFileSmall=null, figureFileBig=null, tableContent=
编号 DSConv 小目标
检测层
SPPF-SimAM GSConv+VoV-
GSCSP
准确率/% 召回率/% mAP/% Parameters/106 GFLOPs FPS/(帧·s-1)
A × × × × 88.8 76.0 84.6 3.0 8.1 60
B × × × 89.3 75.0 84.8 3.0 7.1 73
C × × × 90.2 84.7 90.1 3.0 12.5 68
D × × × 91.6 75.1 85.3 3.0 8.1 66
E × × × 88.9 74.3 84.7 2.8 7.5 62
F × × 91.7 82.8 90.7 3.0 11.5 68
G × 92.7 84.2 90.9 3.0 11.5 51
H 90.4 84.0 90.8 2.8 10.5 53
), ArticleFig(id=1251249361666130762, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=表5, caption=

消融实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
编号 DSConv 小目标
检测层
SPPF-SimAM GSConv+VoV-
GSCSP
准确率/% 召回率/% mAP/% Parameters/106 GFLOPs FPS/(帧·s-1)
A × × × × 88.8 76.0 84.6 3.0 8.1 60
B × × × 89.3 75.0 84.8 3.0 7.1 73
C × × × 90.2 84.7 90.1 3.0 12.5 68
D × × × 91.6 75.1 85.3 3.0 8.1 66
E × × × 88.9 74.3 84.7 2.8 7.5 62
F × × 91.7 82.8 90.7 3.0 11.5 68
G × 92.7 84.2 90.9 3.0 11.5 51
H 90.4 84.0 90.8 2.8 10.5 53
), ArticleFig(id=1251249361745822552, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Table 6, caption=

Results of the mainstream model comparison experiments

, figureFileSmall=null, figureFileBig=null, tableContent=
算法 mAP/% Parameters/
106
GFLOPs FPS/
(帧·s-1)
SSD 59.4 25.3 62.1 15
CenterNet 55.8 32.7 109.7 33
YOLOv3-tiny 66.6 8.7 12.9 63
YOLOv5s 74.2 7.1 16.0 56
YOLOv7 80.2 36.5 103.4 46
YOLOv7-tiny 70.6 6.0 13.1 64
YOLOX-s 68.9 8.9 26.8 69
YOLOX-tiny 63.3 5.0 15.3 70
YOLOv8s 89.5 11.1 28.5 55
YOLOv9s 89.3 9.6 38.8 19
本文算法 90.8 2.8 10.5 53
), ArticleFig(id=1251249361880040295, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=表6, caption=

主流模型对比实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
算法 mAP/% Parameters/
106
GFLOPs FPS/
(帧·s-1)
SSD 59.4 25.3 62.1 15
CenterNet 55.8 32.7 109.7 33
YOLOv3-tiny 66.6 8.7 12.9 63
YOLOv5s 74.2 7.1 16.0 56
YOLOv7 80.2 36.5 103.4 46
YOLOv7-tiny 70.6 6.0 13.1 64
YOLOX-s 68.9 8.9 26.8 69
YOLOX-tiny 63.3 5.0 15.3 70
YOLOv8s 89.5 11.1 28.5 55
YOLOv9s 89.3 9.6 38.8 19
本文算法 90.8 2.8 10.5 53
), ArticleFig(id=1251249361951343474, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Table 7, caption=

Comparison result of the model before and after improvement on PASCAL VOC 2012

, figureFileSmall=null, figureFileBig=null, tableContent=
算法 mAP/% Parameters/
106
GFLOPs FPS/
(帧·s-1)
YOLOv8n 56.4 3.0 8.1 88
本文算法 57.0 2.8 10.5 61
), ArticleFig(id=1251249362052006786, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=表7, caption=

改进前后模型在PASCAL VOC 2012的对比实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
算法 mAP/% Parameters/
106
GFLOPs FPS/
(帧·s-1)
YOLOv8n 56.4 3.0 8.1 88
本文算法 57.0 2.8 10.5 61
), ArticleFig(id=1251249362152670091, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=EN, label=Table 8, caption=

Comparison result of the model before and after improvement on KITTI

, figureFileSmall=null, figureFileBig=null, tableContent=
算法 mAP/% Parameters/
106
GFLOPs FPS/
(帧·s-1)
YOLOv8n 76.9 3.0 8.1 75
本文算法 78.2 2.8 10.5 53
), ArticleFig(id=1251249362240750488, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1149781961000051646, language=CN, label=表8, caption=

改进前后模型在KITTI的对比实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
算法 mAP/% Parameters/
106
GFLOPs FPS/
(帧·s-1)
YOLOv8n 76.9 3.0 8.1 75
本文算法 78.2 2.8 10.5 53
)], 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.1671-1815.2402441, detailUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/10.12404/j.issn.1671-1815.2402441, pdfUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/PDF/10.12404/j.issn.1671-1815.2402441, pdfUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/PDF/10.12404/j.issn.1671-1815.2402441, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
智能驾驶场景下的中小型障碍物检测方法
收藏切换
PDF下载
龙小羽 , 南新元 *
科学技术与工程 | 论文·自动化技术、计算机技术 2025,25(9): 3778-3787
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(9): 3778-3787
智能驾驶场景下的中小型障碍物检测方法
全屏
龙小羽 , 南新元*
作者信息
  • 新疆大学电气工程学院, 乌鲁木齐 830017
  • 龙小羽(2000—),女,汉族,新疆和静人,硕士研究生。研究方向:深度学习、目标检测。E-mail:

