Article(id=1279513094535754456, tenantId=1146029695717560320, journalId=1278651732997652489, issueId=1279511628118986881, articleNumber=null, orderNo=null, doi=10.12086/oee.2026.250340, pmid=null, cstr=32245.14.oee.2026.250340, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1762358400000, receivedDateStr=2025-11-06, revisedDate=1768060800000, revisedDateStr=2026-01-11, acceptedDate=1768147200000, acceptedDateStr=2026-01-12, onlineDate=1782989293940, onlineDateStr=2026-07-02, pubDate=1776960000000, pubDateStr=2026-04-24, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1782989293940, onlineIssueDateStr=2026-07-02, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1782989293940, creator=13701087609, updateTime=1782989293940, updator=13701087609, issue=Issue{id=1279511628118986881, tenantId=1146029695717560320, journalId=1278651732997652489, year='2026', volume='53', issue='4', pageStart='250244', pageEnd='250340', issueExtLink='null', onlineDate='null', pubDate='1776960000000', pubDateStr='2026-04-24', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=1, specialIssue=null, createTime=1782988944320, creator='13701087609', updateTime=1782988944320, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext=null, issueFiles=null, downloadFileDto=null}, startPage=250340, endPage=, ext={EN=ArticleExt(id=1279513096720986842, articleId=1279513094535754456, tenantId=1146029695717560320, journalId=1278651732997652489, language=EN, title=BEV object detection method with multi-layer perceptual enhancement and gated alignment fusion, columnId=1279511634116841602, journalTitle=Opto-Electronic Engineering, columnName=Article, runingTitle=null, highlight=null, articleAbstract=
Objective:

Multimodal three-dimensional object detection combines camera semantics with LiDAR geometry and plays a central role in autonomous driving perception. Yet performance often degrades in complex traffic due to unstable semantics, sparse geometry, and cross-modal misalignment. Camera features become unreliable under glare, low illumination, motion blur, and cluttered backgrounds, introducing uncertainty into bird’s-eye-view (BEV) projection. LiDAR point clouds are sparse at long range or in adverse weather, weakening local geometry and causing boundary blur and missed small objects. In BEV fusion, misalignment between modalities introduces conflicting cues and redundant responses, reducing confidence and accuracy. This study proposes an efficient BEV detector that enhances camera semantics, stabilizes sparse geometry, and achieves adaptive cross-modal fusion for improved accuracy and robustness without excessive computational cost.

Methods:

The proposed framework, MEPFusion (multi-layer perception enhancement and gated alignment fusion), improves camera semantics, stabilizes sparse-geometry modeling, and enables gated alignment fusion in BEV space. For camera encoding, SimPVT enhances global context and salient-region representation with low overhead. A pyramid vision transformer backbone extracts multi-scale features to capture global and local dependencies. A parameter-free saliency reweighting unit amplifies target-related activations and suppresses background distraction, improving semantic separability. To reduce computation, the highest-resolution stage is removed while preserving three-scale outputs for projection. Features are lifted to BEV space via a lift–splat–shoot transformation supervised by depth estimation, stabilizing the depth distribution and reducing projection noise to improve spatial consistency. This camera branch aims to provide more reliable semantics under appearance variations, reducing texture-induced false positives and improving recall for visually small instances.

For LiDAR encoding, raw point clouds are voxelized and processed through a SECOND-style sparse convolution backbone. Standard sparse convolutions are replaced with spatial density-modulated convolution (SDMConv) to address equal-weight aggregation in sparse regions. SDMConv injects a position-dependent density prior into convolution responses via multiplicative modulation, producing a center-enhanced and periphery-decayed pattern within the receptive field. The modulation mitigates feature dilution and preserves boundary cues and shape continuity, especially in long-range or partially occluded areas, without increasing parameter count. The resulting sparse 3D features are compressed along height to produce LiDAR BEV features. In practice, SDMConv acts as a plug-in operator that strengthens local geometry when point evidence is weak, benefiting thin structures and small obstacles that are easily under-sampled.

Cross-modal fusion occurs in BEV space through a gated alignment fusion module (GAFM) designed to suppress misalignment while maintaining complementarity. Each modality undergoes channel recalibration to emphasize discriminative channels. A residual mutual-guidance path performs bidirectional correction: LiDAR features refine camera BEV features in geometry-sensitive regions, while camera semantics complement LiDAR features where geometric evidence is weak. Spatial gates are predicted for each BEV cell and normalized via softmax to assign adaptive fusion weights, allowing the more reliable modality to dominate locally while reducing inconsistent responses. The fused BEV features are decoded by a multi-scale neck and a transformer-based detection head to produce 3D bounding boxes and attributes. This strategy improves optimization stability and encourages reliability-aware weighting rather than uniform mixing.

Results and Discussions:

Extensive experiments on the nuScenes benchmark demonstrate the effectiveness of MEPFusion. With six cameras and one LiDAR input, MEPFusion achieves 67.1% mAP and 70.8% NDS, surpassing the BEVFusion baseline by 2.8% and 1.7%, respectively. Regression errors in translation, scale, orientation, velocity, and attribute are reduced, confirming improved spatial alignment and motion estimation. Category-level analysis shows notable gains for small or geometry-sensitive objects (e.g., traffic cones, barriers, motorcycles, and pedestrians) while maintaining improvements for major vehicle classes. These trends align with module contributions: SimPVT strengthens semantics in cluttered scenes, SDMConv alleviates sparsity-induced degradation, and GAFM reduces cross-modal interference through adaptive gating and complementary fusion. Qualitative inspection suggests fewer duplicated boxes and cleaner spatial responses in crowded regions, consistent with suppressing redundant activations caused by misalignment.

Distance-stratified evaluation further verifies consistent improvements across near, mid, and far ranges. The far range, where point sparsity and scale ambiguity are most severe, still achieves clear gains, highlighting the benefits of density-aware encoding and local gating. Robustness tests under adverse conditions also validate generalization. On the night subset, MEPFusion improves mAP and NDS while reducing error terms, indicating stable semantics and selective fusion under low illumination and glare. On the rain subset, performance again improves, confirming resilience to reduced contrast and weaker LiDAR returns. Ablation studies reveal both independent and synergistic effects: SimPVT alone enhances accuracy with fewer parameters; SDMConv and GAFM each bring additional gains; combining all three yields the best overall performance. Efficiency analysis shows SDMConv introduces minimal overhead, and the full model maintains comparable throughput with reduced total parameters, indicating a favorable accuracy–efficiency trade-off for on-vehicle inference.

Conclusions:

In summary, MEPFusion integrates semantic enhancement, sparse-geometry reinforcement, and gated alignment fusion within a unified BEV framework. The method achieves higher accuracy, improved regression stability, and consistent robustness across distance and environmental variations while retaining deployment-friendly efficiency. Stable semantics before view transformation, density-modulated LiDAR encoding, and adaptive BEV fusion jointly mitigate misalignment and suppress redundant noise. MEPFusion thus offers an effective and reliable multimodal perception solution for urban autonomous driving.

, authors=Hongen Liu1, 2, Xu Li1, *, Sheng Huang1, Ting Kang3, authorsList=Hongen Liu, Xu Li, Sheng Huang, Ting Kang, authorCompany=null, correspAuthors=Xu Li, authorNote=null, correspAuthorsNote=
, copyrightStatement=Copyright © 2026 Opto-Electronic Engineering. All rights reserved., 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, fund=null), CN=ArticleExt(id=1279513127763030930, articleId=1279513094535754456, tenantId=1146029695717560320, journalId=1278651732997652489, language=CN, title=多层感知增强与门控对齐融合的BEV目标检测方法, columnId=1279511637858160772, journalTitle=光电工程, columnName=科研论文, runingTitle=null, highlight=null, articleAbstract=

针对多模态3D目标检测中图像语义受限、点云稀疏致几何退化与跨模态对齐困难等问题,提出一种多层感知增强与门控对齐融合的BEV目标检测方法(MEPFusion)。首先图像端利用SimPVT强化全局语义与关键区域建模;其次点云端设计空间密度调制卷积,以位置依赖权重注入密度先验,稳固稀疏场景的局部几何;最后融合端构建双模态门控对齐模块,通过通道与空间门控抑制错配并实现动态校准,促成BEV特征的动态对齐与互补交互。实验结果表明,本文方法在nuScenes上的mAP与NDS达到67.1%与70.8%,较BEVFusion分别提升2.8%和1.7%,显著提高检测精度与鲁棒性。消融结果验证MEPFusion在语义特征提取、几何建模与跨模态对齐三方面具备独立增益并产生协同效应,整体上在融合质量、检测性能与计算效率之间实现更优平衡,为自动驾驶三维目标检测提供高效可靠的方案。

, authors=刘鸿恩1, 2, 李旭1, *, 黄晟1, 康婷3, authorsList=刘鸿恩, 李旭, 黄晟, 康婷, authorCompany=null, correspAuthors=李旭, authorNote=

刘鸿恩(1987-),男,博士,副教授,研究方向为智能交通系统融合感知、自学习特征建模和协同优化控制方法研究。E-mail:

李旭(2001-),男,硕士研究生,主要研究方向为机器学习、多模态感知等。E-mail:

, correspAuthorsNote=
李旭,
, copyrightStatement=版权所有©《光电工程》编辑部 2026, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=NCrNKvnz6VbTsg9rqXRZZw==, magXml=VHtCdvMYsJoJd6AdvXpanQ==, pdfUrl=null, pdf=8PXz7ZxfbuDC+1bVVO9RZA==, pdfFileSize=7407248, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=IsdnBSpeLtjahz9R3qJosw==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=Y76R4VBlffN8iZwAHMFh+A==, mapNumber=null, fund=null)}, authors=[Author(id=1280951079814808156, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=len3610@126.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1280951079890305631, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951079814808156, language=EN, stringName=Hongen Liu, firstName=Hongen, middleName=null, lastName=Liu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China
2Institute of Permanent-Magnet Maglev Technology and Rail Transit, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1280951079965803104, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951079814808156, language=CN, stringName=刘鸿恩, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000
2江西理工大学永磁磁浮技术与轨道交通研究院,江西 赣州 341000, bio={"img":"bo8nFNNt3aqJ4iyYhCbi6g==","content":"

刘鸿恩(1987-),男,博士,副教授,研究方向为智能交通系统融合感知、自学习特征建模和协同优化控制方法研究。E-mail:

"}, bioImg=bo8nFNNt3aqJ4iyYhCbi6g==, bioContent=

刘鸿恩(1987-),男,博士,副教授,研究方向为智能交通系统融合感知、自学习特征建模和协同优化控制方法研究。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1280951079529595474, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=1, ext=[AuthorCompanyExt(id=1280951079537984083, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079546372692, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000)]), AuthorCompany(id=1280951079621870165, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=2, ext=[AuthorCompanyExt(id=1280951079634453078, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079621870165, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Permanent-Magnet Maglev Technology and Rail Transit, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079642841687, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079621870165, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2江西理工大学永磁磁浮技术与轨道交通研究院,江西 赣州 341000)])]), Author(id=1280951080053883490, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=885239570@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1280951080125186660, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080053883490, language=EN, stringName=Xu Li, firstName=Xu, middleName=null, lastName=Li, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, *, address=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1280951080204878437, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080053883490, language=CN, stringName=李旭, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, *, address=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000, bio={"img":"4oj0L19gjbZj6+6xQK2NFw==","content":"

李旭(2001-),男,硕士研究生,主要研究方向为机器学习、多模态感知等。E-mail:

