欢迎来到天天文库
浏览记录
ID:52175827
大小:4.07 MB
页数:8页
时间:2020-03-23
《基于多特征流形学习和矩阵分解的路面裂缝检测.pdf》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、第37卷第7期2016年7月仪器仪表学报ChineseJournalofScientificInstrumentVoL37No。7Jut.2016基于多特征流形学习和矩阵分解的路面裂缝检测钱彬,唐振民,沈肖波,郭剑辉,吕建勇(南京理工大学计算机科学与工程学院南京210094)摘要:针对单一属性特征的路面裂缝检测方法无法从复杂背景噪声中准确提取裂缝信息的缺陷,提出一种结合多特征流形学习和矩阵分解的路面裂缝检测算法。该算法首先根据路面裂缝子块的统计、形状和纹理特性抽取多重属性特征并构造多个流形正则项,将流形正则项嵌入于矩阵分解的目标函数中,采用交替迭代法在统一框架下实
2、现裂缝子块降维和多特征自适应融合。为进一步提高聚类准确率,对路面裂缝图像采用各向异性算法增强得到少量有效样本标签,实现算法的半监督扩展。在公开数据集(CrackIT)和实际采集的沪宁高速(HN)路面图像库上的实验结果表明,该算法可以有效提高路面裂缝识别率,验证了算法的有效性。关键词:裂缝检测;流形学习;多特征融合;矩阵分解中图分类号:TP391.41TH701文献标识码:A国家标准学科分类代码:520.60Pavementcrackdetectionbasedonmulti.featuremanifoldlearningandmatrixfactorization
3、QianBin,TangZhenmin,ShenXiaobo,GuoJianhui,LvJianyong(SchoolofComputerScienceandEngineering,NanjingUniversityofScienceandTechnology,Na研ng210094,China)Abstract:Aimingatthedefectthatusingonlysingletypeofattributefeature,conventionalpavementcrackdetectionmethodscannotaccuratelyextractcrac
4、kinformationfromcomplicatedbackgroundnoises;inthispaper,anovelpavementcrackdetectionalgorithmisproposedbasedonintegratingmulti—featuremanifoldlearningandmatrixfactorization.Firstly,accordingtothestatistics,shapeandtexturefeaturesofthepavementcracksub-patches,themultipleattributefeatur
5、esaleextractedandmultiplemanifoldregularizedtermsaleestablished.Then,themanifoldregularizedtermsareembeddedintotheobjectivefunctionofmatrixfactorization.Finally,analternatingiterationalgorithmisadoptedtorealizethedimensionreductionofthecracksub—patchesandadaptivemultiplefeaturefusionw
6、ithinaunifiedframework.Tofurtherimprovetheclusteringaccuracy,ananisotropyalgorithmisadoptedtoenhancethepavementcrackimage。partialeffectivesamplelabelsareobtained,andthesemi—supervisedexpansionofthealgorithmisachieved.Theexperimentresultsonthepublicpavementcrackdataset(CraeklT)andthepr
7、acticallyacquiredHu-Ning(HN)highwaypavementcrackimagedatabaseshowthattheproposedalgorithmCaneffectivelyimprnvethepavementcrackrecognitionrate,whichverifiestheeffectivenessofthealgorithm.Keywords:crackdetection;manifoldlearning;multi·featurefusion;matrixfactorization1引言路面裂缝检测是智能交通领域的重要
8、组成部分,
此文档下载收益归作者所有