通讯作者:

* 南新元(1967—),男,汉族,新疆乌鲁木齐人,硕士,教授。研究方向:智能控制系统开发、机器视觉。E-mail:
Small and Medium-sized Obstacle Detection Methods in Intelligent Driving Scenarios
Xiao-yu LONG , Xin-yuan NAN*
Affiliations
  • School of Electrical Engineering, Xinjiang University, Urumqi 830017, China
出版时间: 2025-03-28 doi: 10.12404/j.issn.1671-1815.2402441
文章导航
收藏切换

针对智能驾驶场景下路面中小型障碍物易发生漏检、小目标障碍物难检测、模型参数量大等问题,提出了改进YOLOv8n的障碍物目标检测算法。在主干网络中融入分布移位卷积(distribution shifting convolution,DSConv),将浮点运算替换为整数运算,减少了冗余计算量,通过量化和分布移位的方式模仿原始卷积层,维持了准确率;通过添加小目标检测层,更好地捕捉小目标的特征信息,适配小目标的尺度特征;结合SimAM无参数注意力机制,引入SPPF-SimAM模块,提高特征表示的质量与多样性,在不增加参数量的情况下实现了检测精度的提升;通过组合鬼影混洗卷积(ghost-shuffle convolution,GSConv)和VoV-GSCSP模块的方式轻量化颈部特征融合网络,降低了模型的参数量和计算量。实验结果表明,改进后模型的准确率、召回率、平均精度均值相较于原始模型分别提升了1.6%、8.0%、6.2%,参数量降低了6.7%,所提算法有效提升了智能驾驶场景下中小型障碍物的检测精度,并且在检测性能与模型轻量化之间达到较好的平衡。

障碍物检测  /  YOLOv8n  /  智能驾驶  /  小目标检测  /  注意力机制

Aiming at the problems such as small and medium-sized obstacles on the road are prone to miss detection, small target obstacles are difficult to detect, and the number of model parameters is large in smart driving scenarios, the obstacle target detection algorithm with improved YOLOv8n was proposed. Distribution shifting convolution (DSConv) was used in the backbone network to replace floating point operation with integer operation, reducing the amount of redundant computation, and maintaining the accuracy by imitating the original convolution layer by quantization and distribution shifting. By adding small target detection layer, the feature information of small target can be captured better and the scale characteristics of small target can be adapted. Combined with SimAM parameterless attention mechanism, SPPF-SimAM module was introduced to improve the quality and diversity of feature representation, and the detection accuracy was improved without increasing the number of parameters. By combining ghost-shuffle convolution (GSConv) and VoV-GSCSP modules, the neck feature fusion network was lightweight, reducing the number of parameters and calculation of the model. The experimental results show that the accuracy, recall, and mean average precision of the improved model are improved by 1.6%, 8.0%, and 6.2%, respectively. The number of parameters is reduced by 6.7% compared with the original model, and the proposed algorithm effectively improves the detection accuracy of small and medium-sized obstacles in smart driving scenarios, and achieves a better balance between the detection performance and the model lightweighting.