"}, bioImg=4oj0L19gjbZj6+6xQK2NFw==, bioContent=

李旭(2001-),男,硕士研究生,主要研究方向为机器学习、多模态感知等。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1280951079529595474, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=1, ext=[AuthorCompanyExt(id=1280951079537984083, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079546372692, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000)])]), Author(id=1280951080292958823, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1280951080364261993, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080292958823, language=EN, stringName=Sheng Huang, firstName=Sheng, middleName=null, lastName=Huang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1280951080435565162, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080292958823, language=CN, stringName=黄晟, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1280951079529595474, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=1, ext=[AuthorCompanyExt(id=1280951079537984083, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079546372692, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000)])]), Author(id=1280951080511062636, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1280951080599143022, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080511062636, language=EN, stringName=Ting Kang, firstName=Ting, middleName=null, lastName=Kang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1280951080670446191, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080511062636, language=CN, stringName=康婷, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3江西理工大学电气工程与自动化学院,江西 赣州 341000, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1280951079718339160, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=3, ext=[AuthorCompanyExt(id=1280951079730922073, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079718339160, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079735116378, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079718339160, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3江西理工大学电气工程与自动化学院,江西 赣州 341000)])])], keywords=[Keyword(id=1280951080834024048, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, orderNo=1, keyword=autonomous driving), Keyword(id=1280951080913715825, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, orderNo=2, keyword=multimodal fusion), Keyword(id=1280951080993407602, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, orderNo=3, keyword=BEV representation), Keyword(id=1280951081073099379, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, orderNo=4, keyword=multi-layer perception enhancement), Keyword(id=1280951081148596852, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, orderNo=5, keyword=gated alignment fusion), Keyword(id=1280951081207317109, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, orderNo=1, keyword=自动驾驶), Keyword(id=1280951081266037366, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, orderNo=2, keyword=多模态融合), Keyword(id=1280951083040227961, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, orderNo=3, keyword=BEV表征), Keyword(id=1280951083145085562, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, orderNo=4, keyword=多层感知增强), Keyword(id=1280951083258331771, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, orderNo=5, keyword=门控对齐融合)], refs=[Reference(id=1280951088320856734, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=1, rfOrder=0, authorNames=null, journalName=null, refType=null, unstructuredReference=李文礼, 喻飞, 石晓辉, 等. BEV特征下激光雷达和单目相机融合的目标检测算法研究[J]. 计算机工程与应用, 2024, 60(11): 182−193., articleTitle=null, refAbstract=null), Reference(id=1280951088400548511, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=1, rfOrder=1, authorNames=null, journalName=null, refType=null, unstructuredReference=Li W L, Yu F, Shi X H, et al. Research on target detection algorithm for fusion of LiDAR and monocular camera under BEV features[J]. Comput Eng Appl, 2024, 60(11): 182−193., articleTitle=null, refAbstract=null), Reference(id=1280951088463463072, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=2, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen S, Ma Y, Qiao Y, et al. M-BEV: Masked bev perception for robust autonomous driving[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2024, 38(2): 1183-1191. https://doi.org/10.1609/aaai.v38i2.2788, articleTitle=null, refAbstract=null), Reference(id=1280951088534766241, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=3, authorNames=null, journalName=null, refType=null, unstructuredReference=Lang B, Li X, Chuah M C. BEV-TP: end-to-end visual perception and trajectory prediction for autonomous driving[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(11): 18537−18546, articleTitle=null, refAbstract=null), Reference(id=1280951088601875106, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=4, rfOrder=4, authorNames=null, journalName=null, refType=null, unstructuredReference=Ge C J, Chen J S, Xie E Z, et al. MetaBEV: solving sensor failures for 3D detection and map segmentation[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision, 2023: 8687–8697. https://doi.org/10.1109/ICCV51070.2023.00801., articleTitle=null, refAbstract=null), Reference(id=1280951088681566883, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=5, rfOrder=5, authorNames=null, journalName=null, refType=null, unstructuredReference=王五岳, 徐召飞, 曲春燕, 等. 基于红外与激光雷达融合的鸟瞰图空间三维目标检测算法[J]. 光子学报, 2024, 53(1): 0111002., articleTitle=null, refAbstract=null), Reference(id=1280951088757064356, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=5, rfOrder=6, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang W Y, Xu Z F, Qu C Y, et al. BEV space 3D object detection algorithm based on fusion of infrared camera and LiDAR[J]. Acta Photonica Sin, 2024, 53(1): 0111002., articleTitle=null, refAbstract=null), Reference(id=1280951088832561829, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=7, authorNames=null, journalName=null, refType=null, unstructuredReference=吉宇哲, 陈奕洁, 杨柳青, 等. 基于鸟瞰图的空间-通道注意力多传感器融合[J]. 模式识别与人工智能, 2023, 36(11): 1029−1040., articleTitle=null, refAbstract=null), Reference(id=1280951088899670694, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=8, authorNames=null, journalName=null, refType=null, unstructuredReference=Ji Y Z, Chen Y J, Yang L Q, et al. Spatial-channel attention multi-sensor fusion based on Bird's-Eye View[J]. Pattern Recognit Artif Intell, 2023, 36(11): 1029−1040., articleTitle=null, refAbstract=null), Reference(id=1280951088983556775, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=7, rfOrder=9, authorNames=null, journalName=null, refType=null, unstructuredReference=Philion J, Fidler S. Lift, splat, shoot: Encoding images from arbitrary camera rigs by implicitly unprojecting to 3D[C]//Proceedings of the 16th European Conference on Computer Vision, 2020: 194–210. https://doi.org/10.1007/978-3-030-58568-6_12., articleTitle=null, refAbstract=null), Reference(id=1280951089067442856, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=8, rfOrder=10, authorNames=null, journalName=null, refType=null, unstructuredReference=Huang J J, Huang G, Zhu Z, et al. BEVDet: high-performance multi-camera 3D object detection in bird-eye-view[Z]. arXiv: 2112.11790, 2022. https://doi.org/10.48550/arXiv.2112.11790., articleTitle=null, refAbstract=null), Reference(id=1280951089159717545, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=9, rfOrder=11, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Y H, Ge Z, Yu G Y, et al. BEVDepth: acquisition of reliable depth for multi-view 3D object detection[C]//Proceedings of the 37th AAAI Conference on Artificial Intelligence, 2023: 1477–1485. https://doi.org/10.1609/aaai.v37i2.25233., articleTitle=null, refAbstract=null), Reference(id=1280951089231020714, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=10, rfOrder=12, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Z Q, Wang W H, Li H Y, et al. BEVFormer: learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers[Z]. arXiv: 2203.17270, 2022. https://doi.org/10.48550/arXiv.2203.17270., articleTitle=null, refAbstract=null), Reference(id=1280951089323295403, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=11, rfOrder=13, authorNames=null, journalName=null, refType=null, unstructuredReference=陈旭东, 韩丽丽, 肖英豪, 薛保国, 马留洋. 基于体素-关键点特征聚合网络的点云3维目标检测[J]. 激光技术, 2026, 50(2): 291−299., articleTitle=null, refAbstract=null), Reference(id=1280951089419764396, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=11, rfOrder=14, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen X D, Han L L, Xiao Y H, et al. 3-D object detection for point clouds based on Voxel-Keypoint Feature Aggregation Network[J]. Laser Technol, 2026, 50(2): 291−299., articleTitle=null, refAbstract=null), Reference(id=1280951089499456173, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=12, rfOrder=15, authorNames=null, journalName=null, refType=null, unstructuredReference=Lang A H, Vora S, Caesar H, et al. PointPillars: fast encoders for object detection from point clouds[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 12689–12697. https://doi.org/10.1109/CVPR.2019.01298., articleTitle=null, refAbstract=null), Reference(id=1280951089574953646, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=13, rfOrder=16, authorNames=null, journalName=null, refType=null, unstructuredReference=Yan, Y, Mao Y X, Li B. SECOND: sparsely embedded convolutional detection[J]. Sensors, 2018, 18(10): 3337., articleTitle=null, refAbstract=null), Reference(id=1280951089650451119, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=14, rfOrder=17, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang H, Tao L, Cai Y F, et al. CenterPoint-SE: a single-stage anchor-free 3-D object detection algorithm with spatial awareness enhancement[J]. IEEE Trans Intell Transp Syst, 2023, 24(10): 10760−10773., articleTitle=null, refAbstract=null), Reference(id=1280951089738531504, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=18, authorNames=null, journalName=null, refType=null, unstructuredReference=蔡怀宇, 杨朝乾, 崔子扬, 等. 图像引导和点云空间约束的公路洒落物检测定位方法[J]. 光电工程, 2024, 51(3): 230317., articleTitle=null, refAbstract=null), Reference(id=1280951089826611889, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=19, authorNames=null, journalName=null, refType=null, unstructuredReference=Cai H Y, Yang Z Q, Cui Z Y, et al. Image-guided and point cloud space-constrained method for detection and localization of abandoned objects on the road[J]. Opto-Electron Eng, 2024, 51(3): 230317., articleTitle=null, refAbstract=null), Reference(id=1280951089918886578, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=20, authorNames=null, journalName=null, refType=null, unstructuredReference=Bai X Y, Hu Z Y, Zhu X G, et al. Transfusion: robust LiDAR-camera fusion for 3D object detection with transformers[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 1080–1089. https://doi.org/10.1109/CVPR52688.2022.00116., articleTitle=null, refAbstract=null), Reference(id=1280951090006966963, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=17, rfOrder=21, authorNames=null, journalName=null, refType=null, unstructuredReference=Liu Z J, Tang H T, Amini A, et al. BEVFusion: multi-task multi-sensor fusion with unified bird's-eye view representation[C]//Proceedings of 2023 IEEE International Conference on Robotics and Automation (ICRA), 2023: 2774–2781. https://doi.org/10.1109/ICRA48891.2023.10160968., articleTitle=null, refAbstract=null), Reference(id=1280951090086658740, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=18, rfOrder=22, authorNames=null, journalName=null, refType=null, unstructuredReference=Yin T W, Zhou X Y, Krähenbühl P. Multimodal virtual point 3d detection[C]//Proceedings of the 35th International Conference on Neural Information Processing Systems, 2021: 1261., articleTitle=null, refAbstract=null), Reference(id=1280951090170544821, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=19, rfOrder=23, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen Z H, Li Z Y, Zhang S Q, et al. Deformable feature aggregation for dynamic multi-modal 3D object detection[C]//Proceedings of the 17th European Conference on Computer Vision, 2022: 628–644. https://doi.org/10.1007/978-3-031-20074-8_36., articleTitle=null, refAbstract=null), Reference(id=1280951090262819510, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=20, rfOrder=24, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Y W, Chen Y L, Qi X J, et al. Unifying voxel-based representation with transformer for 3D object detection[C]//Proceedings of the 36th International Conference on Neural Information Processing Systems, 2022: 1340., articleTitle=null, refAbstract=null), Reference(id=1280951091961512631, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=21, rfOrder=25, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang W H, Xie E Z, Li X, et al. PVT v2: improved baselines with pyramid vision transformer[J]. Comput Vis Media, 2022, 8(3): 415−424., articleTitle=null, refAbstract=null), Reference(id=1280951092024427192, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=22, rfOrder=26, authorNames=null, journalName=null, refType=null, unstructuredReference=Yang L X, Zhang R Y, Li L D, et al. SimAm: a simple, parameter-free attention module for convolutional neural networks[C]//Proceedings of the 38th International Conference on Machine Learning, 2021: 11863–11874., articleTitle=null, refAbstract=null), Reference(id=1280951092141867705, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=23, rfOrder=27, authorNames=null, journalName=null, refType=null, unstructuredReference=Cammarasana S, Patanè G. Optimal weighted convolution for classification and denosing[Z]. arXiv: 2505.24558, 2025.https://doi.org/10.48550/arXiv.2505.24558., articleTitle=null, refAbstract=null), Reference(id=1280951092255113914, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=24, rfOrder=28, authorNames=null, journalName=null, refType=null, unstructuredReference=栾庆磊, 常昕昱, 吴叶, 等. PAW-YOLOv7: 河道微小漂浮物检测算法[J]. 光电工程, 2024, 51(4): 240025., articleTitle=null, refAbstract=null), Reference(id=1280951092355777211, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=24, rfOrder=29, authorNames=null, journalName=null, refType=null, unstructuredReference=Luan Q L, Chang X Y, Wu Y, et al. PAW-YOLOv7: algorithm for detection of tiny floating objects in river channels[J]. Opto-Electron Eng, 2024, 51(4): 240025., articleTitle=null, refAbstract=null), Reference(id=1280951092439663292, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=25, rfOrder=30, authorNames=null, journalName=null, refType=null, unstructuredReference=夏金泽, 孙浩铭, 胡盛辉, 等. 基于图像信息约束的三维激光点云聚类方法[J]. 光电工程, 2023, 50(2): 220148., articleTitle=null, refAbstract=null), Reference(id=1280951092527743677, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=25, rfOrder=31, authorNames=null, journalName=null, refType=null, unstructuredReference=Xia J Z, Sun H M, Hu S H, et al. 3D laser point cloud clustering method based on image information constraints[J]. Opto-Electron Eng, 2023, 50(2): 220148., articleTitle=null, refAbstract=null), Reference(id=1280951092590658238, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=26, rfOrder=32, authorNames=null, journalName=null, refType=null, unstructuredReference=张善文, 邵彧. 一种改进U-Net的ARSI目标检测方法[J]. 电光与控制, 2025, 32(9): 1−7., articleTitle=null, refAbstract=null), Reference(id=1280951092661961407, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=26, rfOrder=33, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang S W, Shao Y. Modified U-Net based ARSI object detection[J]. Electron Opt Control, 2025, 32(9): 1−7., articleTitle=null, refAbstract=null), Reference(id=1280951092745847488, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=27, rfOrder=34, authorNames=null, journalName=null, refType=null, unstructuredReference=Vora S, Lang A H, Helou B, et al. PointPainting: sequential fusion for 3D object detection[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 4603–4611. https://doi.org/10.1109/CVPR42600.2020.00466., articleTitle=null, refAbstract=null), Reference(id=1280951092808762049, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=28, rfOrder=35, authorNames=null, journalName=null, refType=null, unstructuredReference=Zeng Y H, Zhang D, Wang C W, et al. LIFT: learning 4D LiDAR image fusion transformer for 3D object detection[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 17151–17160. https://doi.org/10.1109/CVPR52688.2022.01666., articleTitle=null, refAbstract=null), Reference(id=1280951092901036738, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=29, rfOrder=36, authorNames=null, journalName=null, refType=null, unstructuredReference=Cui L C, Li X X, Meng M, et al. MMFusion: a generalized multi-modal fusion detection framework[C]//Proceedings of 2023 IEEE International Conference on Development and Learning (ICDL), 2023: 415–422. https://doi.org/10.1109/ICDL55364.2023.10364441., articleTitle=null, refAbstract=null), Reference(id=1280951092997505731, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=30, rfOrder=37, authorNames=null, journalName=null, refType=null, unstructuredReference=Song Z Y, Wei H Y, Bai L, et al. GraphAlign: enhancing accurate feature alignment by graph matching for multi-modal 3D object detection[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision, 2023: 3335–3346. https://doi.org/10.1109/ICCV51070.2023.00311., articleTitle=null, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1280951079529595474, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=1, ext=[AuthorCompanyExt(id=1280951079537984083, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079546372692, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000)]), AuthorCompany(id=1280951079621870165, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=2, ext=[AuthorCompanyExt(id=1280951079634453078, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079621870165, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Permanent-Magnet Maglev Technology and Rail Transit, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079642841687, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079621870165, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2江西理工大学永磁磁浮技术与轨道交通研究院,江西 赣州 341000)]), AuthorCompany(id=1280951079718339160, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=3, ext=[AuthorCompanyExt(id=1280951079730922073, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079718339160, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079735116378, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079718339160, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3江西理工大学电气工程与自动化学院,江西 赣州 341000)])], figs=[ArticleFig(id=1280951083585487484, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Fig.1, caption=Overall framework of the proposed MEPFusion method, figureFileSmall=Q0xJRZjS20J0a8kkr/7A/A==, figureFileBig=+GdauLr6TvjMUoDQWDcBQQ==, tableContent=null), ArticleFig(id=1280951083677762173, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=图1, caption=MEPFusion方法总体架构图, figureFileSmall=Q0xJRZjS20J0a8kkr/7A/A==, figureFileBig=+GdauLr6TvjMUoDQWDcBQQ==, tableContent=null), ArticleFig(id=1280951083807785598, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Fig.2, caption=Architecture of the SimPVT feature extraction module, figureFileSmall=sOgKayraytQmkVb4gnpiMQ==, figureFileBig=TfqIAMwXEtB2I+OlxwyFFg==, tableContent=null), ArticleFig(id=1280951083891671679, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=图2, caption=特征提取SimPVT结构图, figureFileSmall=sOgKayraytQmkVb4gnpiMQ==, figureFileBig=TfqIAMwXEtB2I+OlxwyFFg==, tableContent=null), ArticleFig(id=1280951083967169152, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Fig.3, caption=Architecture of the point cloud branch enhanced by SDMConv, figureFileSmall=aSCPMr9crbRJvvy+Apet4w==, figureFileBig=UpIjqImHjbk8n9X8sBuTWQ==, tableContent=null), ArticleFig(id=1280951084063638145, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=图3, caption=SDMConv增强的点云分支结构, figureFileSmall=aSCPMr9crbRJvvy+Apet4w==, figureFileBig=UpIjqImHjbk8n9X8sBuTWQ==, tableContent=null), ArticleFig(id=1280951084160107138, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Fig.4, caption=Gated alignment fusion module, figureFileSmall=FsXw9PgwajuPquIc1zk1/w==, figureFileBig=R61YVfReZ4Rn48H2n06WcQ==, tableContent=null), ArticleFig(id=1280951084248187523, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=图4, caption=门控对齐融合模块, figureFileSmall=FsXw9PgwajuPquIc1zk1/w==, figureFileBig=R61YVfReZ4Rn48H2n06WcQ==, tableContent=null), ArticleFig(id=1280951084332073604, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Fig.5, caption=Visualization of object detection results in an urban daytime arterial road environment, figureFileSmall=Hma504TVdx9TKDzOER4RGQ==, figureFileBig=MjQ0BC+ACd13NmVgy6j/Pw==, tableContent=null), ArticleFig(id=1280951084403376773, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=图5, caption=城市白天主干道环境目标检测结果可视化, figureFileSmall=Hma504TVdx9TKDzOER4RGQ==, figureFileBig=MjQ0BC+ACd13NmVgy6j/Pw==, tableContent=null), ArticleFig(id=1280951084491457158, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Fig.6, caption=Visualization of object detection results in complex traffic environments, figureFileSmall=5HbAX/7jHdrv4CMIDfTMbA==, figureFileBig=FkNeY6a0z5hdtVXKwVVobw==, tableContent=null), ArticleFig(id=1280951084566954631, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=图6, caption=复杂交通环境目标检测结果可视化, figureFileSmall=5HbAX/7jHdrv4CMIDfTMbA==, figureFileBig=FkNeY6a0z5hdtVXKwVVobw==, tableContent=null), ArticleFig(id=1280951084638257800, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Fig.7, caption=Visualization of object detection results in a nighttime arterial road environment, figureFileSmall=zpKz6E0EDBI2aJ6m+YAnsg==, figureFileBig=p6C51oXuNr9E6SLfh5x7/w==, tableContent=null), ArticleFig(id=1280951084705366665, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=图7, caption=夜间主干道环境目标检测结果可视化, figureFileSmall=zpKz6E0EDBI2aJ6m+YAnsg==, figureFileBig=p6C51oXuNr9E6SLfh5x7/w==, tableContent=null), ArticleFig(id=1280951084797641354, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Fig.8, caption=Visualization of object detection results in a rainy weather environment, figureFileSmall=xjYhPIVsCzIIuKiaPEC7Ng==, figureFileBig=Zkijce+m4edJuR78PFE1nA==, tableContent=null), ArticleFig(id=1280951084868944523, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=图8, caption=阴雨天环境目标检测结果可视化, figureFileSmall=xjYhPIVsCzIIuKiaPEC7Ng==, figureFileBig=Zkijce+m4edJuR78PFE1nA==, tableContent=null), ArticleFig(id=1280951084957024908, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Tab.1, caption=

