基于改进的Faster R-CNN的古建筑地砖缺陷检测-豆柴文库

您所在位置：网站首页 / 基于改进的Faster R-CNN的古建筑地砖缺陷检测.docx / 文档详情

在线预览结束，喜欢就下载吧，查找使用更方便

5 金币

下载文档

/ 2

下载提示

如果您无法下载资料，请参考说明：

1、部分资料下载需要金币，请确保您的账户上有足够的金币

2、已购买过的文档，再次下载不重复扣费

3、资料包下载后请先用软件解压，在使用对应软件打开

文本预览

基于改进的FasterR-CNN的古建筑地砖缺陷检测
Title:DetectingDefectsinAncientArchitectureFloorTilesbasedonImprovedFasterR-CNN
Abstract:
Withtheincreasingattentiononthepreservationofancientarchitecture,thedetectionofdefectsinhistoricfloortileshasbecomeacrucialtask.Inthispaper,weproposeanimprovedFasterR-CNNmodelfortheautomaticdetectionoffloortiledefectsinancientarchitecture.Theproposedmodelnotonlyachieveshigheraccuracybutalsoenhancesthedetectionspeed.Throughempiricalexperiments,ourmodeldemonstratessuperiorperformanceinidentifyingvarioustypesofdefects,providingavaluabletoolforthepreservationandmaintenanceofancientarchitecture.
1.Introduction
Ancientarchitecturepossesseshistorical,cultural,andartisticvaluesthatmustbepreservedforfuturegenerations.Thefloortilesusedinancientarchitecturearesubjectedtovariousdefectsovertime,suchascracks,colorfading,andmissingtiles.Detectingthesedefectsmanuallyisalabor-intensiveandtime-consumingtask.Hence,theneedforanautomatedsystemtodetectandclassifyfloortiledefectsisimperative.Inthispaper,weproposeanimprovedFasterR-CNNmodeltoaddressthisissue.
2.RelatedWork
Previousstudieshaveexploredtheuseofdeeplearningmodelsforobjectdetectiontasks.TheFasterR-CNNisknownforitsexcellentperformanceinobjectdetection,makingitapplicableinvariousdomains.However,theoriginalFasterR-CNNmodelmaynotachievesatisfactoryresultswhenappliedtoancientarchitecturefloortiledefectdetectionduetothespecificcharacteristicsofthesedefects.Hence,weproposeseveralimprovementstoenhancethemodel'sperformance.
3.Methodology
3.1DataCollectionandPreprocessing
Totrainandevaluatetheproposedmodel,wecollectedadatasetofimagescontainingdifferenttypesoffloortiledefectsfoundinancientarchitecture.Eachimagewaslabeledwithgroundtruthannotationstodenotethepresenceandlocationofdefects.Thedatasetwassplitintotrainingandtestingsets.
3.2ImprovedFasterR-CNNModel
WeproposeseveralmodificationstothestandardFasterR-CNNarchitecturetooptimizeitforfloortiledefectdetection.First,weincorporatearegionproposalnetwork(RPN)withanchorboxesspec