obstacle detection  /  YOLOv8n  /  intelligent driving  /  small target detection  /  attention mechanism
龙小羽, 南新元. 智能驾驶场景下的中小型障碍物检测方法. 科学技术与工程, 2025 , 25 (9) : 3778 -3787 . DOI: 10.12404/j.issn.1671-1815.2402441
Xiao-yu LONG, Xin-yuan NAN. Small and Medium-sized Obstacle Detection Methods in Intelligent Driving Scenarios[J]. Science Technology and Engineering, 2025 , 25 (9) : 3778 -3787 . DOI: 10.12404/j.issn.1671-1815.2402441
近年来,随着数字技术的不断发展与成熟,智能驾驶技术在汽车工业中备受关注,人们对智能驾驶的接受程度也在不断提升,具有智能驾驶功能的汽车在市场上的需求在快速增长。其中路面障碍物的检测是实现更加安全可靠自动驾驶的关键步骤之一。特别是对于中小型障碍物的准确识别与检测,对于提升自动驾驶汽车的安全性和稳定性具有至关重要的意义。
目前智能驾驶系统中的目标检测任务主要是检测出道路上的车辆、行人、交通标志等信息来保证自动驾驶车辆行驶的安全性[1]。随着科学技术的不断发展,乘客对自动驾驶车辆的可靠性和舒适性要求大大提升。在车辆行驶过程中,可能会出现一些异常障碍物,例如路锥、废弃轮胎、石墩、前方掉落的货物等,这些障碍物日常情况下较少见且体积相对较小,车辆需要及时发现并进行避让,这种“罕见又突发”的情况,在智能驾驶行业被称为“边角情况”(corner case,CC)[2],在处理这些情况时,智能驾驶系统必须具备高度的感知能力和快速的响应性,以确保车辆和乘客的安全。
随着深度学习的不断发展,越来越多的学者开始关注智能驾驶领域的异常障碍物检测问题,引发了大量学者的深入研究。文献[3]提出了一种用于小型障碍物检测的MergeNet深度网络架构,通过多阶段的训练过程,每个阶段学习互补的特征,融合这些特征来预测场景的分割图,能够在公共数据集上对障碍物进行高保真度分割,但该方法的场景泛化性不足,且运行速度较慢。文献[4]采用深度学习方法,以检测到的道路消失点为导向,设计了一种基于全局特征的道路消失点检测模型,以确定图像中障碍物的位置区域,并在该区域上进行障碍物检测。文献[5]提出了一种端到端的实时障碍物检测,创造了RODSNet联合学习来自立体RGB图像的语义分割和视差图,利用两种高效的单任务网络架构和多任务学习方案中的模块来识别道路上异常的小型障碍物。文献[6]在SSD模型基础上,提出一种感受野增强和特征融合的小目标检测模型,兼顾检测效率的同时,有效提高了对道路上小目标的检测精度。文献[7]针对高速公路存在抛洒物常出现漏检和误检问题,以YOLOv5s为主体,提出了一种基于特征提纯和上下文信息增强的CIEFRNet抛洒物检测算法,提升了检测效果。
现有的路面异常障碍物检测已经取得了一定的进展,但是其准确性、响应速度、模型轻量化程度及泛化性能均有待提高,且部分研究依靠激光雷达等设备,成本较为高昂。为了能够更容易部署到移动设备以贴合实际,需要在更低成本的计算资源条件下实现更高的检测精度和响应性。针对上述存在的问题,现以YOLOv8n为基准模型,在此基础上构建一种中小型障碍物检测算法。融合DSConv卷积[8]优化主干网络结构,将浮点运算替换为整数运算,减少模型计算量的同时通过量化和分布移位维持检测性能。 智能驾驶场景中包含大量的小目标障碍物,添加特征图为160×160的小目标检测层提升小目标的检测精度。 为了提高目标特征表示的质量,对SPPF模块进行优化,将无参数注意力机制SimAM[9]与SPPF模块相结合构成SPPF-SimAM模块,通过特征加权、空间关系建模和全局信息整合等方式,更高效地提取到目标的关键信息。 由于添加的小目标检测层造成模型复杂程度增加,结合Slim-neck模块[10]将颈部网络的卷积替换为轻量级卷积GSConv,部分C2f模块替换为VoV-GSCSP模块,降低模型参数量和计算量。 以期在检测性能与模型轻量化之间达到较好的平衡,同时为智能驾驶过程中的安全问题提供新的解决思路。
YOLOv8是一种单阶段目标检测算法,具有广泛的应用领域,还可用于目标跟踪、姿态估计、图像分割等任务。YOLOv8的架构主要包含主干(Backbone)特征提取网络、颈部(Neck)特征融合网络和检测头(Head)这3个部分。Backbone负责网络的特征提取功能,采用了一系列卷积和C2f模块。Neck部分采用PANet结构,利用上采样和通道融合,增强特征的表示能力,最终将PANet的3个输出分支输入Head中。Head部分负责产生最后的检测结果,采用目前较为主流的解耦头结构,包括一个检测头和一个分类头,有助于加快模型的收敛速度。
YOLOv8包括5个模型:YOLOv8-n、YOLOv8-s、YOLOv8-m、YOLOv8-l、YOLOv8-x,模型的参数量和计算量随着模型深度和宽度的增大而增加。YOLOv8n是目前模型最小、速度最快且易于在低成本、低算力设备上部署的版本,适用于智能驾驶场景中的障碍物检测任务。为了满足智能驾驶场景中对障碍物检测实时性的需求,并在保持检测准确率的同时降低成本和算力消耗,故选用YOLOv8n作为实验基础模型。
基于YOLOv8n模型进行改进,着眼于实现低延迟、高精度、轻量化的网络模型,旨在提高模型对小目标的感知能力和检测准确率,同时减小模型的体积和计算量,以适应实际应用的需求。在YOLOv8n的基础上,通过在Backbone中引入DSConv替换原有的传统卷积,减少冗余计算量的同时提升检测精度;增加小目标检测层提升了对小目标障碍物的检测精度;引入一种SPPF-SimAM模块,通过无参数注意力机制更高效地提取关键特征;采用GSConv和VoV-GSCSP模块轻量化Neck部分,总体实现网络模型的轻量化和精确度的提升。改进后的YOLOv8n网络结构如图1所示。
为了降低模型计算的复杂性,在YOLOv8n中引入DSConv。DSConv是一种深度可分离卷积的变体,总体目标是通过量化和分布移位来模拟卷积层的行为,用成本更低的整数运算代替单精度运算,同时保证卷积内核权重及输出的概率分布。DSConv模块结构如图2所示。
DSConv可以分为可变量化核(variable quantized kernel,VQK)和分布移位(distribution shifts)两个组件。VQK是DSConv的量化组件,可实现更快、内存效率更高的乘法运算;而分布移位用于改变VQK分布,以尝试模仿原始卷积的分布,“移位”是指缩放和偏置操作,是通过使用两个张量在两个域中移动来实现的。第一个张量是核分布移位器(kernel distribution shifter,KDS),用于移位每个(1,BLK,1,1)切片中的分布,BLK是块大小超参数,它的大小为2[cho,CEIL$\left(\frac{\mathrm{c}{\mathrm{h}}_{i}}{\mathrm{B}\mathrm{L}\mathrm{K}}\right)$,k,k];cho为下一层的通道数;CEIL(x)为向上舍入运算;k为内核的宽和高。第二个张量是通道分布移位器(channel distribution shifter,CDS),用于移动每个通道中的分布,即改变每个(1,chi,k,k)切片中的分布,chi为当前层通道数,它的大小为2(cho)的单精度张量。
在得到VQK和KDS两个组件后,明确量化程序,通常量化函数需要将量化的位数作为输入,存储时使用有符号的整数形式表达,使用wq表示张量中各个参数的值,对于输入多个位数b时,量化函数如式(1)所示。