Comparison of detection performance across categories on the nuScenes test set (unit: %)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodmAPCarTruckC.V.BusT.L.BarrierM.T.BikePde.T.C.
BEVFusion64.387.059.725.973.543.268.270.158.686.371.9
Ours67.188.763.428.775.742.871.276.458.387.377.6
Gain+2.8+1.7+3.7+2.8+2.2−0.4+3+6.5−0.3+1.0+5.7
), ArticleFig(id=1280951085057688205, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=表1, caption=

nuScenes测试集类别目标检测精度对比(单位:%)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodmAPCarTruckC.V.BusT.L.BarrierM.T.BikePde.T.C.
BEVFusion64.387.059.725.973.543.268.270.158.686.371.9
Ours67.188.763.428.775.742.871.276.458.387.377.6
Gain+2.8+1.7+3.7+2.8+2.2−0.4+3+6.5−0.3+1.0+5.7
), ArticleFig(id=1280951085137379982, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Tab.2, caption=

Comparison of fine-grained evaluation metrics on the nuScenes test set (unit: %)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodNDS↑mATE↓mASE↓mAOE↓mAVE↓mAAE↓
BEVFusion69.128.425.531.726.518.6
Ours70.828.125.130.325.318.3
Gain+1.7−0.3−0.4−1.4−1.2−0.3
), ArticleFig(id=1280951085208683151, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=表2, caption=

nuScenes测试集细粒度指标对比(单位:%)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodNDS↑mATE↓mASE↓mAOE↓mAVE↓mAAE↓
BEVFusion69.128.425.531.726.518.6
Ours70.828.125.130.325.318.3
Gain+1.7−0.3−0.4−1.4−1.2−0.3
), ArticleFig(id=1280951085292569232, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Tab.3, caption=

Comparison of detection performance across distance ranges on the nuScenes validation set (unit: %)

, figureFileSmall=null, figureFileBig=null, tableContent=
Method0-20 m20-30 m30-50 m
mAPNDSmAPNDSmAPNDS
BEVFusion77.877.261.067.430.446.0
Ours78.678.062.668.431.846.8
Gain+0.8+0.8+1.6+1.0+1.4+0.8
), ArticleFig(id=1280951085376455313, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=表3, caption=

nuScenes 验证集距离分段性能对比(单位:%)

, figureFileSmall=null, figureFileBig=null, tableContent=
Method0-20 m20-30 m30-50 m
mAPNDSmAPNDSmAPNDS
BEVFusion77.877.261.067.430.446.0
Ours78.678.062.668.431.846.8
Gain+0.8+0.8+1.6+1.0+1.4+0.8
), ArticleFig(id=1280951085447758482, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Tab.4, caption=

Comparison of category-wise ap across distance ranges on the nuScenes validation set (unit: %)

, figureFileSmall=null, figureFileBig=null, tableContent=
Method0-20 m20-30 m30-50 m
Pde.M.T.CarTruckPde.M.T.CarTruckPde.M.T.CarTruck
BEVFusion92.087.296.774.582.366.788.763.343.426.273.147.3
MEPFusion/Ours92.489.996.779.383.568.888.863.543.730.474.349.9
ΔAP+0.4+2.7+0.0+4.8+1.2+2.1+0.1+0.2+0.3+4.2+1.2+2.6
), ArticleFig(id=1280951085510673043, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=表4, caption=

nuScenes验证集距离分段类别AP对比(单位:%)

, figureFileSmall=null, figureFileBig=null, tableContent=
Method0-20 m20-30 m30-50 m
Pde.M.T.CarTruckPde.M.T.CarTruckPde.M.T.CarTruck
BEVFusion92.087.296.774.582.366.788.763.343.426.273.147.3
MEPFusion/Ours92.489.996.779.383.568.888.863.543.730.474.349.9
ΔAP+0.4+2.7+0.0+4.8+1.2+2.1+0.1+0.2+0.3+4.2+1.2+2.6
), ArticleFig(id=1280951085670056596, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Tab.5, caption=

Comparison of 3D object detection performance on the nuScenes dataset (unit: %)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodModalityTestValidation
mAPNDSmAPNDS
BEVDetC42.248.239.347.2
BEVFormerC48.156.941.651.7
BEVDepthC50.360.035.147.5
PointPillarsL30.545.3
CenterPointL58.065.550.360.2
TransFusion-LL65.570.265.170.1
PointPainting[27]C+L46.458.1
MVPC+L66.470.567.170.8
LIFT[28]C+L65.170.2
mmFusion[29]C+L65.469.8
GraphAlign[30]C+L66.570.662.868.5
BEVFusionC+L64.369.165.269.7
OursC+L67.170.867.170.9
), ArticleFig(id=1280951085804274325, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=表5, caption=

nuScenes上三维目标检测性能对比结果(单位:%)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodModalityTestValidation
mAPNDSmAPNDS
BEVDetC42.248.239.347.2
BEVFormerC48.156.941.651.7
BEVDepthC50.360.035.147.5
PointPillarsL30.545.3
CenterPointL58.065.550.360.2
TransFusion-LL65.570.265.170.1
PointPainting[27]C+L46.458.1
MVPC+L66.470.567.170.8
LIFT[28]C+L65.170.2
mmFusion[29]C+L65.469.8
GraphAlign[30]C+L66.570.662.868.5
BEVFusionC+L64.369.165.269.7
OursC+L67.170.867.170.9
), ArticleFig(id=1280951087519744662, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Tab.6, caption=

Comparison of ablation study results on the nuScenes test set

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodSimPVTSDMConvGAFMmAP/%NDS/%FPS/%Params/M
Bevfusion64.369.16.839.80
E165.569.76.637.03
E264.669.56.939.08
E365.469.86.936.31
E466.270.07.037.03
E5/Ours67.170.86.836.31
), ArticleFig(id=1280951087603630743, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=表6, caption=

在nuScenes测试集上的消融实验对比结果

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodSimPVTSDMConvGAFMmAP/%NDS/%FPS/%Params/M
Bevfusion64.369.16.839.80
E165.569.76.637.03
E264.669.56.939.08
E365.469.86.936.31
E466.270.07.037.03
E5/Ours67.170.86.836.31
), ArticleFig(id=1280951087674933912, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Tab.7, caption=

Statistics of computational overhead and inference latency of different modules

, figureFileSmall=null, figureFileBig=null, tableContent=
ModuleParams/MGFLOPsInference time/ms
Conv2.7×10−56.71×10−33.5×10−2
SDMConv2.7×10−51.34×10−23.8×10−2
), ArticleFig(id=1280951087742042777, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=表7, caption=

模块计算开销与推理延迟统计

, figureFileSmall=null, figureFileBig=null, tableContent=
ModuleParams/MGFLOPsInference time/ms
Conv2.7×10−56.71×10−33.5×10−2
SDMConv2.7×10−51.34×10−23.8×10−2
), ArticleFig(id=1280951087821734554, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Tab.8, caption=

Quantitative robustness evaluation results on the night subset of the nuScenes validation set (unit: %)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodmAP↑NDS↑mATE↓mASE↓mAOE↓mAVE↓mAAE↓
BEVFusion37.843.648.945.143.157.361.1
Ours39.844.447.644.542.857.060.7
Gain+2.0+0.8−1.3−0.6−0.3−0.3−0.4
), ArticleFig(id=1280951087926592155, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=表8, caption=

nuScenes验证集Night子集鲁棒性定量评测结果(单位:%)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodmAP↑NDS↑mATE↓mASE↓mAOE↓mAVE↓mAAE↓
BEVFusion37.843.648.945.143.157.361.1
Ours39.844.447.644.542.857.060.7
Gain+2.0+0.8−1.3−0.6−0.3−0.3−0.4
), ArticleFig(id=1280951087997895324, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=EN, label=Tab.9, caption=

Quantitative robustness evaluation results on the rain subset of the nuScenes validation Set (unit: %)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodmAP↑NDS↑mATE↓mASE↓mAOE↓mAVE↓mAAE↓
BEVFusion64.770.029.526.933.120.013.8
Ours67.070.829.226.833.119.612.7
Gain+2.3+0.8−0.3−0.10−0.4−1.1
), ArticleFig(id=1280951088073392797, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, language=CN, label=表9, caption=

nuScenes验证集Rain子集鲁棒性定量评测结果(单位:%)

