基于低秩稀疏結(jié)構(gòu)分解的紡織品瑕疵檢測論文設(shè)計(jì)

摘要現(xiàn)有的紡織品瑕疵檢測算法多采用傳統(tǒng)的模式識(shí)別方法，如統(tǒng)計(jì)法、頻譜法和訓(xùn)練法等。近些年來，低秩稀疏結(jié)構(gòu)分解模型在顯著性檢測等領(lǐng)域得到了廣泛的應(yīng)用。低秩稀疏矩陣結(jié)構(gòu)分解模型將待檢測的特征圖像矩陣分解為低秩矩陣和稀疏矩陣兩個(gè)部分，其中低秩矩陣用來表示背景，稀疏矩陣用來表示稀疏矩陣。而紡織品在視覺上具有高度的冗余性，因此可以利用低秩稀疏結(jié)構(gòu)分解來進(jìn)行紡織品的瑕疵檢測。本文以周期性紡織品為樣本作為瑕疵檢測的研究對象，采用低秩稀疏結(jié)構(gòu)分解作為研究方法，所做的工作及研究成果如下：提出了一種基于模板校正與低秩分解的紡織品瑕疵檢測方法。首先對原紡織品圖像進(jìn)行模板校正，以減輕圖像拉伸、變形及光照對檢測結(jié)果的影響；然后提出低秩校正分解模型，包含低秩項(xiàng)、稀疏項(xiàng)和校正項(xiàng)，采用交替方向法進(jìn)行優(yōu)化求解，將原特征矩陣分解為低秩矩陣和稀疏矩陣，其中低秩矩陣表示背景，稀疏矩陣表示瑕疵區(qū)域；最后利用最優(yōu)閾值分割算法，對由稀疏矩陣產(chǎn)生的顯著圖進(jìn)行閾值分割，得到二值化的檢測結(jié)果。提出了一種基于權(quán)重低秩分解模型的紡織品瑕疵檢測方法。采用塊分割法獲取瑕疵先驗(yàn)，瑕疵先驗(yàn)用于增加對大的瑕疵塊的檢測率。通過瑕疵先驗(yàn)來指導(dǎo)模型的分解，懲罰缺陷區(qū)域，降低算法的誤檢率，提高檢測精度。提出了一種基于權(quán)重低秩分解與拉普拉斯正則項(xiàng)模型的紡織品瑕疵檢測算法。在權(quán)重低秩分解模型的基礎(chǔ)上加入拉普拉斯正則項(xiàng)，提高對細(xì)小瑕疵的檢測精度，通過拉普拉斯正則項(xiàng)將具有相似像素的塊共享相似的表示，不同的像素采用不同的表示方式，以此增加瑕疵與背景之間的距離。并且采用交替方向法對所提的凸優(yōu)化模型進(jìn)行優(yōu)化求解，最后，采用自適應(yīng)的閾值分割算法對由稀疏矩陣產(chǎn)生的顯著圖進(jìn)行分割，定位出瑕疵區(qū)域。關(guān)鍵詞:紡織品瑕疵檢測;低秩分解;模板校正；拉普拉斯正則項(xiàng)1緒論1.1研究背景和意義在紡織工業(yè)中，紡織品的生產(chǎn)通常在針織機(jī)上進(jìn)行ADDINZOTERO_ITEMCSL_CITATION{"citationID":"X2x21zmn","properties":{"formattedCitation":"\\super[1]\\nosupersub{}","plainCitation":"[1]","noteIndex":0},"citationItems":[{"id":215,"uris":["/users/local/xIMhsWAM/items/R7HS83TF"],"uri":["/users/local/xIMhsWAM/items/R7HS83TF"],"itemData":{"id":215,"type":"article-journal","container-title":"Optik","DOI":"10.1016/j.ijleo.2016.09.110","ISSN":"00304026","issue":"24","language":"en","page":"11960-11973","source":"Crossref","title":"Fabricdefectdetectionsystemsandmethods—Asystematicliteraturereview","volume":"127","author":[{"family":"Hanbay","given":"Kaz?m"},{"family":"Talu","given":"MuhammedFatih"},{"family":"?zgüven","given":"?merFaruk"}],"issued":{"date-parts":[["2016",12]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[1]。因此，這個(gè)過程總會(huì)產(chǎn)生各種各樣的瑕疵。研究表明，紡織品的瑕疵可以造成紡織品利潤下降45%-65%ADDINZOTERO_ITEMCSL_CITATION{"citationID":"NsJCBLKu","properties":{"formattedCitation":"\\super[2]\\nosupersub{}","plainCitation":"[2]","noteIndex":0},"citationItems":[{"id":218,"uris":["/users/local/xIMhsWAM/items/YKWDZDBB"],"uri":["/users/local/xIMhsWAM/items/YKWDZDBB"],"itemData":{"id":218,"type":"article-journal","container-title":"Neurocomputing","DOI":"10.1016/j.neucom.2015.09.011","ISSN":"09252312","language":"en","page":"1386-1401","source":"Crossref","title":"Differentialevolution-basedoptimalGaborfiltermodelforfabricinspection","volume":"173","author":[{"family":"Tong","given":"Le"},{"family":"Wong","given":"W.K."},{"family":"Kwong","given":"C.K."}],"issued":{"date-parts":[["2016",1]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[2]。因此，在紡織品生產(chǎn)過程中，對其進(jìn)行紡織品的瑕疵檢測是不可缺少的一步。