欢迎来到天天文库
浏览记录
ID:32149470
大小:2.54 MB
页数:70页
时间:2019-01-31
《基于视觉的输电线路除冰机器人障碍识别方法》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、●硕士学位论文AbstractObstaclerecognitionisoneofthekeytechniquesofautonomousdeicingrobotonhighVoltagetransmissionline.Theobstaclessuchascounterweight,strainclampandsuspensiononhighVoltagetransmissionlineshouldbeef.fectiVelyrecognizedflordeicingrobottoautonomouslycrossobst
2、acles.Accordingtothestructureof220kVtransmissionline,aseriesofmethodsfbfobstacIerecognitionbasedonVisionareputfIorwardinthisthesis.Theproposedmethodsdon’trequirestructureconstraintandcanachieVegoodobstacleclassificationresults.Themainstudiesinthispaperareasfollows.
3、1)Theobstacleimagesarepretreatedfirstly,andthentheedgesofobstacleimagesaredetectedbyusingwaVeletmodulusmaximumalgorithmandcannyalgorithmrespectiVely.TheresultshowsthattheedgesofobstacIeimageswhichdetectedbywaveletmodulusmaximumaIgorithmwhichhaVeastrongeranti-jammin
4、gcapabilityhaVeabetteredgedetectionperf.0rmancethancannyalgorithm.2)Themomentfeaturesofobstacle’sedgeimagesareselectedasobstacleclassifierinputVector.ThentheunitedmomentsandthewaVeletmomentsarecalculatedandthebothmomentfeaturesarereduceddimensionandoptimizedbyusing
5、sub—optimalsearchalgorithm.3)Obstacleclassi6cationmethodsbasedonneuralnetworkareresearched.Firstofall,amultilayerfeed-forwardneuralnetworkandawaveletneuralnetworkbothbasedonBPalgorithmareproposed。SinceBPalgorithmhassIowconVergenceandiseasytendencytopartialoptimizat
6、ion,theparticleswarmoptimization(PSO)algorithmwithaf.asterconvergenceandastrongerglobaIsearchcapabilityisintroducedtoreplacetheBPalgorithmtotrainingthewaVeletneuralnetwork,thenawaVeletnetworkbasedonPSOisestablished.HoweVer,PSOalgorithmmaystillf.allintopartialoptimi
7、zation,drawingontheideaofsuddenjumpinsimulatedannealing,atlast,aimprovedPSOwaVeletnetworkbasedonsimulatedannealingalgorithmisdeVelopedanddemonstratesexcellentclassificationperfbrmance.4)Taking—intoaccountthesupportVectormachine(SVM)ismoresuitablgforsmallsamplepatte
8、rnfecognitionproblem,obstacleclassificationmethodsbasedonSVMarealsostudied.Tbbeginwith,aSVMclassifierbaSedon盯idsearchandcross—Validationforparame
此文档下载收益归作者所有