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodmAP↑NDS↑mATE↓mASE↓mAOE↓mAVE↓mAAE↓
BEVFusion64.770.029.526.933.120.013.8
Ours67.070.829.226.833.119.612.7
Gain+2.3+0.8−0.3−0.10−0.4−1.1
)], attaches=null, journal=Journal(id=1278641367198941188, delFlag=0, nameCn=光电工程, nameEn=Opto-Electronic Engineering, nameHistory1=null, nameHistory2=null, issn=1003-501X, eissn=2097-4019, cn=51-1346/O4, coden=null, periodic=0, language=CN, oaType=null, 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=4Vimkd+qXLWNxtdpr9mFNw==, journalPrice=null, startedYear=null, abbrevIsoEn=Opto-Electronic Engineering, journalRemark=null, publicationField=null, createdTime=1782781457950, updatedTime=1784021784694, createdBy=18614031015, updatedBy=13041195026, firstLetterCn=G, firstLetterEn=G, subjectCode=Engineering, subjectName=null, subjectCodeEn=Engineering, subjectNameEn=null, picCn=4Vimkd+qXLWNxtdpr9mFNw==, picEn=vAq9s20WLs1ODDfbWq+Gjg==, jcr=null, cjcr=null, exts=[JournalExt(id=1283843675721536154, language=CN, name=光电工程, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1784021784954, updatedTime=1784021784954, createdBy=13041195026, updatedBy=13041195026, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.manuscripts.com.cn/gdgc, submissionEditorUrl=http://www.manuscripts.com.cn/gdgc, submissionReviewUrl=http://www.manuscripts.com.cn/gdgc, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1283843675771867803, language=EN, name=Opto-Electronic Engineering, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1784021784966, updatedTime=1784021784966, createdBy=13041195026, updatedBy=13041195026, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.manuscripts.com.cn/gdgc, submissionEditorUrl=http://www.manuscripts.com.cn/gdgc, submissionReviewUrl=http://www.manuscripts.com.cn/gdgc, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1278651732997652489, websiteList=[Website(id=1278723867418018151, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1278651732997652489, 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/oee/CN, language=CN, createTime=1782801127533, createBy=18614031015, updateTime=1782804494442, updateBy=18614031015, name=光电工程-中文, tplId=1146099689490845704, title=光电工程, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1278738091150128034, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=articleTextType, value=kx, createTime=1782804518735, updateTime=1782804518735, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091120767903, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=banner, value=null, createTime=1782804518728, updateTime=1782804518728, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091171099557, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=grayFlag, value=0, createTime=1782804518740, updateTime=1782804518740, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091108184990, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=logo, value=https://castjournals.cast.org.cn/joweb/oee/CN/file/pic?fileId=A1C6uwqtMazluiWkEpR0Mg==, createTime=1782804518725, updateTime=1782804518725, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091179488167, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=minRunFlag, value=0, createTime=1782804518742, updateTime=1782804518742, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091141739425, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/oee/CN/file/pic, createTime=1782804518733, updateTime=1782804518733, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091175293862, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=silenceFlag, value=0, createTime=1782804518741, updateTime=1782804518741, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091129156512, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1782804518730, updateTime=1782804518730, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091154322339, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=themeColor, value=null, createTime=1782804518736, updateTime=1782804518736, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091162710948, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=themeStyle, value=null, createTime=1782804518738, updateTime=1782804518738, creator=18614031015, updator=18614031015)]), Website(id=1278723867522875769, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1278651732997652489, 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/oee/EN, language=EN, createTime=1782801127558, createBy=18614031015, updateTime=1782804490442, updateBy=18614031015, name=光电工程-英文, tplId=1146101810881728533, title=Opto-Electronic Engineering, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1278738063660659607, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=articleTextType, value=kx, createTime=1782804512181, updateTime=1782804512181, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063635493780, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=banner, value=null, createTime=1782804512175, updateTime=1782804512175, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063924900762, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=grayFlag, value=0, createTime=1782804512244, updateTime=1782804512244, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063606133651, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=logo, value=https://castjournals.cast.org.cn/joweb/oee/EN/file/pic?fileId=A1C6uwqtMazluiWkEpR0Mg==, createTime=1782804512168, updateTime=1782804512168, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063937483676, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=minRunFlag, value=0, createTime=1782804512247, updateTime=1782804512247, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063652270998, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/oee/EN/file/pic, createTime=1782804512179, updateTime=1782804512179, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063933289371, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=silenceFlag, value=0, createTime=1782804512246, updateTime=1782804512246, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063639688085, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1782804512176, updateTime=1782804512176, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063664853912, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=themeColor, value=null, createTime=1782804512182, updateTime=1782804512182, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063916512153, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=themeStyle, value=null, createTime=1782804512242, updateTime=1782804512242, creator=18614031015, updator=18614031015)])], journalTitle=光电工程, weixinUrl=null, journalUrl=https://www.oejournal.org/oee, iacademicId=null, status=1, seqNo=null, journalTitleEn=Opto-Electronic Engineering, journalPhotoCn=4Vimkd+qXLWNxtdpr9mFNw==, journalPhotoEn=vAq9s20WLs1ODDfbWq+Gjg==, journalFirstLetter=G, 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=, provinceCode=null, provinceName=null, collectFlag=false, interPubPlatform=, interPubPlatformUrl=null), detailUrlCn=https://castjournals.cast.org.cn/joweb/oee/CN/10.12086/oee.2026.250340, detailUrlEn=https://castjournals.cast.org.cn/joweb/oee/EN/10.12086/oee.2026.250340, pdfUrlCn=https://castjournals.cast.org.cn/joweb/oee/CN/PDF/10.12086/oee.2026.250340, pdfUrlEn=https://castjournals.cast.org.cn/joweb/oee/EN/PDF/10.12086/oee.2026.250340, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1776960000000, fullTextJson=null, articleText=null, reference=null)
收藏切换
多层感知增强与门控对齐融合的BEV目标检测方法
收藏切换
PDF下载
刘鸿恩 1, 2 , 李旭 1, * , 黄晟 1 , 康婷 3
光电工程 | 科研论文 2026,53(4): 250340
收起
收藏切换
光电工程 |科研论文 2026 , 53 (4) : 250340
多层感知增强与门控对齐融合的BEV目标检测方法
全屏
[Author(id=1280951079814808156, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=len3610@126.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1280951079890305631, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951079814808156, language=EN, stringName=Hongen Liu, firstName=Hongen, middleName=null, lastName=Liu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China
2Institute of Permanent-Magnet Maglev Technology and Rail Transit, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1280951079965803104, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951079814808156, language=CN, stringName=刘鸿恩, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000
2江西理工大学永磁磁浮技术与轨道交通研究院,江西 赣州 341000, bio={"img":"bo8nFNNt3aqJ4iyYhCbi6g==","content":"

刘鸿恩(1987-),男,博士,副教授,研究方向为智能交通系统融合感知、自学习特征建模和协同优化控制方法研究。E-mail:

"}, bioImg=bo8nFNNt3aqJ4iyYhCbi6g==, bioContent=

刘鸿恩(1987-),男,博士,副教授,研究方向为智能交通系统融合感知、自学习特征建模和协同优化控制方法研究。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1280951079529595474, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=1, ext=[AuthorCompanyExt(id=1280951079537984083, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079546372692, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000)]), AuthorCompany(id=1280951079621870165, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=2, ext=[AuthorCompanyExt(id=1280951079634453078, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079621870165, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Institute of Permanent-Magnet Maglev Technology and Rail Transit, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079642841687, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079621870165, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2江西理工大学永磁磁浮技术与轨道交通研究院,江西 赣州 341000)])]), Author(id=1280951080053883490, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=885239570@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1280951080125186660, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080053883490, language=EN, stringName=Xu Li, firstName=Xu, middleName=null, lastName=Li, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, *, address=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1280951080204878437, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080053883490, language=CN, stringName=李旭, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, *, address=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000, bio={"img":"4oj0L19gjbZj6+6xQK2NFw==","content":"

李旭(2001-),男,硕士研究生,主要研究方向为机器学习、多模态感知等。E-mail:

"}, bioImg=4oj0L19gjbZj6+6xQK2NFw==, bioContent=

李旭(2001-),男,硕士研究生,主要研究方向为机器学习、多模态感知等。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1280951079529595474, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=1, ext=[AuthorCompanyExt(id=1280951079537984083, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079546372692, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000)])]), Author(id=1280951080292958823, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1280951080364261993, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080292958823, language=EN, stringName=Sheng Huang, firstName=Sheng, middleName=null, lastName=Huang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1280951080435565162, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080292958823, language=CN, stringName=黄晟, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1280951079529595474, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=1, ext=[AuthorCompanyExt(id=1280951079537984083, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079546372692, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079529595474, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000)])]), Author(id=1280951080511062636, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1280951080599143022, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080511062636, language=EN, stringName=Ting Kang, firstName=Ting, middleName=null, lastName=Kang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1280951080670446191, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, authorId=1280951080511062636, language=CN, stringName=康婷, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3江西理工大学电气工程与自动化学院,江西 赣州 341000, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1280951079718339160, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, xref=3, ext=[AuthorCompanyExt(id=1280951079730922073, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079718339160, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China), AuthorCompanyExt(id=1280951079735116378, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1279513094535754456, companyId=1280951079718339160, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3江西理工大学电气工程与自动化学院,江西 赣州 341000)])])]
刘鸿恩1, 2 , 李旭1, * , 黄晟1, 康婷3
作者信息
  • 1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000
  • 2江西理工大学永磁磁浮技术与轨道交通研究院,江西 赣州 341000
  • 3江西理工大学电气工程与自动化学院,江西 赣州 341000
通讯作者:
作者简介:

刘鸿恩(1987-),男,博士,副教授,研究方向为智能交通系统融合感知、自学习特征建模和协同优化控制方法研究。E-mail:

李旭(2001-),男,硕士研究生,主要研究方向为机器学习、多模态感知等。E-mail:

BEV object detection method with multi-layer perceptual enhancement and gated alignment fusion
Hongen Liu1, 2 , Xu Li1, * , Sheng Huang1, Ting Kang3
Affiliations
  • 1Jiangxi Provincial Key Laboratory of Maglev Rail Transit Equipment, Ganzhou, Jiangxi 341000, China
  • 2Institute of Permanent-Magnet Maglev Technology and Rail Transit, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China
  • 3School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China
出版时间: 2026-04-24 doi: 10.12086/oee.2026.250340
文章导航
收藏切换

针对多模态3D目标检测中图像语义受限、点云稀疏致几何退化与跨模态对齐困难等问题,提出一种多层感知增强与门控对齐融合的BEV目标检测方法(MEPFusion)。首先图像端利用SimPVT强化全局语义与关键区域建模;其次点云端设计空间密度调制卷积,以位置依赖权重注入密度先验,稳固稀疏场景的局部几何;最后融合端构建双模态门控对齐模块,通过通道与空间门控抑制错配并实现动态校准,促成BEV特征的动态对齐与互补交互。实验结果表明,本文方法在nuScenes上的mAP与NDS达到67.1%与70.8%,较BEVFusion分别提升2.8%和1.7%,显著提高检测精度与鲁棒性。消融结果验证MEPFusion在语义特征提取、几何建模与跨模态对齐三方面具备独立增益并产生协同效应,整体上在融合质量、检测性能与计算效率之间实现更优平衡,为自动驾驶三维目标检测提供高效可靠的方案。

自动驾驶  /  多模态融合  /  BEV表征  /  多层感知增强  /  门控对齐融合
Objective:

Multimodal three-dimensional object detection combines camera semantics with LiDAR geometry and plays a central role in autonomous driving perception. Yet performance often degrades in complex traffic due to unstable semantics, sparse geometry, and cross-modal misalignment. Camera features become unreliable under glare, low illumination, motion blur, and cluttered backgrounds, introducing uncertainty into bird’s-eye-view (BEV) projection. LiDAR point clouds are sparse at long range or in adverse weather, weakening local geometry and causing boundary blur and missed small objects. In BEV fusion, misalignment between modalities introduces conflicting cues and redundant responses, reducing confidence and accuracy. This study proposes an efficient BEV detector that enhances camera semantics, stabilizes sparse geometry, and achieves adaptive cross-modal fusion for improved accuracy and robustness without excessive computational cost.

Methods:

The proposed framework, MEPFusion (multi-layer perception enhancement and gated alignment fusion), improves camera semantics, stabilizes sparse-geometry modeling, and enables gated alignment fusion in BEV space. For camera encoding, SimPVT enhances global context and salient-region representation with low overhead. A pyramid vision transformer backbone extracts multi-scale features to capture global and local dependencies. A parameter-free saliency reweighting unit amplifies target-related activations and suppresses background distraction, improving semantic separability. To reduce computation, the highest-resolution stage is removed while preserving three-scale outputs for projection. Features are lifted to BEV space via a lift–splat–shoot transformation supervised by depth estimation, stabilizing the depth distribution and reducing projection noise to improve spatial consistency. This camera branch aims to provide more reliable semantics under appearance variations, reducing texture-induced false positives and improving recall for visually small instances.

For LiDAR encoding, raw point clouds are voxelized and processed through a SECOND-style sparse convolution backbone. Standard sparse convolutions are replaced with spatial density-modulated convolution (SDMConv) to address equal-weight aggregation in sparse regions. SDMConv injects a position-dependent density prior into convolution responses via multiplicative modulation, producing a center-enhanced and periphery-decayed pattern within the receptive field. The modulation mitigates feature dilution and preserves boundary cues and shape continuity, especially in long-range or partially occluded areas, without increasing parameter count. The resulting sparse 3D features are compressed along height to produce LiDAR BEV features. In practice, SDMConv acts as a plug-in operator that strengthens local geometry when point evidence is weak, benefiting thin structures and small obstacles that are easily under-sampled.

Cross-modal fusion occurs in BEV space through a gated alignment fusion module (GAFM) designed to suppress misalignment while maintaining complementarity. Each modality undergoes channel recalibration to emphasize discriminative channels. A residual mutual-guidance path performs bidirectional correction: LiDAR features refine camera BEV features in geometry-sensitive regions, while camera semantics complement LiDAR features where geometric evidence is weak. Spatial gates are predicted for each BEV cell and normalized via softmax to assign adaptive fusion weights, allowing the more reliable modality to dominate locally while reducing inconsistent responses. The fused BEV features are decoded by a multi-scale neck and a transformer-based detection head to produce 3D bounding boxes and attributes. This strategy improves optimization stability and encourages reliability-aware weighting rather than uniform mixing.

Results and Discussions:

Extensive experiments on the nuScenes benchmark demonstrate the effectiveness of MEPFusion. With six cameras and one LiDAR input, MEPFusion achieves 67.1% mAP and 70.8% NDS, surpassing the BEVFusion baseline by 2.8% and 1.7%, respectively. Regression errors in translation, scale, orientation, velocity, and attribute are reduced, confirming improved spatial alignment and motion estimation. Category-level analysis shows notable gains for small or geometry-sensitive objects (e.g., traffic cones, barriers, motorcycles, and pedestrians) while maintaining improvements for major vehicle classes. These trends align with module contributions: SimPVT strengthens semantics in cluttered scenes, SDMConv alleviates sparsity-induced degradation, and GAFM reduces cross-modal interference through adaptive gating and complementary fusion. Qualitative inspection suggests fewer duplicated boxes and cleaner spatial responses in crowded regions, consistent with suppressing redundant activations caused by misalignment.