wq∈Z|-2b-1wq≤2b-1-1

首先对每个卷积层的权重进行缩放,使得原始权重w的最大绝对值与上述量化函数的约束最大值相匹配,将所有权重量化为最接近的整数,并将新的整数值权重wq存储起来,量化程序仅用于VQK组件的缩放。
分布移位的目的是通过KDS和CDS改变VQK的值,以便输出与原始权重张量的值匹配,首先需要对它们进行初始化。通过最小化KL散度的方法,由KDS移位后得到的VQK应该具有与原始权重相似的分布,用ξ表示缩放,采用移位后的VQK和原始分布的softmax值来计算,如式(2)所示。
$\left\{\begin{array}{l}{T}_{j}=\frac{{\mathrm{e}}^{{w}_{j}}}{\sum _{i}{\mathrm{e}}^{{w}_{i}}},\\ {I}_{j}=\frac{{\mathrm{e}}^{\stackrel{\wedge }{\xi }{w}_{{q}_{j}}}}{\sum _{i}{\mathrm{e}}^{\stackrel{\wedge }{\xi }{w}_{{q}_{i}}}}\end{array}\right.$
式(2)中:Tj为第j块卷积核的VQK的初始值;Ij为第j块卷积核的softmax值;i为卷积层数,j为当前层的第j块卷积核;$\stackrel{\wedge }{\xi }$为偏置,初始值为0。通过梯度下降最小化损失得到ξ,ξ为张量中每个参数的值,表达式为
ξ=$\underset{\stackrel{\wedge }{\xi }}{\mathrm{m}\mathrm{i}\mathrm{n}}\sum _{j}$Tjln$\left(\frac{{T}_{j}}{{I}_{j}}\right)$,∀(1,BLK,1,1)slice
DSConv相较于传统卷积的优势在于计算速度快,占用内存更少。因此,在主干特征提取网络中引入DSConv,能够在维持精度的同时有效降低模型的参数量和计算量。
数据集中存在着大量的小目标障碍物,这些小目标障碍物与常见的车辆、行人相比,所占尺寸相对较小,形状和外观也更加多样化。小目标通常具有更小的像素区域和更低的分辨率,细节与边界信息不明显,且目标的特征更加微弱、模糊。输入图像尺寸对小目标障碍物的识别也产生了干扰,原YOLOv8n的检测层会输出20×20、40×40、80×80三种不同尺寸的特征图,对原始图像进行缩放会使小目标的识别更加困难。
基于此,对YOLOv8n网络进行优化,添加一个专门处理小目标的检测层,引入更多的上下文信息和特征表示,从而提高模型对小目标检测的鲁棒性以及识别的准确性。当输入为尺寸640×640的特征图时,在原来的网络中增加一个160×160特征图的检测层,从主干网络的第六层提取特征信息,与颈部网络上采样后的信息进行特征融合,能够对驾驶视角中距离较远的障碍物进行更加准确的检测,一定程度上降低了各种尺度上对小目标障碍物的漏检与误检,但模型的复杂程度也因此而增大。
面向驾驶场景的障碍物检测通常具有复杂的环境背景,这需要模型提升对目标障碍物关键特征的提取能力。如图3所示,SimAM注意力机制作为一种具有全三维权重的无参数注意力机制, 可以通过测量一个目标神经元和其他神经元之间的线性可分离性来估计各个神经元的重要性实现其功能,最小能量函数${e}_{t}^{\mathrm{*}}$
${e}_{t}^{\mathrm{*}}$=$\frac{4({\stackrel{\wedge }{\sigma }}^{2}+\lambda )}{{(t-\stackrel{\wedge }{\mu })}^{2}+2{\stackrel{\wedge }{\sigma }}^{2}+2\lambda }$
$\stackrel{\wedge }{\mu }$=$\frac{1}{M}\stackrel{M}{\sum _{i=1}}$xi
${\stackrel{\wedge }{\sigma }}^{2}$=$\frac{1}{M}\stackrel{M}{\sum _{i=1}}({x}_{i}{-\stackrel{\wedge }{\mu })}^{2}$
式中:λ为超参数,设置为0.000 1;t为输入特征的值;$\stackrel{\wedge }{\mu }$${\stackrel{\wedge }{\sigma }}^{2}$分别为通道中单个神经元的平均值和方差;x为其余神经元;i为空间上的维度;M为通道中神经元数量。由式(4)可知,${e}_{t}^{\mathrm{*}}$越低神经元的区别越大,即重要性越高。整个SimAM模块通过Sigmoid优化,可以表示为
$\stackrel{\wedge }{X}$=Sigmoid$\left(\frac{1}{E}\right)$X
式(7)中:E为神经元的最小能量函数集合;X为输入特征;☉表示矩阵元素相乘。Sigmoid可以在不影响神经元之间相对重要性的前提下抑制E的值不能过大。
SPPF通过3次连续的池化操作,结合不同尺度的输出进行融合,将SimAM引入SPPF模块中,能够更好的整合不同池化层的特征信息,提高关键特征信息的表征能力,进而提升特征融合的效果,保证特征提取的信息更加丰富,改进的SPPF-SimAM结构如图4所示。
在模型中添加小目标检测层时增加了额外的网络层数,这导致模型的计算量增加,从而降低整体的性能和速度。为了在维持精度的同时减少模型的计算资源,利用Slim-Neck方法中的轻量级卷积GSConv和VoV-GSCSP模块实现颈部特征融合网络的轻量化,可以有效缓解高昂的计算成本。
GSConv是一种新型的轻量化卷积模块,由标准卷积(standard convolution,SC)、深度可分离卷积(depth-wise separable convolution,DSC)和Shuffle组成,使用Shuffle将SC生成的信息渗透到DSC生成的信息的每个部分中。Shuffle是一种统一的混合策略,该方法通过在不同通道上统一交换局部特征信息,可以将来自SC的信息完全混合到DSC的输出中,而无需任何附加功能,使得卷积计算的输出尽可能接近SC的输出,降低了计算成本,计算量约为SC的50%。其模块结构如图5所示。