但由于紡織品圖像本身具有復(fù)雜多變的紋理以及各式各樣的瑕疵類型，給瑕疵檢測算法的研究帶來了一定的挑戰(zhàn)性。在目前的工業(yè)生產(chǎn)中，紡織品的瑕疵檢測主要采用人工檢測的方式，如圖1-1所示。但是人工檢測存在許多的缺點(diǎn)，如，誤檢率高，檢測速度慢，成本高等。然而，紡織品的自動(dòng)瑕疵檢測能彌補(bǔ)上述的這些缺點(diǎn)ADDINZOTERO_ITEMCSL_CITATION{"citationID":"FKAtLEgL","properties":{"formattedCitation":"\\super[3]\\nosupersub{}","plainCitation":"[3]","noteIndex":0},"citationItems":[{"id":217,"uris":["/users/local/xIMhsWAM/items/32JYVAL9"],"uri":["/users/local/xIMhsWAM/items/32JYVAL9"],"itemData":{"id":217,"type":"article-journal","container-title":"IEEETransactionsonPatternAnalysisandMachineIntelligence","DOI":"10.1109/TPAMI.1982.4767309","ISSN":"0162-8828","issue":"6","page":"557-573","source":"Crossref","title":"AutomatedVisualInspection:ASurvey","title-short":"AutomatedVisualInspection","volume":"PAMI-4","author":[{"family":"Chin","given":"RolandT."},{"family":"Harlow","given":"CharlesA."}],"issued":{"date-parts":[["1982",11]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[3]?，F(xiàn)如今，越來越多的紡織品企業(yè)采用了自動(dòng)化瑕疵檢測設(shè)備，檢測速度可以達(dá)到每分鐘120米，并且檢測成功率一般在90%左右ADDINZOTERO_ITEMCSL_CITATION{"citationID":"M6QjD9Dp","properties":{"formattedCitation":"\\super[4]\\nosupersub{}","plainCitation":"[4]","noteIndex":0},"citationItems":[{"id":130,"uris":["/users/local/xIMhsWAM/items/VAQKQEI6"],"uri":["/users/local/xIMhsWAM/items/VAQKQEI6"],"itemData":{"id":130,"type":"article-journal","container-title":"IEEETransactionsonIndustryApplications","issue":"5","page":"1267-1276","title":"FabricdefectdetectionbyFourieranalysis","volume":"36","author":[{"family":"Chan","given":"ChiHo"},{"family":"Pang","given":"G.K.H."}],"issued":{"date-parts":[["2000"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[4]。對于各種各樣的瑕疵類型，自動(dòng)瑕疵檢測系統(tǒng)都有統(tǒng)一的評(píng)價(jià)標(biāo)準(zhǔn)，避免了主觀判斷給檢測帶來的影響。近年來，隨著低秩稀疏結(jié)構(gòu)分解模型的快速發(fā)展，利用低秩稀疏結(jié)構(gòu)分解模型進(jìn)行顯著性檢測日漸成熟。低秩稀疏結(jié)構(gòu)分解模型的原理是將圖像的特征數(shù)據(jù)矩陣分解為低秩矩陣和稀疏矩陣，用低秩矩陣來表示背景，用稀疏矩陣來表示顯著性區(qū)域。如利用低秩稀疏結(jié)構(gòu)分解模型進(jìn)行目標(biāo)檢測ADDINZOTERO_ITEMCSL_CITATION{"citationID":"a1vnsnot92p","properties":{"formattedCitation":"\\super[5,6]\\nosupersub{}","plainCitation":"[5,6]","noteIndex":0},"citationItems":[{"id":312,"uris":["/users/local/xIMhsWAM/items/P39KSXI5"],"uri":["/users/local/xIMhsWAM/items/P39KSXI5"],"itemData":{"id":312,"type":"article-journal","abstract":"Low-rankrecoverymodelshaveshownpotentialforsalientobjectdetection,whereamatrixisdecomposedintoalow-rankmatrixrepresentingimagebackgroundandasparsematrixidentifyingsalientobjects.Twodeficiencies,however,stillexist.First,previousworktypicallyassumestheelementsinthesparsematrixaremutuallyindependent,ignoringthespatialandpatternrelationsofimageregions.Second,whenthelow-rankandsparsematricesarerelativelycoherent