Distance-stratified evaluation further verifies consistent improvements across near, mid, and far ranges. The far range, where point sparsity and scale ambiguity are most severe, still achieves clear gains, highlighting the benefits of density-aware encoding and local gating. Robustness tests under adverse conditions also validate generalization. On the night subset, MEPFusion improves mAP and NDS while reducing error terms, indicating stable semantics and selective fusion under low illumination and glare. On the rain subset, performance again improves, confirming resilience to reduced contrast and weaker LiDAR returns. Ablation studies reveal both independent and synergistic effects: SimPVT alone enhances accuracy with fewer parameters; SDMConv and GAFM each bring additional gains; combining all three yields the best overall performance. Efficiency analysis shows SDMConv introduces minimal overhead, and the full model maintains comparable throughput with reduced total parameters, indicating a favorable accuracy–efficiency trade-off for on-vehicle inference.

Conclusions:

In summary, MEPFusion integrates semantic enhancement, sparse-geometry reinforcement, and gated alignment fusion within a unified BEV framework. The method achieves higher accuracy, improved regression stability, and consistent robustness across distance and environmental variations while retaining deployment-friendly efficiency. Stable semantics before view transformation, density-modulated LiDAR encoding, and adaptive BEV fusion jointly mitigate misalignment and suppress redundant noise. MEPFusion thus offers an effective and reliable multimodal perception solution for urban autonomous driving.