GSConv更适合于在颈部网络使用,因为在主干网络中使用会导致模型的网络层数更深,会加剧对数据流的阻力,增加推理时间,特征图传送到颈部网络时,其通道尺寸已达到最大,宽高维度最小,不再需要变换。而在颈部网络使用时冗余信息更少,不需要压缩。
以GSConv模块为基础引入GS bottleneck结构,使用一次性聚合方法设计跨阶段部分网络模块VoV-GSCSP,能够有效降低网络结构的复杂性,模型结构如图6所示。
通过组合GSConv和VoV-GSCSP的方式能够轻量化颈部特征融合网络,有效降低模型的参数量和计算量,从而使模型在不牺牲检测精度的条件下,加快模型的推理速度。
智能驾驶过程中出现的中小型障碍物通常存在日常情况下“罕见又突发”的特点,在单一数据集上训练容易导致模型出现过拟合等问题,为了提高模型的泛化能力,应对不同的驾驶场景,需要制作场景丰富的数据集。
因此,结合公开数据集、网络资源、实地采集等方式建立障碍物联合数据集。首先,从Lost And Found[11]、Small Obstacle[12]、Street Obstacle Sequences[13]、CODA[14]这4个公开数据集中筛选出2 564张图像,Lost And Found包含13个驾驶场景,是一个专注于小型道路危险检测的数据集;Small Obstacle数据集的研究对象主要为施工活动、货物掉落等场景出现的小障碍物;Street Obstacle Sequences包含20个真实场景的视频序列,涵盖13个不同的异常障碍物类型;CODA数据集具有丰富的驾驶场景,各个场景至少包含一个对智能驾驶车辆或周围环境有危险性或较为新奇的物体。为了在不同场景下获得更丰富可靠的数据样本,通过网络引擎搜索、真实道路场景自行采集的方式获得1 437张图像。为了保证自行采集数据的可用性,每个图像中需要包含至少一个障碍物类别,且在多个时段、多种角度和距离进行采集。通过上述方式累计获得共4 001张图像,联合多种场景的数据集包含了多样的光照情况及天气条件,不仅仅局限于单一的数据场景,有助于提升模型的泛化能力。
由于驾驶视角的特殊性,仅采取镜像翻转扩充的方式数据增强至8 002张构成障碍物联合数据集。按照8∶1∶1的比例划分为训练集、验证集和测试集供本实验使用。
选用14类边角情况下的异常中小型障碍物构建数据集,包括:路锥、石墩、废弃轮胎、小动物、球、箱式货物、水瓶、水马、施工标志牌、防撞桶、路桩、石块、塑料袋、三角警示牌。对联合数据集进行标签及图片格式的统一,使用LabelImg进行标注,以障碍物名称作为标签类别,共计19 878个障碍物标签,各集合标签类别及其数量如表1所示。
实验均在Windows11操作系统中完成,选择Pytorch2.0.1深度学习框架,CUDA版本为11.7,编程语言为Python3.8。CPU型号为Inter(R) Core(TM) i7-12700H,运行内存16 GB,GPU采用NVIDIA GeForce RTX 4060 Laptop,8 G显存。
为消除参数变化的影响,在模型训练过程中均设置相同的实验参数,初始学习率为0.01,Batchsize设置为8,Workers设置为4,Epoch设置为100。
采用准确率(precision)、召回率(recall)、平均精度均值(mean average precision,mAP)、参数量(parameters)、浮点运算量(giga floating-point operations per second,GFLOPs)、每秒帧数(frames per second,FPS)指标对模型性能进行评价。
准确率、召回率、平均精度均值用于衡量模型的检测准确程度,参数量用于衡量模型的复杂程度和规模大小,浮点运算量用于评估模型的计算资源消耗,模型的检测速度由每秒检测帧数来表示。
智能驾驶场景下的障碍物检测需要满足高精度、实时性、低计算量的要求,这就需要在满足模型轻量化的同时维持检测精度。为了验证在主干网络引入DSConv卷积的有效性,与常见的卷积模块ODConv[15]、DWConv[16]、GhostConv[17]进行对比实验,实验结果如表2所示。
表2可以看出,4种卷积模块都能够降低基准模型的GFLOPs,一定程度上能够减少模型的计算资源。在检测速度方面,仅DWConv、DSConv有所提升。但是除了DSConv外,其他3种卷积方式的mAP都呈现小幅度的降低,未达到轻量化的同时维持检测精度的需求,其中ODConv降低了0.9%,DWConv降低了0.1%,GhostConv降低了0.3%,而DSConv提升了0.2%。
由此可知,DSConv模块更具有优越性,能够取代原模型中传统的卷积模块,在降低模型计算量的同时提升了检测精度,实现了更高的检测速度。
为提升模型对障碍物关键特征的提取能力,优化SPPF结构能够更有效地进行多尺度融合,在SPPF模块的相同位置引入不同注意力机制,包括Triplet Attention[18]、Shuffle Attention[19]、LSKA[20]、CoT Attention[21]进行对比试验,结果如表3所示。可以看出,在LSKA、CoT增加了一定的参数量和计算量,但检测精度的提升幅度较低,在同等参数量与计算量的情况下,采用的SimAM模块在mAP指标表现较突出,在不增加计算负担的前提下有效提升了检测性能。
为验证所提出的中小型障碍物检测算法相较于原模型而言对不同类别障碍物识别精度的优越性,将模型改进前后的障碍物检测结果进行比较,对比结果如表4所示。
原YOLOv8n模型对石墩、水瓶、路桩、石块的mAP值较低,均位于80%以下,分析其原因可知这四类障碍物的共同特点是数据集中包含大量的小目标,并且路桩会出现极端长宽比的情况,同时石墩与石块的形态颜色相似,易发生混淆。本文模型对石墩、水瓶、路桩、石块四类障碍物的mAP分别提升了9.0%、9.9%、13.1%、18.5%,说明本文所提模型对小目标的检测能力有所提高,且对其余障碍物类别的mAP均在80%以上,改进前后模型的检测效果如图7所示,可以看出改进后的检测算法对于小目标障碍物具备更高的特征提取能力,对于较远的目标也可以检测到,与原始模型相比,误检漏检的情况有明显改善。
