TFT-LCDCircuitDefectsDetectionBasedonFasterR-CNN

doi:10.3969/j.issn.1006-2475.2018.07.007

Abstract

Abstract: ThedetectionoftinyandcomplexdefectsinthebordercircuitofThinFilmTransistor-LiquidCrystalDisplay(TFT-LCD)hasbeenadifficultpointinAutomaticOpticalInspection(AOI).ThispaperdetectsTFT-LCDbordercircuitdefectsbyusingimprovedFasterRegion-basedConvolutionalNeuralNetwork(FasterR-CNN).Thealgorithmextractsfeaturesfromsharedconvolutionallayersfirstly,andthengeneratescandidateregionsaccuratelythroughthemultilayeredRegionProposalNetwork(RPN),whichcanrecognizeandlocatethetargetscombiningwithclassificationinformation.Weanalyzetheperformancesofthemethodwithdifferentnetworkstructureswedesigned,andcomparewithdifferentalgorithms.Theexperimentstrainedinabordercircuitdatasetshowthatthemethodachievesexcellentperformance,andthedetectionsystemcanrecognizeandlocatesixkindsofTFT-LCDbordercircuitdefectsinoneimagesimultaneouslywithin0.12sandachieveanaccuracyof94.6%.

Key words: defectdetection, industrialintelligence, region-basedconvolutionalneuralnetwork, deeplearning

CLC Number:

TP312

HEJun-jie,XIAOKe,LIUChang,CHENSong-yan. TFT-LCDCircuitDefectsDetectionBasedonFasterR-CNN[J]. Computer and Modernization, 2018, 0(07): 33-.

References

［1］LoweDG.Distinctiveimagefeaturesfromscale-invariantkeypoints［J］.InternationalJournalofComputerVision,2004,60(2):91-110.
［2］DalalN,TriggsB.Histogramsoforientedgradientsforhumandetection［C］//IEEEComputerSocietyConferenceonComputerVisionandPatternRecognition.2005:886-893.
［3］OjalaT,PietikinenM,HarwoodD.PerformanceevaluationoftexturemeasureswithclassificationbasedonKullbackdiscriminationofdistributions［C］//Proceedingsofthe12thIAPRInternationalConferenceonPatternRecognition.1994:582-585.
［4］朱光,朱学芳,张华坤.复杂背景下TFT-LCD表面缺陷检测系统的设计［J］.电子测量与仪器学报,2011,25(12):1054-1059.
［5］苏小红,何志广,马培军.TFT-LCD微米级显示缺陷的自动检测算法［J］.哈尔滨工业大学学报,2008,40(11):1756-1760.
［6］GallJ,LempitskyV.Class-specifichoughforestsforobjectdetection［C］//2009IEEEConferenceonComputerVisionandPatternRecognition.2009:1022-1029.
［7］FelzenszwalbPF,GirshickRB,McAllesterD.Cascadeobjectdetectionwithdeformablepartmodels［C］//2010IEEEConferenceonComputerVisionandPatternRecognition.2010:2241-2248.
［8］GirshickR,DonahueJ,DarrellT,etal.Richfeaturehierarchiesforaccurateobjectdetectionandsemanticsegmentation［C］//2014IEEEConferenceonComputerVisionandPatternRecognition.2014:580-587.
［9］GirshickR.FastR-CNN［C］//ProceedingsoftheIEEEInternationalConferenceonComputerVision.2015:1440-1448.
［10］HeKaiming,ZhangXiangyu,RenShaoqing,etal.Spatialpyramidpoolingindeepconvolutionalnetworksforvisualrecognition［C］//EuropeanConferenceonComputerVision.2014:346-361.
［11］RenShaoqing,HeKiaming,GirshickR,etal.FasterR-CNN:Towardsreal-timeobjectdetectionwithregionproposalnetworks［C］//Proceedingsofthe28thInternationalConferenceonNeuralInformationProcessingSystems.2015:91-99.
［12］UijlingsJRR,VanDeSandeKEA,GeversT,etal.Selectivesearchforobjectrecognition［J］.InternationalJournalofComputerVision,2013,104(2):154-171.
［13］NeubeckA,VanGoolL.Efficientnon-maximumsuppression［C］//IEEE18thInternationalConferenceonPatternRecognition.2006,3:850-855.
［14］ZeilerMD,FergusR.Visualizingandunderstandingconvolutionalnetworks［C］//EuropeanConferenceonComputerVision.2014:818-833.
［15］KrizhevskyA,SutskeverI,HintonGE.Imagenetclassificationwithdeepconvolutionalneuralnetworks［C］//Proceedingsofthe25thInternationalConferenceonNeuralInformationProcessingSystems.2012:1097-1105.
［16］SimonyanK,ZissermanA.Verydeepconvolutionalnetworksforlarge-scaleimagerecognition［J］.ComputerScience,2014:arXiv:1409.1556.
［17］NairV,HintonGE.Rectifiedlinearunitsimproverestrictedboltzmannmachines［C］//Proceedingsofthe27thInternationalConferenceonMachineLearning(ICML-10).2010:807-814.
［18］SrivastavaN,HintonG,KrizhevskyA,etal.Dropout:Asimplewaytopreventneuralnetworksfromoverfitting［J］.JournalofMachineLearningResearch,2014,15(1):1929-1958.

[1]	LINXue-feng1,XIAYuan-yi2,GUOJin-long1,YUXiao-wen1. SensitiveFileDetectionMethodBasedonCNN [J]. Computer and Modernization, 2018, 0(07): 28-.
[2]	ZHANGWen. Multi-taskFaceAttributeRecognitionBasedonUnbalancedClasses [J]. Computer and Modernization, 2018, 0(06): 58-.