autonomous driving  /  multimodal fusion  /  BEV representation  /  multi-layer perception enhancement  /  gated alignment fusion
刘鸿恩, 李旭, 黄晟, 康婷. 多层感知增强与门控对齐融合的BEV目标检测方法. 光电工程, 2026 , 53 (4) : 250340 - . DOI: 10.12086/oee.2026.250340
Hongen Liu, Xu Li, Sheng Huang, Ting Kang. BEV object detection method with multi-layer perceptual enhancement and gated alignment fusion[J]. Opto-Electronic Engineering, 2026 , 53 (4) : 250340 - . DOI: 10.12086/oee.2026.250340
自动驾驶的规模化应用使环境感知成为衡量整车智能水平的关键指标,其对复杂场景的语义理解与快速响应能力决定行车安全与稳定性[1]。作为底层支撑模块,环境感知的重要性持续提升,并对精度、效率与实时性的协同优化提出更高要求。三维目标检测是感知系统的关键环节,除承担类别识别外,还需准确估计目标的尺寸、位置、姿态与速度等关键状态量,为轨迹预测、路径规划与紧急避障提供可信输入。在真实道路环境中,远距小目标、强遮挡、复杂气象与多源噪声交织出现,单一传感器难以长期保持稳定性能[2]。近年来,纯视觉方案在目标检测任务中取得了显著进展,在多个自动驾驶公开数据集上实现了接近最优的性能。然而,纯视觉方案尽管能够提供丰富的语义信息,但在光照变化、空间结构感知和深度估计方面仍然存在较大挑战。相比之下,激光雷达(LiDAR)能够提供高精度的几何数据与深度信息,但在远距离与雨雾等工况下回波稀疏[3]。然而,单一传感器的局限性在遮挡和低可见度的情况下愈发明显。由此引发的语义表达受限与几何建模缺陷在远距离与遮挡等条件下进一步放大,成为制约整体检测表现的关键瓶颈[4]。为突破上述限制,结合了激光雷达和相机的多模态融合系统被广泛认为是提升自动驾驶感知质量的有效路径,能够更有效地融合几何信息,从而在处理远距离、小目标、遮挡及恶劣天气等场景时,提供不可替代的安全性优势。其核心是在统一的时空与表征基准上构建可学习的互补推理机制[5]。该范式在尺度锚定、稠密语义补全与冗余互证环节产生协同增益,从而显著降低远距与遮挡场景的不确定性,并增强在复杂工况下的鲁棒性。为将上述协同转化为工程可部署方案,亟须一种统一且计算友好的交互载体。鸟瞰图表示(Bird's-eye view, BEV)作为平面化的三维空间表征,通过视图变换将多视角图像与点云对齐至同一地面坐标系,在二维计算预算内保留关键三维几何关系,弱化跨视角配准误差,并为后续的特征对齐与融合提供统一空间基准,从而有助于提升远距小目标与复杂遮挡场景的召回与定位精度[6]
围绕BEV空间的点云与图像的相关算法,现有研究依输入模态大体可分为三类[1]。基于BEV的视觉三维检测,典型流程通常遵循“视图变换→BEV空间聚合→BEV头检测”。其中,LSS[7]通过为每个像素预测离散深度分布,将二维特征沿光线方向反投影到三维,并在地面坐标系中进行体素池化或直接splat至BEV网格。代表性方法如BEVDet[8]以高效视图变换将多视角特征映射到BEV,兼顾速度与精度;BEVDepth[9]在训练中加入显式深度监督,并配合体素池化等设计,缓解由尺度与距离不确定引起的误差传播。此外,BEVFormer[10]将BEV查询与图像平面对齐并融合时序Transformer,以提升动态场景建模能力。然而,由于相机缺乏真实深度且易受照明、遮挡与纹理重复影响,纯视觉BEV在远距小目标与强遮挡工况下仍容易出现深度分配不稳、框体尺度偏差与虚警增多的问题。
与此相对,基于LiDAR的BEV检测以几何精度为基础,通过体素或柱体化将不规则点集转化为规则网格,再在BEV或体素上进行特征提取与目标回归[11]。PointPillars[12]以柱状体素和二维卷积实现实时检测;SECOND[13]通过稀疏卷积在计算代价与几何完整性之间取得平衡;CenterPoint[14]以中心点回归替代锚框设计,兼顾精度与后处理效率。尽管LiDAR在几何上更可靠,但在远距、雨雾与稀疏回波情况下点云密度骤降,卷积核“等权聚合”的倾向会弱化细窄结构与物体边界,导致小目标漏检与姿态估计不稳;体素尺寸的缩小虽有利于细节保留,却显著增加内存与算力,带来部署约束。
在单模态路径逼近瓶颈后,多模态BEV融合成为进一步突破方向。其基本思想是在统一的BEV空间内将视觉的语义与LiDAR的几何进行互补耦合,以在维持推理效率的同时提升性能上限。按融合时机与粒度划分,早期策略在投影前于图像平面或点级别交互,信息充分但对标定与时序同步更敏感;后期策略在各自BEV表征形成后再交互,工程稳定但对细粒度错配的纠正能力有限;中期策略则在多尺度间反复交互,力求兼顾语义与几何一致性[15]。代表工作方面,TransFusion[16]在LiDAR-BEV上引入基于注意力的软关联机制,以自适应汇聚图像分支信息,从而减轻模态分布差异导致的“生硬拼接”;BEVFusion[17]将图像经视图变换与LiDAR特征同时统一到BEV,再端到端融合,在精度与效率之间取得良好平衡;MVP[18]、AutoAlignV2[19]、UVTR[20]等方法分别从显式几何约束、可学习对齐与Transformer深度交互角度强化跨模态协同。尽管体系日趋成熟,多模态BEV感知仍普遍面临:视图变换量化误差与视觉深度不确定共同引发的语义——几何错位;点云稀疏性与视角覆盖差异导致远距与遮挡处边界线索被弱化,诱发小目标漏检与尺度偏差;跨模态噪声与冗余在融合阶段被放大;且缺乏逐位置、逐通道的选择性机制时,全局交互难以兼顾局部几何约束与计算预算,限制分辨率、时延与鲁棒性的协同优化空间。
基于上述分析,本文提出面向自动驾驶的多层感知增强与门控对齐融合的BEV目标检测算法(multilayer perception enhancement and gated alignment fusion, MEPFusion),围绕语义表征增强、稀疏几何建模与跨模态对齐三条主线协同推进。在图像分支,构建SimPVT强化全局与多尺度语义一致性并放大关键区域响应,为后续视图变换提供更稳定的语义先验;在点云分支,提出空间密度调制卷积,在不增加参数量的前提下对卷积核实施位置相关的密度调制,强化远距与稀疏场景中的边界与细节表达;在融合阶段,设计门控对齐融合模块,于BEV空间学习逐位置与逐通道的门控与软对齐,动态突出优势模态、抑制冗余互扰,从而在复杂交通场景下实现高精度、低时延与强鲁棒的三维感知。在与基线等参数与等时延配置下的系统对比与消融验证中,所提方案在mAP/NDS指标上获得稳定增益,并在视觉语义、几何建模与跨模态对齐三个层面呈现相互支撑的协同效应,整体体现出在多场景条件下的精度提升与鲁棒性增强。
本文的主要贡献概括如下:
1)语义全局增强。在图像分支构建SimPVT,强化全局多尺度语义一致性并放大关键区域响应,稳固视图变换前的语义先验。
2)几何建模强化。在点云分支提出空间密度调制卷积,以位置相关的空间密度调制替代等权聚合,在不增加参数量的前提下提升稀疏、远距场景的几何边界与细节建模能力。
3)融合对齐机制。设计双模态门控对齐融合模块,于BEV空间学习逐位置与逐通道的门控与软对齐,动态突出优势模态、抑制冗余互扰,实现高效稳健的跨模态互补。
4)性能与实证。在等参数与等时延条件下取得稳定性能增益(mAP/NDS均提升),并通过系统消融验证各模块的独立贡献与协同效应,兼顾精度、效率与可部署性。
在多模态三维感知任务中,图像与点云分别在语义表达与几何建模方面具有互补优势,但两者在特征分布、空间组织及时间采样频率上的差异,常导致语义特征提取不足、几何结构刻画不精确以及跨模态对齐偏差等问题。为此,本文提出一种基于BEVFusion结构优化的多层感知增强与门控对齐融合的BEV目标检测算法MEPFusion,从语义建模、几何表达与多模态融合三方面对原框架进行协同增强。本文MEPFusion方法的总体架构如图1所示。
本文MEPFusion方法采用相机&LiDAR双分支结构,并在BEV空间实现可学习对齐与融合。图像分支以PVTv2[21]为金字塔主干,结合SimAM[22]的显著性重标定获取多尺度且判别性更强的语义特征;随后结合显式深度监督与LSS视图变换,将二维特征反投影并聚合为统一BEV表示,实现由图像平面到鸟瞰坐标的几何对齐。点云分支在体素编码阶段利用空间密度调制卷积,依据局部点密度对卷积响应进行位置相关调制,以强化稀疏与远距场景下的结构建模,并通过BEVPooling沿高度维聚合将体素特征压缩至BEV网格。两路BEV特征输入门控对齐融合模块,以逐位置门控与软对齐实现动态加权与互补交互,得到统一的多模态BEV表征;最终经FPN与TransFusion检测头进行多尺度解码,完成高精度三维目标预测[23]
在算力受限下,图像分支旨在生成跨视角一致且对目标边界更敏感的语义表征,从而降低视图变换后的深度不确定性,提升BEV投影的稳定性。为此,本文构建兼顾全局建模与局部显著性的特征提取网络SimPVT (图2),采用PVTv2全局建模和SimAM局部显著性重标定的两段式结构。其中,PVTv2作为金字塔式主干,在多尺度空间内高效刻画长程依赖实现全局表征学习。同时,在各尺度输出端嵌入轻量无参注意力SimAM,以元素级显著性重标定放大关键区域响应并抑制背景噪声,从而形成兼具全局一致性与局部判别力的输入特征。该分支在不显著增加参数与计算开销下,能够放大关键区域响应、抑制背景噪声,并提升跨视角语义分布稳定性。
1) PVTv2全局建模。首先输入经重叠式Patch Embedding得到初始特征,随后进入四阶段Transformer编码逐级构建由高分辨率到低分辨率的金字塔特征{F1, F2, F3, F4}。鉴于后续结构仅需三尺度输入,本工作舍弃最高分辨率的F1,仅保留{F2, F3, F4} 作为多尺度特征集,以降低高分辨率分支带来的额外计算开销,并抑制低层纹理细节对高层语义表征的干扰。其中,各Transformer编码阶段通过LSRA (linear spatial-reduction attention),先对Key、Value的空间维度进行线性压缩以缩短序列,再执行缩放点积注意力完成全局依赖聚合,最后在理论复杂度上由$ L^2 $缩减为$ L^2/r^2 $,其中$ L=H\times W $表示输入特征展平后的序列长度,$ r $表示空间压缩比,从而在长程依赖建模与计算开销之间取得平衡。LSRA的正式定义为:
$ {LS RA}\left(x\right)={Concat}\left({{\boldsymbol{h}}}_{0},\cdots ,{{\boldsymbol{h}}}_{N-1}\right){{\boldsymbol{W}}}^{{\mathrm{o}}} \;, $
$ {{\boldsymbol{h}}}_{j}={Att}(x{\boldsymbol{W}}_{j}^{Q},{S}(x){\boldsymbol{W}}_{j}^{K},{S}(x){\boldsymbol{W}}_{j}^{V}),j\in \left\{0,\cdots ,N-1\}\right.\;, $
$ {S}(x)={LN}({Avg}(x,p)\cdot {{\boldsymbol{W}}}^{S})\;, $
式中:$ {{x}}\in {\mathbb{R}}^{H\times W\times {{C}_{i}}} $表示输入特征,$ H $$ W $$ C $分别表示特征图的高、宽和通道数;$ {{\boldsymbol{h}}}_{j} $是第j个注意力头的输出;N为头的数量;Concat(·)表示通道级联;$ {{\boldsymbol{W}}}^{\rm{o}}\in {\mathbb{R}}^{{{C}_{\rm{o}}}\times {{C}_{\rm{o}}}} $是将级联特征映射到输出维度$ {C}_{\rm{o}} $的输出投影矩阵;矩阵$ {\boldsymbol{W}}_{j}^{Q}、{\boldsymbol{W}}_{j}^{K}、{\boldsymbol{W}}_{j}^{V}\in {\mathbb{R}}^{{{C}_{i}}\times d} $是第j个头的查询、键和值投影,$ d={C}_{\rm{o}}/N $作为每个头通道维度;Att(·)表示缩放的点积注意力;S($x $)表示输入x的空间简化表示;Avg($x $, p)表示$x $上具有$ p\times p $窗口的平均池化;$ {\boldsymbol{W}}^{S}\in {\mathbb{R}}^{{{C}_{i}}\times {{C}_{i}}} $为信道投影矩阵;LN(·)为层归一化。
2) SimAM局部重标定。在获得的多尺度特征{F2, F3, F4}基础上,逐尺度执行无参的SimAM重标定。首先依据通道内均值与方差构造能量函数$ e_{\min }^{t} $,度量单个位置与同通道其余位置的可分性;接着将由能量函数得到的响应值${\boldsymbol{W}} $经Sigmoid映射为权重张量$ \tilde{{\boldsymbol{W}}} $,最后与原特征逐元素相乘完成显著性增强。该过程零参数、仅含元素级运算,能够增强关键区域与远距小目标响应并抑制背景噪声,提升跨视角语义稳健性。过程数学公式如下。
$ e_{\min }^{t}=\dfrac{4({\tilde{\sigma }}^{2}+\lambda )}{{(t-{{\mu }_{t}})}^{2}+2\tilde{\sigma }_{t}^{2}+2\lambda }\;, $
$ \tilde{{\boldsymbol{W}}}=S igmoid\left(\dfrac{1}{{\boldsymbol{E}}}\right)\odot {\boldsymbol{W}} \;, $
式中:$ {\mu }_{t} $$ \tilde{\sigma }_{t}^{2} $λ分别表示在排除目标神经元后的均值、方差和数值稳定常数;t为神经元;E为所有神经元能量值组合的张量。
鉴于LiDAR点云在空间分布上高度不均且整体稀疏,标准3×3稀疏卷积对核内各相对位置等权聚合,易在远距稀疏、遮挡与噪声条件下产生“均值化”效应,进而削弱窄带边界与线状几何细节的判别性。为此,本文在点云分支设计FLN (feature learning network)→SECOND-SDM的二级路径,其结构如图3所示。其中,D×H×W 表示体素化后特征张量的空间尺寸,D表示竖直方向的离散层数,HW分别表示BEV平面两个方向上的离散网格数。
具体而言,首先将原始点云按规则网格进行体素化,并在每个体素内通过随机采样获得定长点集,随后经Voxel Encoding和Stacked VFE将点级属性逐层聚合与压缩,构建稀疏四维体素张量并输入SECOND主干。其中,主干由三阶段SECOND-SDM顺序组成,各阶段以空间密度调制卷积(spatial density-modulated convolution, SDMConv)为核心,通过位置密度调制抑制等权聚合导致的信息稀释,逐级提炼特征并维持合理的尺度组织;阶段之间通过stride=2的稀疏卷积完成尺度切换与感受野扩展[24]。最终,Sparse2Dense沿高度维聚合得到BEV平面表征并输入融合模块。该路径在网络深度、通道配置与下采样倍率上保持稳定层级,确保从点级信息到BEV表示的高效映射与可复现实验流程。
在此总体路径中,SDMConv作为默认的3×3空间算子贯穿主干残差/堆叠块:前向阶段预置固定、中心对称、秩为1的位置先验$ \varPhi $,并以Hadamard逐元素方式对卷积核实施位置依赖的权重调制,形成“中心增强—外围递减”的响应分布。由此,在保持计算可比性的同时,有效抑制远距稀疏与遮挡噪声引发的等权“均值化”,显著强化BEV平面上细窄边界与局部几何回波的保真度,提升几何判别性、小目标可分性与整体检测稳健性。SDMConv数学公式如下。
$ ({\boldsymbol{I}}\ast {\boldsymbol{W}}_{\phi })_{ij}^{{f}}=\displaystyle\sum\limits_{a=1}^{K}\displaystyle\sum\limits_{b=1}^{K}{\varPhi }_{a,b}\cdot \omega _{a,b}^{{f}}\cdot {{\boldsymbol{I}}}_{i+a-{{K}_{a}}+1,j+b-{{K}_{{b}}}+1}\;, $
式中:$ {\boldsymbol{I}}\in {\mathbb{R}}^{{R}\times C} $为输入信号;$ (i,j) $为网格坐标;$ {\boldsymbol{W}}_{\phi }\in {\mathbb{R}}^{{{K}_{a}}\times {{K}_{b}}\times F} $为卷积核张量;$ \omega _{a,b}^{{f}} $为第f个核在相对位置$ (a,b) $处的权重;离散密度函数$ \varPhi \in {\mathbb{R}}^{{{K}_{a}}\times {{K}_{b}}} $为空间位置先验;元素$ {\varPhi }_{ab} $为调制核内位置$ (a,b) $贡献。
在多模态3D目标检测中,尽管将图像与激光雷达特征统一至BEV空间有助于发挥两者在语义理解与几何建模方面的互补优势,但两种模态在信息来源、感知维度与采样密度上的差异往往导致BEV特征语义不一致与空间错位,简单拼接或均值融合不仅难以缓解这一矛盾,反而易将噪声引入最终表征,从而削弱检测性能[25]。为此,本文设计门控对齐融合模块(gated alignment fusion module, GAFM),通过轻量化空间注意力机制在BEV网格逐位置学习空间相关的门控权重,使关键区域突出、冗余区域抑制,从而提高融合表征的一致性与判别性[26]。具体而言,本文在BEV网格位置$ (x,y) $上将对检测贡献更显著且更稳定的局部响应区域定义为关键区域,其特征为该位置至少存在一种模态提供相对可靠的语义或几何证据;而背景噪声占优、有效响应较弱,或两模态在该位置出现不一致表征的区域定义为冗余区域。GAFM在BEV网格上为两模态学习逐位置的空间门控权重,以选择性加权的方式进行融合,使空间上呈现“优势模态主导、弱证据抑制”的聚合特性,从而避免简单拼接将冗余信息一并引入检测头。
图4所示,GAFM采用分离-聚合的总体范式,先完成模态内显著性重标定,再进行跨模态互导校正,最后实施空间门控聚合操作。具体而言,先在各自模态内部做针对性增强,使图像侧更聚焦全局语义与类别判别、点云侧更突出几何边界与局部结构;继而在两模态之间进行轻量互导以削弱系统性偏差并互补细节;最后在模态维实施可解释的门控加权,输出语义完整且空间一致的融合BEV特征,用作检测头的高质量输入。在此过程中,首先,BEV图像特征$ {\boldsymbol{T}}_{\rm{p}} $和点云特征$ {\boldsymbol{T}}_{\rm{c}} $分别经通道注意力强化全局上下文信息表达,得到增强后特征$ {\boldsymbol{T}}_{\text{p}}^{\text{f}} $$ {\boldsymbol{T}}_{\text{c}}^{\text{f}} $。该过程本质上是通道维的显著性重标定,增强与类别语义判别及目标整体轮廓和结构表征相关的通道响应,同时抑制背景噪声与点云稀疏性导致的激活不稳定,从而在模态内先行净化特征表示,为后续跨模态互导校正与空间门控聚合提供更具判别性的输入。其数学表达式分别为
$ {\boldsymbol{T}}_{\rm{p}}^{\rm{f}}=\sigma ({\boldsymbol{W}}_{2}\cdot {Re}LU({\boldsymbol{W}}_{1}\cdot Gap(Concat({\boldsymbol{T}}_{{\mathrm{p}}},{\boldsymbol{T}}_{\rm{c}}))))\cdot {\boldsymbol{T}}_{\rm{p}} \;, $
$ {\boldsymbol{T}}_{\rm{c}}^{\rm{f}}=\sigma ({\boldsymbol{W}}_{2}\cdot {Re}LU({\boldsymbol{W}}_{1}\cdot Gap(Concat({\boldsymbol{T}}_{{\mathrm{p}}},{\boldsymbol{T}}_{\rm{c}}))))\cdot {\boldsymbol{T}}_{\rm{c}} \;, $
式中:$\sigma $表示Sigmoid激活函数;$ G{ap}(\cdot ) $表示全局平均池化操作;$ {\boldsymbol{W}}_{1} $$ {\boldsymbol{W}}_{2} $为两层全连接层的权重,用于生成通道加权向量。
其次,为增强模态间信息协同而引入残差校正策略,构建点云对图像的引导关系,对图像特征进行自适应校准以有效去除干扰信息,获得优化后的图像特征表示$ {\boldsymbol{T}}_{\rm{p}}^{{\mathrm{r}}} $。同时,利用图像丰富的语义与细节信息对点云特征进行补充,增强其空间结构表达能力,得到互补点云特征$ {\boldsymbol{T}}_{\rm{c}}^{{\mathrm{r}}} $,进一步提升融合特征的表达能力。需要强调的是,残差校正除跨模态互补外,还承担局部消噪与一致性约束:在错配位置削弱不可靠分量、保留一致证据,从而提升关键区域的有效响应并抑制冗余干扰。其数学表达式分别为
$ {\boldsymbol{T}}_{\rm{p}}^{{\mathrm{r}}}={\boldsymbol{T}}_{\rm{c}}^{\rm{f}}+{\boldsymbol{T}}_{{\mathrm{p}}} \;, $
$ {\boldsymbol{T}}_{\rm{c}}^{{\mathrm{r}}}={\boldsymbol{T}}_{\rm{p}}^{\rm{f}}+{\boldsymbol{T}}_{\rm{c}} \;, $