为了更直观地体现本文模型相对于原模型的优化效果,将原模型与本文模型训练过程中的mAP值变化幅度可视化,如图8所示,可以看出从80轮开始模型mAP近趋于平稳,且改进后模型的平均精度均值收敛速度更快,精度更高。
为了验证各个模块对YOLOv8n模型改进的有效性,以YOLOv8n为基准模型,通过不同的模块组合方式设计了编号A~H共8组消融实验,实验结果如表5所示。
表5中结果可知,提出的4种改进方式均对基准模型产生了优化效果。如实验B所示,将主干网络中的标准卷积替换为DSConv卷积后,在维持检测精度与速度的同时降低了模型的计算量,且FPS每秒提升13帧,说明DSConv通过量化和分布移位维持了和原始卷积相似的输出;实验C在原模型中加入小目标检测层后,检测精度提升了5.5%,说明小目标检测层能够关注到数据集中更多的特征信息,检测精度提升的幅度较大,但由于网络模型的层数的扩增,增加了模型的计算量;实验D引入SPPF-SimAM模块,通过SimAM注意力机制在不增加参数量的前提下提升了检测精度;实验E通过轻量化颈部网络,将颈部网络中的全部卷积与部分C2f模块替换为GSConv与VoV-GSCSP模块,在mAP、参数量、GFLOPs、FPS四个评价指标上均有优化效果,其参数量和GFLOPs均有所下降,在模型轻量化的同时提升了0.1%的mAP,表明轻量化颈部网络能够在模型性能与轻量化之间达到较好的平衡。
由实验F可知,在主干网络中替换DSConv的同时加入小目标检测层,模型的mAP和FPS得到大幅优化,相比原模型来看,mAP增大了6.1%,FPS每秒提升8帧,同时由于DSConv计算量更小,减少了仅加入小目标检测层时的计算量,维持了检测性能;实验G在F的基础上引入SPPF-SimAM模块,充分发挥注意力机制对关键信息提取的作用,提升了模型的检测精度。实验H为本文模型,在G的基础上引入轻量化模块,mAP相较于原模型提升6.2%,参数量降低了6.7%,但是GFLOPs稍有提升。消融实验结果表明,改进后的模型能够更有效地检测小目标障碍物,改善了误检和漏检情况,并且基本满足了轻量化模型和实时性要求。
为了进一步评估改进的YOLOv8n在障碍物检测研究中的性能,考虑到智能驾驶领域对目标检测精度和实时性的双重需求,仅将改进后的模型与部分主流的单阶段目标检测算法:SSD[22]、CenterNet[23]、YOLOv3[24]、YOLOv5、YOLOv7[25]、YOLOX[26]、YOLOv9[27]进行比较,对比实验结果如表6所示。
相对于表6中其他主流模型,本文模型的参数量和计算量是最小的,同时检测精度高于其他检测算法。本文模型的检测精度和速度相较于SSD和CenterNet而言有较大的提升,mAP分别提升了31.4%和35.0%,FPS也有较大幅度提升。与YOLO系列其他版本的模型相比,本文模型的mAP值有较为显著的提升,其中提升幅度最大的是相对于YOLOX-tiny提高了27.5%,但FPS值略低;YOLOv8s与本文模型的mAP值最接近,但其模型参数量和计算量较大,未满足轻量化需求,而本文模型的参数量和计算量仅仅相当于YOLOv8s的25.2%和36.8%;对于YOLOv9s模型,虽然与本文模型的mAP值较为接近,但计算复杂度相对较高,导致检测速度变慢。因此本文模型更易部署到低成本、低算力的移动设备,兼顾了检测性能和模型的轻量化。
为了验证改进后模型的泛化能力,选择在目标检测数据集PASCAL VOC 2012[28]和自动驾驶领域数据集KITTI[29]上进行实验与分析。本次实验中使用的PASCAL VOC 2012共包含17 125张图片,其中训练集和测试集分别包含13 870张和3 255张;KITTI共包含7 481张图片,训练集和测试集分别包含6 058张和1 423张。实验结果如表7表8所示。
表7表8中结果可知,改进后模型的mAP在PASCAL VOC2012和KITTI上分别相较于原始模型提升了0.6%和1.3%,速度上仍满足实时性需求。在自动驾驶领域的KITTI数据集上涨幅相对较高,由此可说明,本文模型具有较好的泛化能力,验证了本文模型的有效性。
为了提高智能驾驶汽车对路面中小型障碍物的检测精度,提出一种改进YOLOv8n的目标检测算法。在Backbone网络中融入DSConv减少了冗余计算量,针对数据集中存在大量的小目标障碍物这一问题,添加160×160特征图的小目标检测层加强对小目标的特征提取能力,提出一种SPPF-SimAM模块更加关注目标的重要特征,最后采用GSConv与VoV-GSCSP的组合模块轻量化Neck网络。实验结果表明,本文模型的mAP相较于原YOLOv8n模型提升了6.2%,参数量降低了6.7%,在检测性能与模型轻量化之间达到较好的平衡。面对大量的小目标障碍物,本文模型能够实现更好的检测效果,且具有较好的泛化能力。
对比其他主流模型,所提模型的综合性能更强,但是在检测速度上仍有提升的空间。由于实际场景中的障碍物目标与驾驶路况更为复杂多变,未来研究将优先从提升数据集质量出发,扩充障碍物类别数量,解决样本不均衡问题,提升模型应用的泛化能力与应用价值。
  • 国家自然科学基金(62303394)
  • 新疆维吾尔自治区自然科学基金(2022D01C694)
参考文献 引证文献
排序方式:
[1]
茅智慧, 朱佳利, 吴鑫, . 基于YOLO的自动驾驶目标检测研究综述[J]. 计算机工程与应用, 2022, 58(15): 68-77.
Mao Zhihui, Zhu Jiali, Wu Xin, et al. Review of YOLO based target detection for autonomous driving[J]. Computer Engineering and Applications, 2022, 58(15): 68-77.
[2]
