欢迎来到天天文库
浏览记录
ID:37046406
大小:4.45 MB
页数:68页
时间:2019-05-15
《Research on Fast R-CNN for Visual Object Detection》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、硕士留学生学位论文ResearchonFastR-CNNforVisualObjectDetection作者姓名GHULAMARBI学科专业电气与计算机工程指导教师Prof.YUZHULIANG所在学院自动化科学与工程学院论文提交日期2018年5月ResearchonFastR-CNNforVisualObjectDetectionADissertationSubmittedfortheDegreeofMasterADissertationSubmittedfortheDegreeofMasterCandidate
2、:GHULAMARBISupervisor:Prof.YUZHULIANGSouthChinaUniversityofTechnologyGuangzhou,ChinaJune2018AbstractOverthelastdecade,objectdetectionhasemergedasanimportantareaofresearchforcomputervision.Inthisregard,somenovelandimpressingCNNarchitectureshavebeenproposed.Fort
3、hisworkFastR-CNNandVGGNetarchitectureshavebeenincorporatedforthegiventaskofvisualobjectdetection.DifferentproposedmethodshavebeenexercisedandempiricallyitisrealizedthatFastR-CNNarchitectureexhibitsconsiderableperformanceforthegiventask.Inparticularforgivenprob
4、lem,emphasishasbeenrenderedonvisualobjectCapraFalconeri(Markhor),indifferentchallengingenvironmentandbackgroundimages.Themostissueforobjectdetectionusingdeeplearningtechniquesistosavethetrainingandtestingtimewithoutsacrificingitsaccuracy.However,existingFastR-
5、CNNtechniqueprovidegoodaccuracybutitsuffersfrommoretrainingandtestingtimethatlimitstheoverallperformanceofsystem.Inthisthesis,anewdatasetiscreatednamedasMarkhor-VIbyvariousdataaugmentationmethodssuchastranslation,rotation,scalingandsoon.Thenwehavemodifiedexist
6、ingFastR-CNNtechniquebytransferlearningbyvariousmethodsfromlayer1tolayer5.Itisfoundthatconvolutionalobjectdetectionisstillevolvingasatechnology,despiteoutrankingotherobjectdetectionmethods.Itisalsofoundthattransferlearningcansavealotofresourcesiffinetuningispe
7、rformedinaparticularway.Keywords:ArtificialIntelligence,DeepLearning,ObjectDetection,ComputerVision.ITableofContentsAbstract........................................................................................................................................
8、............ITableofContents....................................................................................................................................IIListofFigures.....
此文档下载收益归作者所有