然后,通过1×1卷积计算每种模态特征的空间注意力权重,并利用Softmax函数归一化空间权重,确保在每个BEV位置上融合权重的动态分配,从而在目标证据显著处由更可靠模态主导,而背景或冲突区域的贡献被自动压低,从而避免噪声放大并提升融合判别性。最后,依据注意力权重对校准后的图像与点云特征进行加权求和,生成语义完整、空间一致的融合特征表示$ {\boldsymbol{T}}_{{\mathrm{output}}} $,实现关键区域突出与冗余区域抑制,为后续目标检测提供高质量的BEV融合特征输入。其数学表达式分别为
$ {\boldsymbol{W}}_{\mathrm{p}}^{\rm{spatial}}={S oftmax}\left({Conv1}\times \text{1}\left({Concat}\left({\boldsymbol{T}}_{\rm{c}}^{\rm{f}},{\boldsymbol{T}}_{\rm{p}}^{\rm{f}}\right)\right)\right) \;, $
$ {\boldsymbol{W}}_{\rm{c}}^{{{\mathrm{spatial}}}}={{{S oftmax}}}\left({Conv1}\times \text{1}\left({Concat}\left({\boldsymbol{T}}_{\rm{c}}^{\rm{f}},{\boldsymbol{T}}_{\rm{p}}^{\rm{f}}\right)\right)\right) \;, $
$ {\boldsymbol{T}}_{\text{output}}={\boldsymbol{W}}_{\mathrm{p}}^{\rm{spatial}}\cdot {\boldsymbol{T}}_{\rm{p}}^{\rm{f}}+{\boldsymbol{W}}_{\rm{c}}^{\text{spatial}}\cdot {\boldsymbol{T}}_{\rm{p}}^{\rm{f}} \;, $
式中:$ {\boldsymbol{W}}_{\text{p}}^{\text{spatial}} $$ {\boldsymbol{W}}_{\rm{c}}^{\text{spatial}} $为动态空间加权权重;$ Conv1\times 1 $为1×1卷积层操作。
本文基于大规模多模态自动驾驶数据集nuScenes开展实验。该数据集包含约1000个真实城市驾驶场景,配备6个环视相机获取环境纹理信息,1个32线LiDAR感知环境深度信息,以及5个毫米波雷达提供速度与运动特征信息。同时提供23类精细目标标注与高清地图,覆盖多样交通参与体与复杂环境,是自动驾驶领域最具代表性的大规模多模态3D感知数据集之一,为本文方法提供了有效性验证。本文仅使用环视相机与LiDAR两种模态数据进行训练与评估,严格遵循官方训练集和验证集划分与评测协议。采用nuScenes官方指标体系,使用平均精度均值(mean average precision, mAP)和nuScenes检测分数(nuScenes detection score, NDS)作为主要评价指标。评价指标的定义如下:
1) mAP衡量预测结果在不同类别上的整体检测性能,其定义为所有类别平均精度的均值,值越高表示模型检测效果越优,公式如下:
$ mAP=\dfrac{1}{C}\displaystyle\sum\limits_{c=1}^{C}\displaystyle\int\limits_{0}^{1}{P}_{\rm{c}}(r){\mathrm{d}}r \;, $
式中:C表示类别总数;$ {P}_{\rm{c}}(r) $表示在召回率r下的精确率。
2) NDS在mAP的基础上进一步综合预测目标检测分数,数值越高表示模型整体检测性能越优公式如下:
$ NDS=\dfrac{1}{10}\left(5\cdot mAP+\displaystyle\sum\limits_{{m}\in M}(1-\min (1,m))\right) \;, $
式中:$ M=\left\{mAT E,mAS E,mAOE,mAVE,mAAE\right\} $,其中每个误差度量限制在[0,1]。具体而言,mATE为平均位移误差、mASE为平均尺度误差、mAOE为平均方向误差、mAVE为平均速度误差和mAAE为平均属性误差。
本文基于Ubuntu 22.04 LTS环境开展仿真验证实验,使用的软件栈主要包括Python 3.8、PyTorch 1.10.0、CUDA 11.3与cuDNN 8.6,并基于MMDetection3D 1.3.0实现模型训练与推理。硬件平台采用Intel至强(Xeon) Platinum 8336C处理器(32核,主频2.3~3.4 GHz)与NVIDIA RTX 4090D×2,为大规模多模态数据处理与并行训练提供算力保障。
训练策略方面,选用AdamW优化器,初始学习率设为2.5×10−5,权重衰减0.01,配合余弦退火策略在训练过程中自适应降低学习率,以提升收敛稳定性与泛化能力。整体训练历时26个epoch,其中前20个epoch以LiDAR分支为主进行预训练,用于稳固几何表征;随后6个epoch进行多模态联合训练,在保持LiDAR表征稳定的基础上促进跨模态特征对齐与融合性能提升。为保证结果可复现,训练过程固定随机种子,并采用常规梯度稳定化配置。
网络与超参数设置如下:点云体素化分辨率为(0.075 m, 0.075 m, 0.2 m),处理空间范围设为x∈[−54, 54]m,y∈[−54, 54] m,z∈[−5, 3] m,覆盖主要可驾驶区域并兼顾远距目标。数据增强策略与BEVFusion基线保持一致,点云进行整体随机缩放、旋转、平移与点顺序打乱,以降低采样偏差并增强姿态鲁棒性;图像采用旋转与翻转等几何增强。所有几何变换均通过显式记录并传递对应的内外参与同构变换矩阵,确保经由LSS视图转换映射后的图像特征与点云BEV表征在空间上严格对齐。
为在稀疏远距与类别不均衡条件下同时保证中心定位稳健性和几何回归精度,本文在TransFusion框架下构建以中心热力图和多分支回归为核心的检测损失L。训练时,先将三维标注投影到BEV平面生成类别中心热力图H,并把每个真值框编码为回归向量$ (\text{Δ}x,\;\text{Δ}y,\;\text{Δ}z,\;w,\;l,\;h,\;\theta ,\;{v}_{x},\;{v}_{y}) $。随后在LiDAR坐标系下对预测查询与真值执行Hungarian一对一匹配,联合分类代价(focal loss cost)、BEV几何L1代价(BBox BEV-L1 Cost)与三维IoU代价(IoU3D Cost)确定正负样本。得到配对后,在BEV网格上以Gaussian Focal Loss计算中心热力图损失$ {\lambda }_{\text{hm}} $,对全部查询以Focal Loss计算分类损失$ {\lambda }_{\text{cls}} $,对匹配到的查询以加权L1计算回归损失$ {\lambda }_{\text{box}} $(对速度通道赋予较小权重以抑制低速噪声);三项分别按有效样本数归一化,并以$ {\lambda }_{\text{hm}}、{\lambda }_{\text{cls}}、{\lambda }_{\text{box}} $加权求和得到总损失L。最终以L反向传播,使“中心命中”与“框体质量”(位置、尺度、朝向与速度)协同优化,从而提升mAP与NDS。
总体损失定义为
$ L={\lambda }_{{\mathrm{hm}}}{L}_{{\mathrm{hm}}}+{\lambda }_{{\mathrm{cls}}}{L}_{{\mathrm{cls}}}+{\lambda }_{{\mathrm{box}}}{L}_{{\mathrm{box}}} \;, $
式中:$ {L}_{\text{hm}} $是BEV中心热力图损失函数;$ {L}_{\text{cls}} $为查询分类损失函数,用于解码查询的类别判别;$ {L}_{{\mathrm{box}}} $为几何与运动回归损失函数;$ {\lambda }_{\text{hm}},{\lambda }_{\text{cls}},{\lambda }_{\text{box}} $为三项损失的权重系数。并对各通道施加权重$ [1,1,1,1,1,1,1,1,0.2,0.2] $ ($ {\lambda }_{\text{box}}=0.25 $),以降低速度项的不确定性干扰。正负样本一一匹配由Hungarian Assigner 3D在LiDAR坐标系下完成,综合Focal Loss Cost、BBox BEV-L1 Cost与IoU3D Cost进行联合度量,从而在语义置信、BEV平面几何与体积重叠度之间取得平衡,确保监督信号稳定且与几何一致。
为系统评估所提方法的有效性,实验遵循nuScenes官方评价体系,报告总体指标与细粒度指标两部分结果。首先,在本地实现BEVFusion基线复现实验,其NDS与mAP分别为69.1%与64.3%。在相同的本地硬件与训练配置下,本文方法mAP提升2.8%、NDS提升1.7% (见表1表2)。表明本文MEPFusion方法在检测精度与综合感知能力在同等算力预算下均有实质提升。其中“C.V.”、“T.L.”、“M.T.”、“Ped.”和“T.C.”分别代表施工车辆、拖车、摩托车、行人和交通锥。
表2展示各目标类别的检测精度,本文MEPFusion方法在小目标与复杂几何目标上取得稳定增益。小体积高语义类别方面,Pde.、T.C.、Barrier与M.T.分别提升1.0%、5.7%、3.0%与6.5%,表明在多目标密集且局部遮挡并存的路口场景中,行人轮廓与交通锥边界的响应更集中、漏检明显减少。大型交通工具方面,虽Trailer略低于基线 (−0.4%),但Car、Truck与Bus分别提升1.7%、3.7%与2.2%,体现出在复杂交通流与视角变化下对大型车辆的更强稳健性,并显著抑制基线中的漏检与框漂移。上述增益源于多模态协同带来的对小尺寸、细长几何与远距稀疏回波上的综合改善。SimPVT在金字塔多尺度建模基础上引入显著性重标定,提升小体积目标的语义判别与边界响应,使行人与交通锥在密集与局部遮挡场景下更易被凸显;SDMConv依据点云密度实施位置加权,强化稀疏远距与细窄结构的几何连续性;GAFM在BEV空间完成跨模态动态对齐与门控融合,在遮挡条带或远距回波弱时优先几何证据、在纹理混淆时提高语义权重,抑制框漂移与漏检,在大型交通工具上取得稳定收益。
三者协同形成功能分工,最终在小目标、远距与遮挡场景下同时提升检出与定位精度,基线存在的漏检与框漂移在本方法中明显缓解。
为更全面评估目标检测的回归质量,除mAP与NDS外,本文依据nuScenes官方评测协议进一步报告五类误差:平移误差mATE (m)、尺度误差mASE (1−IoU)、朝向误差mAOE (rad)、速度误差mAVE (m/s)与属性误差mAAE (1−acc)。与基线方法的细粒度指标对比见表2:MEPFusion在上述误差项上均呈现改善,其中对跨模态对齐更为敏感的朝向与速度误差下降更为显著,mAOE与mAVE分别降低1.4%和1.2%。这表明所提出的选择性融合与互导校正能够有效削弱错配与冗余干扰对回归分支的影响,从而提升融合表示的空间一致性与回归稳定性;此外,mATE、mASE与mAAE的小幅下降亦说明定位、尺度与属性估计得到稳步优化。
同时,为系统检验MEPFusion在远距小目标场景下的检测增益,本文在nuScenes验证集上补充了基于距离分段的细粒度评测结果。表3将评测范围划分为0~20 m (近距)、20~30 m (中距)与30~50 m (远距)三个区间。结果表明,MEPFusion在各距离段均取得一致提升,且在更具挑战的30~50 m远距段仍能获得稳定增益,mAP/NDS分别提升1.4%和0.8%。鉴于远距场景更易受到目标尺度缩小、遮挡累积以及点云回波稀疏等因素影响,基线性能在该区间显著退化,而MEPFusion仍能保持稳定改进,说明所提方法在远距离条件下具备更强的目标可检出性与综合感知质量。
此外,为定位增益来源并聚焦“远距小目标”这一关键情形,表4给出分距离段类别AP,并选取Pde.、M.T.作为典型小型障碍物代表,Car、Truck作为大型目标代表进行对比。结果显示,在远距区间内,小型障碍物的增益更为显著,M.T.的AP由26.2%提升至30.4%(+4.2%),Pde.也有小幅提升;同时,大型目标亦保持稳定提升(Car+1.2%,Truck+2.6%)。
上述结果表明,MEPFusion的性能增益并非主要来源于近距易检出样本的累积贡献,而是在远距段面对更具挑战性的小型障碍物时,仍能显著降低漏检并提升判别精度,验证了本文方法在远距小目标检测方面的有效性与稳健性。该优势源于多模态表征与对齐的协同:SimPVT增强远距小目标语义与边界响应,SDMConv提升稀疏回波下的几何连续性,GAFM通过动态对齐与门控融合抑制远距错位,从而获得更稳健的跨模态互补增益。
进一步,将本文方法与不同技术路径的代表性三维目标检测方法进行对比评估。表5汇总了相机(C)、激光雷达(L)与多模态(C+L)方法在nuScenes的公开结果(均引自原论文或官方基准,含Validation与Test),用于量级参照。与相机组对比,本文MEPFusion方法在Test集达到67.1% mAP和70.8% NDS,相对该组最优的BEVDepth分别领先16.8%和10.8%;在Validation集达到67.1% mAP和70.9% NDS,相对该组公开最高的BEVFormer分别领先25.5%和19.2%。这表明在仅依赖视觉时常见的尺度与深度不确定、遮挡敏感等问题,经由跨模态补偿后得到根本性缓解。与激光雷达组对比,本文MEPFusion方法在Test集,MEPFusion相比TransFusion-L仍有1.6% mAP和0.6% NDS的增益;在Validation集,相比TransFusion-L,本文方法领先了2.0% mAP和0.8% NDS。说明在保持高质量几何先验可达水准的同时,通过更稳健的语义、几何互证与对齐机制,MEPFusion在方向与动态估计上进一步压缩误差。与多模态组对比,MEPFusion在与BEVFusion保持相同算力与实现设置下,于Test/Validation分别实现+2.8%/+1.9% mAP与+1.7%/+1.2% NDS的提升,并在Validation取得全表最高NDS (70.9%)。同时,相比GraphAlign、LIFT、MVP等近期方法,Test集在mAP上分别领先+0.6%、+2.0%、+0.7%,NDS亦小幅占优;在Validation集与MVP的mAP并列第一(67.1%),而NDS居首位(70.9%)。上述结果说明,在不增加计算预算的前提下,跨模态对齐与信息互补被更有效地发挥。
综上,本文方法在精度、效率与鲁棒性之间实现更优平衡,体现出面向复杂交通场景的工程可用性与推广价值。在不增加计算预算的前提下,本文MEPFusion方法相较基线不仅在总体指标与关键类别、关键动态量上实现协同提升。并且,三项改进SimPVT强语义判别、SDMConv稀疏几何增强与GAFM跨模态动态对齐分工明确且相互补偿,有效缓解小目标语义弱、稀疏导致的几何退化、跨模态对齐偏差的问题,使系统在遮挡、远距与复杂背景下保持更高的检测稳定性与工程可用性。
为系统评估所提出的独立增益与协同效应,本文在nuScenes上开展控制变量的消融实验。为避免外部干扰,除被替换模块外,其余训练策略、数据增强、视图变换与检测头均与基线BEVFusion保持一致。实验结果如表6所示。基线在测试集上取得64.3% mAP、69.1% NDS、6.8 f/s、39.80 M参数,可以得出,SimPVT、SDMConv与GAFM在不同组合下均能稳定带来性能增益,并在三者联合时取得最优。
首先,从单模块出发,仅替换图像主干并引入无参注意力的E1 (SimPVT)即可将mAP/NDS提升至65.5%/69.7%,提升1.2%和0.6%,同时模型参数量下降至37.03 M。需要强调的是,参数量的下降主要归因于SimPVT主干自身的轻量化特性,表明更高效的全局建模与显著性重标定能够在几乎不改变实时性水平的前提下有效释放视觉分支潜力;其次,转向双模块组合,可以观察到三条互补路径:其一,E2 (SDMConv+GAFM)在不改动图像分支的前提下,将mAP/NDS提升至64.6%/69.5%,并将参数降至39.08 M,说明密度调制卷积对稀疏点云的几何刻画与门控对齐对跨模态稳定融合均具备正向作用;其二,E3 (SimPVT+GAFM)的mAP/NDS达到65.4%/69.8%,提升1.1%和0.7%,参数进一步降至36.31 M,显示语义增强与门控对齐之间存在互补;其三,E4将SimPVT与SDMConv结合时,性能提升至66.2%/70.0%,且推理速度提升至7.0 FPS,同时参数维持37.03 M,因而在“精度—速度—参数”三维度上取得较为均衡的改进。最后,在三模块联合的情形下,E5在保持6.8 f/s实时性的同时,取得67.1% mAP与70.8% NDS的最佳结果,相对基线提升+2.8%和+1.7%。参数降至36.31 M,较基线减少约8.8%。值得注意的是,E3与E5的参数量(Params)保持一致,表明以SDMConv替换标准稀疏卷积不会引入额外的参数开销。如表7所示,SDMConv作用于BEV点云几何表征特征图时,其平均计算量由6.71×10−3 GFLOPs增至1.34×10−2 GFLOPs,绝对增量为6.71×10−3 GFLOPs;对应推理时间仅由3.5×10−2 ms增至3.8×10−2 ms,仅增加3×10−3 ms。尽管GFLOPs与单模块推理延迟存在小幅上升,但其增量处于毫秒与10−3GFLOPs量级,对整体实时推理链路的影响较为有限。更重要的是,SDMConv通过引入空间密度调制机制,将传统等权聚合转化为对点云稀疏分布的自适应加权聚合,从而强化BEV空间的几何结构建模能力。综合来看,SDMConv在不增加参数量的前提下,以可控的计算与时延代价换取更显著的表征与性能收益,有助于提升模型在复杂场景下的鲁棒性与检测精度。
综上,SimPVT主要强化视觉语义的全局一致性与关键区域显著性,SDMConv面向稀疏与远距场景提升点云几何细节的可分性,GAFM则在BEV空间实现跨模态对齐与互补交互;三者在效率与精度维度形成“可分—可对齐—可互补”的链式增益,使模型在参数规模更小、实时性不降的前提下,显著提高检测鲁棒性与上限。这表明本文的结构设计在保持实时性的同时具备更优的参数效率与检测性能。
为直观评估所提方法的有效性,图5~图8给出典型场景下的实验结果可视化对比。本文方法和BEVFusion基线每组样例均按上下排列,左侧多视角图像与右侧BEV点云以同色虚线同步圈注关注区域,便于在纹理语义与几何回波之间进行一一对应的对比。这表明本文检测结果与LiDAR证据相互印证,性能提升来源于更稳健的跨模态对齐与几何表征增强,而非由视觉噪声诱发的语义误检。
首先,从城市白天主干道场景中观察到(见图5),基线方法在远距小车与细长护栏处易出现漏检或边界残缺,而本文方法在相同圈注区域给出更完整的外接框并保持更稳定的朝向估计,在多车相互遮挡处,框体分离更清晰,边界与BEV点云的高密度回波贴合度更高。该现象主要得益于SimPVT对关键区域响应的强化以及视图变换阶段更稳定的深度分配,同时SDMConv对稀疏边界与细长结构的几何敏感性更强,从而提升了远距与细粒度目标的可检出性与定位一致性。
其次,在多目标密集、尺度跨度大与远距小目标并存的路口/干道场景下(见图6),基线更易出现尺寸低估、姿态漂移与局部漏检等问题。本文方法在远距车辆与施工障碍物上维持连贯框体与正确朝向,圈注区域与点云轮廓高度一致;长边物体(如拖车侧面、长护栏)呈现更好的贴边性与姿态稳定,体现SDMConv在复杂遮挡与多目标干扰条件下对稀疏/细长几何结构的建模优势与GAFM的跨模态错位抑制能力。
此外,为验证所提算法在夜间低照度与阴雨天气等困难工况下的检测能力,除可视化对比外,本文进一步在nuScenes验证集上构建Night与Rain子集开展单独评测,并按官方评价协议统计mAP、NDS及误差分解指标以验证鲁棒性。其中,Night子集包含602个样本,Rain子集包含1088个样本。
在夜间低照度与强反射干扰条件下(见图7表8),强反光、车灯眩光与低照度易诱发基线高置信度误检与框体漂移。相比之下,本文方法在圈注处明显降低虚警,并在暗部区域保持对车辆、路障等关键目标连续检出,同时框体在BEV中仍能稳定对齐稀疏但结构明确的几何回波。与此一致,表8显示本文方法在Night子集上mAP/NDS分别提升+2.0%/+0.8%,且多项误差指标整体下降,表明在低照度与强反射干扰下仍能提升检测稳定性与定位质量。其收益主要来自SimPVT的背景纹理抑制与GAFM的逐位置门控,使融合结果对噪声与反射更不敏感。
同时,在阴雨场景中(见图8表9),雨滴散射与路面反光会造成成像对比度下降与边缘纹理退化,并伴随点云有效回波减少,从而加剧框体抖动与局部漏检的风险。本文方法在圈注区域能够维持更稳定的框体尺度与位置一致性,并对受反光影响较大的目标类型表现更稳健。表9进一步表明,本文方法在Rain子集上mAP/NDS分别提升+2.3%/+0.8%,且速度/属性等误差项同步降低,说明其对降雨导致的感知退化与点云扰动具有更强的鲁棒性。
最后,综合图5~图8的可视化对比与表8~表9的子集定量结果,本文方法在小目标、远距稀疏、强遮挡以及夜间/阴雨强干扰等困难工况下,均较BEVFusion展现出更高的边界贴合度、更稳定的朝向与更低的假阳性率,验证了所提方法在复杂环境下的稳健性与泛化能力。圈注区域在点云BEV上的几何证据与最终检测结果相互支撑,与定量结果(mAP/NDS提升)保持一致,验证了SimPVT语义判别、SDMConv几何建模与GAFM精细对齐与冗余抑制的协同作用。
针对多模态3D目标检测中图像语义信息受限、点云稀疏以及跨模态特征对齐不稳定等关键问题,本文提出的MEPFusion方法有效强化了多模态特征的表达与交互能力,显著提升了检测精度与系统鲁棒性。在nuScenes数据集上的实验验证了该框架的优越性,其mAP与NDS指标相较于BEVFusion基线分别提升了2.8%与1.7%。进一步消融研究揭示,各模块在语义建模、几何结构增强与跨模态对齐方面具有独立增益,并通过协同融合进一步放大了整体性能。本文方法为多模态3D感知任务的深度融合与高效建模提供了可行思路。未来工作将重点探索MEPFusion在极端天气条件、超远距离小目标检测等复杂场景下的适应性与稳定性,并进一步研究面向边缘计算设备的轻量化部署方案,以推动高精度3D感知技术在实际自动驾驶与智能交通系统中的广泛应用。