Breitenstein J, Termöhlen J A, Lipinski D, et al. Corner cases for visual perception in automated driving: some guidance on detection approaches[J]. ArXiv Preprint, 2021: 2102.05897.
[3]
Gupta K, Javed S A, Gandhi V, et al. Mergenet: a deep net architecture for small obstacle discovery[J]. ArXiv Preprint, 2018: 1803.06508.
[4]
俞骏威, 张黎明, 陈凯, . 基于道路消失点的远距离路面微小障碍物检测[J]. 同济大学学报(自然科学版), 2019, 47(S1): 213-216.
Yu Junwei, Zhang Liming, Chen Kai, et al. Long-distance small road obstacles detection based on road vanishing point[J]. Journal of Tongji University(Natural Science), 2019, 47(S1): 213-216.
[5]
Song T J, Jeong J, Kim J H. End-to-end real-time obstacle detection network for safe self-driving via multi-task learning[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(9): 16318-16329.
[6]
吴晨曦, 应保胜, 许小伟, . 基于改进单步多框目标检测的道路小目标检测算法[J]. 科学技术与工程, 2023, 23(5): 2051-2058.
Wu Chenxi, Ying Baosheng, Xu Xiaowei, et al. Road small target detection algorithm based on improved single shot multibox detector[J]. Science Technology and Engineering, 2023, 23(5): 2051-2058.
[7]
李旭, 宋焕生, 史勤, . CIEFRNet: 面向高速公路的抛洒物检测算法[J]. 计算机工程与应用, 2024, 60(5): 336-346.
Li Xu, Song Huansheng, Shi Qin, et al. CIEFRNet: abandoned objects detection algorithm for highway[J]. Computer Engineering and Applications, 2024, 60(5): 336-346.
[8]
Nascimento M G, Fawcett R, Prisacariu V A. DSConv: efficient convolution operator[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 5148-5157.
[9]
Yang L, Zhang R Y, Li L, et al. SimAM:a simple, parameter-free attention module for convolutional neural networks[C]// International Conference on Machine Learning. Online: PMLR, 2021: 11863-11874.
[10]
Li H, Li J, Wei H, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[J]. ArXiv Preprint, 2022: 2206.02424.
[11]
Pinggera P, Ramos S, Gehrig S, et al. Lost and found:detecting small road hazards for self-driving vehicles[C]//2016 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS). Daejeon: IEEE, 2016: 1099-1106.
[12]
Singh A, Kamireddypalli A, Gandhi V, et al. Lidar guided small obstacle segmentation[C]//2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Las Vegas: IEEE, 2020: 8513-8520.
[13]
Maag K, Chan R, Uhlemeyer S, et al. Two video data sets for tracking and retrieval of out of distribution objects[C]// Proceedings of the Asian Conference on Computer Vision. Macau: Springer, 2022: 3776-3794.
[14]
Li K, Chen K, Wang H, et al. Coda: a real-world road corner case dataset for object detection in autonomous driving[C]// European Conference on Computer Vision. Tel Aviv: Springer, 2022: 406-423.
[15]