,e.g.,whentherearesimilaritiesbetweenthesalientobjectsandbackgroundorwhenthebackgroundiscomplicated,itisdifficultforpreviousmodelstodisentanglethem.Toaddresstheseproblems,weproposeanovelstructuredmatrixdecompositionmodelwithtwostructuralregularizations:(1)atree-structuredsparsity-inducingregularizationthatcapturestheimagestructureandenforcespatchesfromthesameobjecttohavesimilarsaliencyvalues,and(2)aLaplacianregularizationthatenlargesthegapsbetweensalientobjectsandthebackgroundinfeaturespace.Furthermore,high-levelpriorsareintegratedtoguidethematrixdecompositionandboostthedetection.Weevaluateourmodelforsalientobjectdetectiononfivechallengingdatasetsincludingsingleobject,multipleobjectsandcomplexsceneimages,andshowcompetitiveresultsascomparedwith24state-of-the-artmethodsintermsofsevenperformancemetrics.","container-title":"IEEETransactionsonPatternAnalysis&MachineIntelligence","DOI":"10.1109/TPAMI.2016.2562626","issue":"4","page":"818-832","source":"BaiduScholar","title":"SalientObjectDetectionviaStructuredMatrixDecomposition","volume":"39","author":[{"family":"Peng","given":"Houwen"},{"family":"Li","given":"Bing"},{"family":"Ling","given":"Haibin"},{"family":"Hu","given":"Weiming"},{"family":"Xiong","given":"Weihua"},{"family":"Maybank","given":"StephenJ."}],"issued":{"date-parts":[["2017"]]}},"label":"page"},{"id":434,"uris":["/users/local/xIMhsWAM/items/9SJHRW7G"],"uri":["/users/local/xIMhsWAM/items/9SJHRW7G"],"itemData":{"id":434,"type":"article-journal","container-title":"IEEESignalProcessingLetters","issue":"8","note":"publisher:IEEE","page":"739–742","title":"Visualsaliencydetectionviasparsitypursuit","volume":"17","author":[{"family":"Yan","given":"Junchi"},{"family":"Zhu","given":"Mengyuan"},{"family":"Liu","given":"Huanxi"},{"family":"Liu","given":"Yuncai"}],"issued":{"date-parts":[["2010"]]}},"label":"page"}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[5,6]、人臉檢測ADDINZOTERO_ITEMCSL_CITATION{"citationID":"aldsb4j4uo","properties":{"formattedCitation":"\\super[7]\\nosupersub{}","plainCitation":"[7]","noteIndex":0},"citationItems":[{"id":435,"uris":["/users/local/xIMhsWAM/items/NCMGC9DJ"],"uri":["/users/local/xIMhsWAM/items/NCMGC9DJ"],"itemData":{"id":435,"type":"article-journal","container-title":"Patternrecognition","issue":"11","note":"publisher:Elsevier","page":"3502–3511","title":"Fisherdiscriminationbasedlowrankmatrixrecoveryforfacerecognition","volume":"47","author":[{"family":"Zheng","given":"Zhonglong"},{"family":"Yu","given":"Mudan"},{"family":"Jia","given":"Jiong"},{"family":"Liu","given":"Huawen"},{"family":"Xiang","given":"Daohong"},{"family":"Huang","given":"Xiaoqiao"},{"family":"Yang","given":"Jie"}],"issued":{"date-parts":[["2014"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[7]等。