参考文献 引证文献
排序方式:
1
李文礼, 喻飞, 石晓辉, 等. BEV特征下激光雷达和单目相机融合的目标检测算法研究[J]. 计算机工程与应用, 2024, 60(11): 182−193.
Li W L, Yu F, Shi X H, et al. Research on target detection algorithm for fusion of LiDAR and monocular camera under BEV features[J]. Comput Eng Appl, 2024, 60(11): 182−193.
2
Chen S, Ma Y, Qiao Y, et al. M-BEV: Masked bev perception for robust autonomous driving[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2024, 38(2): 1183-1191. https://doi.org/10.1609/aaai.v38i2.2788
3
Lang B, Li X, Chuah M C. BEV-TP: end-to-end visual perception and trajectory prediction for autonomous driving[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(11): 18537−18546
4
Ge C J, Chen J S, Xie E Z, et al. MetaBEV: solving sensor failures for 3D detection and map segmentation[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision, 2023: 8687–8697. https://doi.org/10.1109/ICCV51070.2023.00801.
5
王五岳, 徐召飞, 曲春燕, 等. 基于红外与激光雷达融合的鸟瞰图空间三维目标检测算法[J]. 光子学报, 2024, 53(1): 0111002.
Wang W Y, Xu Z F, Qu C Y, et al. BEV space 3D object detection algorithm based on fusion of infrared camera and LiDAR[J]. Acta Photonica Sin, 2024, 53(1): 0111002.
6
吉宇哲, 陈奕洁, 杨柳青, 等. 基于鸟瞰图的空间-通道注意力多传感器融合[J]. 模式识别与人工智能, 2023, 36(11): 1029−1040.
Ji Y Z, Chen Y J, Yang L Q, et al. Spatial-channel attention multi-sensor fusion based on Bird's-Eye View[J]. Pattern Recognit Artif Intell, 2023, 36(11): 1029−1040.
7
Philion J, Fidler S. Lift, splat, shoot: Encoding images from arbitrary camera rigs by implicitly unprojecting to 3D[C]//Proceedings of the 16th European Conference on Computer Vision, 2020: 194–210. https://doi.org/10.1007/978-3-030-58568-6_12.
8
Huang J J, Huang G, Zhu Z, et al. BEVDet: high-performance multi-camera 3D object detection in bird-eye-view[Z]. arXiv: 2112.11790, 2022. https://doi.org/10.48550/arXiv.2112.11790.
9
Li Y H, Ge Z, Yu G Y, et al. BEVDepth: acquisition of reliable depth for multi-view 3D object detection[C]//Proceedings of the 37th AAAI Conference on Artificial Intelligence, 2023: 1477–1485. https://doi.org/10.1609/aaai.v37i2.25233.
10
Li Z Q, Wang W H, Li H Y, et al. BEVFormer: learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers[Z]. arXiv: 2203.17270, 2022. https://doi.org/10.48550/arXiv.2203.17270.
11
陈旭东, 韩丽丽, 肖英豪, 薛保国, 马留洋. 基于体素-关键点特征聚合网络的点云3维目标检测[J]. 激光技术, 2026, 50(2): 291−299.
Chen X D, Han L L, Xiao Y H, et al. 3-D object detection for point clouds based on Voxel-Keypoint Feature Aggregation Network[J]. Laser Technol, 2026, 50(2): 291−299.
12
Lang A H, Vora S, Caesar H, et al. PointPillars: fast encoders for object detection from point clouds[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 12689–12697. https://doi.org/10.1109/CVPR.2019.01298.
13
Yan, Y, Mao Y X, Li B. SECOND: sparsely embedded convolutional detection[J]. Sensors, 2018, 18(10): 3337.
14
Wang H, Tao L, Cai Y F, et al. CenterPoint-SE: a single-stage anchor-free 3-D object detection algorithm with spatial awareness enhancement[J]. IEEE Trans Intell Transp Syst, 2023, 24(10): 10760−10773.
15
蔡怀宇, 杨朝乾, 崔子扬, 等. 图像引导和点云空间约束的公路洒落物检测定位方法[J]. 光电工程, 2024, 51(3): 230317.
Cai H Y, Yang Z Q, Cui Z Y, et al. Image-guided and point cloud space-constrained method for detection and localization of abandoned objects on the road[J]. Opto-Electron Eng, 2024, 51(3): 230317.
16
Bai X Y, Hu Z Y, Zhu X G, et al. Transfusion: robust LiDAR-camera fusion for 3D object detection with transformers[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 1080–1089. https://doi.org/10.1109/CVPR52688.2022.00116.
17
Liu Z J, Tang H T, Amini A, et al. BEVFusion: multi-task multi-sensor fusion with unified bird's-eye view representation[C]//Proceedings of 2023 IEEE International Conference on Robotics and Automation (ICRA), 2023: 2774–2781. https://doi.org/10.1109/ICRA48891.2023.10160968.
18
Yin T W, Zhou X Y, Krähenbühl P. Multimodal virtual point 3d detection[C]//Proceedings of the 35th International Conference on Neural Information Processing Systems, 2021: 1261.
19
Chen Z H, Li Z Y, Zhang S Q, et al. Deformable feature aggregation for dynamic multi-modal 3D object detection[C]//Proceedings of the 17th European Conference on Computer Vision, 2022: 628–644. https://doi.org/10.1007/978-3-031-20074-8_36.
20
Li Y W, Chen Y L, Qi X J, et al. Unifying voxel-based representation with transformer for 3D object detection[C]//Proceedings of the 36th International Conference on Neural Information Processing Systems, 2022: 1340.
21
Wang W H, Xie E Z, Li X, et al. PVT v2: improved baselines with pyramid vision transformer[J]. Comput Vis Media, 2022, 8(3): 415−424.
22
Yang L X, Zhang R Y, Li L D, et al. SimAm: a simple, parameter-free attention module for convolutional neural networks[C]//Proceedings of the 38th International Conference on Machine Learning, 2021: 11863–11874.
23
Cammarasana S, Patanè G. Optimal weighted convolution for classification and denosing[Z]. arXiv: 2505.24558, 2025.https://doi.org/10.48550/arXiv.2505.24558.
24
栾庆磊, 常昕昱, 吴叶, 等. PAW-YOLOv7: 河道微小漂浮物检测算法[J]. 光电工程, 2024, 51(4): 240025.
Luan Q L, Chang X Y, Wu Y, et al. PAW-YOLOv7: algorithm for detection of tiny floating objects in river channels[J]. Opto-Electron Eng, 2024, 51(4): 240025.
25
夏金泽, 孙浩铭, 胡盛辉, 等. 基于图像信息约束的三维激光点云聚类方法[J]. 光电工程, 2023, 50(2): 220148.
Xia J Z, Sun H M, Hu S H, et al. 3D laser point cloud clustering method based on image information constraints[J]. Opto-Electron Eng, 2023, 50(2): 220148.
26
张善文, 邵彧. 一种改进U-Net的ARSI目标检测方法[J]. 电光与控制, 2025, 32(9): 1−7.
Zhang S W, Shao Y. Modified U-Net based ARSI object detection[J]. Electron Opt Control, 2025, 32(9): 1−7.
27
Vora S, Lang A H, Helou B, et al. PointPainting: sequential fusion for 3D object detection[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 4603–4611. https://doi.org/10.1109/CVPR42600.2020.00466.
28
Zeng Y H, Zhang D, Wang C W, et al. LIFT: learning 4D LiDAR image fusion transformer for 3D object detection[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 17151–17160. https://doi.org/10.1109/CVPR52688.2022.01666.
29
Cui L C, Li X X, Meng M, et al. MMFusion: a generalized multi-modal fusion detection framework[C]//Proceedings of 2023 IEEE International Conference on Development and Learning (ICDL), 2023: 415–422. https://doi.org/10.1109/ICDL55364.2023.10364441.
30
Song Z Y, Wei H Y, Bai L, et al. GraphAlign: enhancing accurate feature alignment by graph matching for multi-modal 3D object detection[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision, 2023: 3335–3346. https://doi.org/10.1109/ICCV51070.2023.00311.
2026年第53卷第4期
PDF下载
140
60
引用本文
BibTeX
文章信息
doi: 10.12086/oee.2026.250340
  • 接收时间:2025-11-06
  • 首发时间:2026-07-02
  • 出版时间:2026-04-24
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2025-11-06
  • 修回日期:2026-01-11
  • 录用日期:2026-01-12
基金
作者信息
    1磁浮轨道交通装备江西省重点实验室,江西 赣州 341000
    2江西理工大学永磁磁浮技术与轨道交通研究院,江西 赣州 341000
    3江西理工大学电气工程与自动化学院,江西 赣州 341000

通讯作者:

参考文献
分享链接
https://castjournals.cast.org.cn/joweb/oee/CN/10.12086/oee.2026.250340
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