Li C, Zhou A, Yao A. Omni-dimensional dynamic convolution[J]. ArXiv Preprint, 2022: 2209.07947.
[16]
Chollet F. Xception: deep learning with depthwise separable convolutions[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 1251-1258.
[17]
Han K, Wang Y, Tian Q, et al. Ghostnet: more features from cheap operations[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. Seattle: IEEE, 2020: 1580-1589.
[18]
Misra D, Nalamada T, Arasanipalai A, et al. Rotate to attend: convolutional triplet attention module[C]// IEEE/CVF Winter Conference on Applications of Computer Vision. Waikoloa: IEEE, 2021: 3139-3148.
[19]
Zhang Q L, Yang Y B. SA-Net: shuffle attention for deep convolutional neural networks[C]// ICASSP-IEEE International Conference on Acoustics. Toronto: IEEE, 2021: 2235-9.
[20]
Lau K W, Po L M, Rehman Y A U. Large separable kernel attention: rethinking the large kernel attention design in CNN[J]. Expert Systems with Applications, 2024, 236: 121352.
[21]
Li Y H, Yao T, Pan Y W, et al. Contextual transformer networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(2): 1489-1500.
[22]
Liu W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector[C]// European Conference on Computer Vision. Amsterdam: Springer, 2016: 21-37.
[23]
Zhou X, Wang D, Krähenbühl P. Objects as points[J]. ArXiv Preprint, 2019: 1904.07850.
[24]
Redmon J, Farhadi A. YOLOv3: an incremental improvement[J]. ArXiv Preprint, 2018: 1804.02767.
[25]
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. Vancouver: IEEE, 2023: 7464-7475.
[26]
Ge Z, Liu S, Wang F, et al. YOLOX: exceeding YOLO series in 2021[J]. ArXiv Preprint, 2021: 2107.08430.
[27]
Wang C Y, Yeh I H, Liao H Y M. YOLOv9: learning what you want to learn using programmable gradient information[J]. ArXiv Preprint, 2024: 2402.13616.
[28]
Everingham M, Eslami S A, Van Gool L, et al. The pascal visual object classes challenge: a retrospective[J]. International Journal of Computer Vision, 2015, 111: 98-136.
[29]
Geiger A, Lenz P, Stiller C, et al. Vision meets robotics: the KITTI dataset[J]. The International Journal of Robotics Research, 2013, 32(11): 1231-1237.
2025年第25卷第9期
PDF下载
420
170
引用本文
BibTeX
文章信息
doi: 10.12404/j.issn.1671-1815.2402441
  • 接收时间:2024-04-06
  • 首发时间:2025-07-09
  • 出版时间:2025-03-28
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2024-04-06
  • 修回日期:2024-12-06
基金
国家自然科学基金(62303394)
新疆维吾尔自治区自然科学基金(2022D01C694)
作者信息
    新疆大学电气工程学院, 乌鲁木齐 830017

通讯作者:

* 南新元(1967—),男,汉族,新疆乌鲁木齐人,硕士,教授。研究方向:智能控制系统开发、机器视觉。E-mail:
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2402441
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