基于机器视觉的通信装备故障识别分析

《基于机器视觉的通信装备故障识别分析》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、万方数据西南交通大学硕士研究生学位论文第1I页关键词：机器视觉；故障识别；支持向量机；BP神经网络；特征提取万方数据西南交通大学硕士研究生学位论文第1II页ABSTRACTMachinevisionisveryimportantfortheidentificationanddetectionofobjects，becauseofhighdetectionprecision，non·contactproperty,stabilityandSOon．Thesystemforcommunicationequipmentmalfunctiondetectionbasedonmachinevisi

2、oncoulddetectthedevicesinrealtime，moreoveritCanreducetheconsumptionofhumanresourcesandthemalfunctionlocationtimesignificantlyThethesishasstudiedthemalfunctiondetectionforlargecommunicationsequipmentbycombiningimageprocessingalgorithmswithneuralnetworks，afteranalyzingthecurrentresearchonmachinevis

3、ion．Itincludesthesegmentationandclassificationforcommunicationsequipmenttoolboards，andthemalfunctionidentificationfornetworkportsandISV3board’Sdualinterfaces．Inordertosolvetherecognitionproblemoflarge—scalecommunicationequipmentcomponents，theclassificationmethodcombiningSIFTandSVMisproposed．Extra

4、ctthefeaturesofsampleimagesbySIFTalgorithm，andmakeclusteranalyzingbyKmeansalgorithm．Evenmore，gaintheinputdatabasedontheminimumEuclideandistance．AfteranalyzingtherelationshipbetweentheSVMdetectionaccuracyandkernelfunctiontype，7value，penakyfactorC，theSVMclassifierwiththebestperformanceisusedtodetec

5、tthetargetimages，andtheaccuracyrateismorethan92％．IdentifyingwhetherthenetportsarepluggedthroughthedetectionalgorithmcombininggrayhistogramandBPneuralnetwork．ExtractthefeaturepointsbySIFTalgorithm，andmakefeaturematchingwithusingthenearestneighbormethod．Thencorrectthedetectedimageusingtheoptimaltra

6、nsformationmatrixgainedbyRANSACalgorithm．Moreover,gettheinputdatabymakinghistogramstatisticsforthenetportsamples．AfteranalyzingtheimpactofhiddenlayernodesandthenumbersofiterationsonBPneuralnetwork，whichwiththebestperformanceisusedtodetectthetargetnetports，andtheaccuracyrateisreached96％Characterre

7、cognitionmethodisusedtodeterminewhetherthecablesandnetportsarematched．Makegrayingandbinarizingfortheoriginalimage，andseparatethecharactersbyfindingboundarypointsaccordingtothedistributionofpixelsonthebinaryimage．Letter