文档名：基于深度学习的小目标检测基准研究进展
摘要：小目标检测是计算机视觉中极具挑战性的任务.它被广泛应用于遥感、交通、国防军事和日常生活等领域.相比其他视觉任务,小目标检测的研究进展相对缓慢.制约因素除了学习小目标特征的内在困难,还有小目标检测基准,即小目标检测数据集的稀缺以及建立小目标检测评估指标的挑战.为了更深入地理解小目标检测,本文首次对基于深度学习的小目标检测基准进行了全新彻底的调查.系统介绍了现存的35个小目标数据集,并从相对尺度和绝对尺度(目标边界框的宽度或高度、目标边界框宽高的乘积、目标边界框面积的平方根)对小目标的定义进行全面总结.重点从基于交并比及其变体、基于平均精度及其变体以及其他评估指标这3方面详细探讨了小目标检测评估指标.此外,从锚框机制、尺度感知与融合、上下文信息、超分辨率技术以及其他改进思路这5个角度对代表性小目标检测算法进行了全面阐述.与此同时,在6个数据集上对典型评估指标(评估指标+目标定义、评估指标+单目标类别)下的代表性小目标检测算法进行性能的深入分析与比较,并从小目标检测新基准、小目标定义的统一、小目标检测新框架、多模态小目标检测算法、旋转小目标检测以及高精度且实时的小目标检测这6个方面指出未来可能的发展趋势.希望该综述可以启发相关研究人员,进一步促进小目标检测的发展.
Abstract：Smallobjectdetectionisanextremelychallengingtaskincomputervision.Itiswidelyusedinremotesensing,intelligenttransportation,nationaldefenseandmilitary,dailylifeandotherfields.Comparedtoothervisualtaskssuchasimagesegmentation,actionrecognition,objecttracking,genericobjectdetection,imageclassification,videocaptionandhumanposeestimation,theresearchprogressofsmallobjectdetectionisrelativelyslow.Webelievethattheconstraintsmainlyincludetwoaspects:theintrinsicdifficultyoflearningsmallobjectfeaturesandthescarcityofsmallobjectdetectionbenchmarks.Inparticular,thescarcityofsmallobjectdetectionbenchmarkscanbeconsideredfromtwoaspects:thescarci-tyofsmallobjectdetectiondatasetsandthedifficultyofestablishingevaluationmetricsforsmallobjectdetection.Togainadeeperunderstandingofsmallobjectdetection,thisarticleconductsabrand-newandthoroughinvestigationonsmallobjectdetectionbenchmarksbasedondeeplearningforthefirsttime.Theexisting35smallobjectdetectiondatasetsareintro-ducedfrom7differentapplicationscenarios,suchasremotesensingimages,trafficsignandtrafficlightdetection,pedestri-andetection,facedetection,syntheticapertureradarimagesandinfraredimages,dailylifeandothers.Meanwhile,compre-hensivelysummarizethedefinitionofsmallobjectsfrombothrelativescaleandabsolutescale.Fortheabsolutescale,itmainlyincludes3categories:thewidthorheightoftheobjectboundingbox,theproductofthewidthandheightoftheob-jectboundingbox,andthesquarerootoftheareaoftheobjectboundingbox.Thefocusisonexploringtheevaluationmet-ricsofsmallobjectdetectionindetailfrom3aspects:basedonIoU(IntersectionoverUnion)anditsvariants,basedonaver-ageprecisionanditsvariants,andotherevaluationmetrics.Inaddition,in-depthanalysisandcomparisonoftheperfor-manceofsomerepresentativesmallobjectdetectionalgorithmsundertypicalevaluationmetricsareconductedon6datas-ets.Thesecategoriesoftypicalevaluationmetricscanbefurthersubdivided,includingtheevaluationmetricplusthedefini-tionofobjects,theevaluationmetricplussingleobjectcategory.Moreconcretely,theevaluationmetricsplusthedefinitionofobjectscanbedividedinto4categories:averageprecisionplusthedefinitionofobjects,missrateplusthedefinitionofobjects,DoR-AP-SM(DegreeofReductioninAveragePrecisionbetweenSmallobjectsandMediumobjects)andDoR-AP-SL(DegreeofReductioninAveragePrecisionbetweenSmallobjectsandLargeobjects).Fortheevaluationmetricsplussingleobjectcategory,itmainlyincludes2types:averageprecisionplussingleobjectcategory,OLRP(OptimalLocaliza-tionRecallPrecision)plussingleobjectcategory.Theserepresentativesmallobjectdetectionmethodsmainlyincludean-chormechanism,scale-awareandfusion,contextinformation,super-resolutiontechniqueandotherimprovementideas.Fi-nally,wepointoutthepossibletrendsinthefuturefrom6aspects:anewbenchmarkforsmallobjectdetection,aunifieddefinitionofsmallobjects,anewframeworkforsmallobjectdetection,multi-modalsmallobjectdetectionalgorithms,rotat-ingsmallobjectdetection,andhighprecisionandrealtimesmallobjectdetection.Wehopethatthispapercouldprovideatimelyandcomprehensivereviewoftheresearchprogressofsmallobjectdetectionbenchmarksbasedondeeplearning,andinspirerelevantresearcherstofurtherpromotethedevelopmentofthisfield.
作者：童康 吴一全Author：TONGKang WUYi-quan
作者单位：南京航空航天大学电子信息工程学院,江苏南京211106
刊名：电子学报 ISTICEIPKU
Journal：ActaElectronicaSinica
年，卷(期)：2024, 52(3)
分类号：TP389.1TP391.41
关键词：小目标检测 深度学习 小目标评估指标 小目标数据集 小目标定义 小目标检测基准
Keywords：smallobjectdetection deeplearning evaluationmetricofsmallobjects smallobjectdataset thedefini-tionofsmallobjects smallobjectdetectionbenchmark
机标分类号：TP391.41TN911.73U445.7
在线出版日期：2024年5月16日
基金项目：基于深度学习的小目标检测基准研究进展[
期刊论文] 电子学报--2024, 52(3)童康 吴一全小目标检测是计算机视觉中极具挑战性的任务.它被广泛应用于遥感、交通、国防军事和日常生活等领域.相比其他视觉任务,小目标检测的研究进展相对缓慢.制约因素除了学习小目标特征的内在困难,还有小目标检测基准,即小目标检...参考文献和引证文献
参考文献
引证文献
本文读者也读过
相似文献
相关博文
基于深度学习的小目标检测基准研究进展 Research Advances on Deep Learning Based Small Object Detection Benchmarks
基于深度学习的小目标检测基准研究进展.pdf
- 文件大小:
- 1.8 MB
- 下载次数:
- 60
-
高速下载
|
|