紡織品圖像是一種人造的紋理，雖然形態(tài)各異，結(jié)構(gòu)復(fù)雜，但是其中存在著大量的視覺冗余，其圖案由重復(fù)的紋理單元構(gòu)成，而紡織品的瑕疵是相對孤立的，是稀疏的。因此紡織品的瑕疵檢測相對于目前的顯著性檢測，更好地符合了低秩稀疏結(jié)構(gòu)分解模型的特性。但是基于低秩稀疏結(jié)構(gòu)分解的顯著性檢測算法大都是針對自然場景中的圖像，其特征提取方式重點(diǎn)考慮了色差、亮度等特性。而紡織品圖像由于存在著拉伸變形、光照等問題，不能直接應(yīng)用以往的顯著性檢測算法。因此，需要提出一種適用于紡織品瑕疵檢測領(lǐng)域的低秩稀疏結(jié)構(gòu)分解模型，實(shí)現(xiàn)背景與瑕疵區(qū)域的分離。目前根據(jù)紡織品種類的不同，可以將紡織品分為兩大類，第一類是背景簡單沒有復(fù)雜圖案的紡織品，如平紋和斜紋，第二類是背景復(fù)雜且具有周期變化圖案的紡織品，根據(jù)周期圖案的不同又可細(xì)分為盒狀，星狀和點(diǎn)狀紡織品。幾種常見的紡織品瑕疵類型如圖1-2所示，其中第一、二幅圖像為第一類紡織品圖像，后三幅為第二類紡織品圖像。第一類紡織品圖像的檢測相對簡單，目前已經(jīng)有許多成熟有效的算法。而第二類周期性紡織品圖像的瑕疵檢測則更為復(fù)雜，通常是具有花紋的紡織品，如，地毯、床罩等。Ng等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"qS0fBlu2","properties":{"formattedCitation":"\\super[8]\\nosupersub{}","plainCitation":"[8]","noteIndex":0},"citationItems":[{"id":346,"uris":["/users/local/xIMhsWAM/items/NK4DXLPZ"],"uri":["/users/local/xIMhsWAM/items/NK4DXLPZ"],"itemData":{"id":346,"type":"article-journal","abstract":"?2017SocietyforIndustrialandAppliedMathematics.Inthispaper,westudyanimagedecompositionmodelforpatternedfabricinspection.Itisimportanttorepresentfabricpatternseffectivelysothatfabricdefectscanbeseparated.Oneconcernisthatbothpatternedfabric(e.g.,star-orbox-patternedfabrics)andfabricdefectscontainmainlylowfrequencycomponents.ThemainideaofthispaperistousetheconvolutionofalatticewithaDiraccombtocharacterizeapatternedfabricimagesothatitsrepetitivecomponentscanbeeffectivelyrepresentedintheimagedecompositionmodel.Weformulateamodelwithtotalvariation,sparsity,andlow-ranktermsforpatternedfabricinspection.Thetotalvariationtermisusedtoregularizethedefectiveimage,andthesparsityandthelow-ranktermsareemployedtocontroltheDiraccombfunction.Theproposedmodelcanbesolvedefficientlyviaaconvexprogrammingsolver.Ourexperimentalresultsfordifferenttypesofpatternedfabricsshowthattheproposedmodelcaninspectdefectsatahigheraccuracycomparedwithsomeclassicalmethodsintheliterature.","container-title":"SIAMJournalonImagingSciences","DOI":"10.1137/17M1113138","issue":"4","page":"2140-2164","source":"BaiduScholar","title":"Lattice-BasedPatternedFabricInspectionbyUsingTotalVariationwithSparsityandLow-RankRepresentations","volume":"10","author":[{"family":"Ng","given":"MichaelK."},{"family":"Ngan","given":"HenryY.T."},{"family":"Yuan","given":"Xiaoming"},{"family":"Zhang","given":"Wenxing"}],"issued":{"date-parts":[["2017"]]}},"label":"page"}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[8]將組成紡織品的基本圖案稱為基本周期圖案塊(Lattice)，將可以通過自身的旋轉(zhuǎn)復(fù)制而得到基本周期圖案塊的部分稱為基礎(chǔ)單元(Motif)，如圖1-3所示。圖1-3(a)是含有周期變化圖案的周期性紡織品，分別為盒型及星型紡織品，圖1-3(b)是對應(yīng)的基本周期圖案塊，圖1-3(c)是基礎(chǔ)單元。因此對于第二類紡織品的瑕疵檢測相比于第一類將會(huì)更加復(fù)雜。本節(jié)主要介紹紡織品瑕疵檢測的國內(nèi)外研究現(xiàn)狀及其相關(guān)算法，將針對第一類和第二類紡織品兩個(gè)部分分別闡述。第一類無圖案紡織品的出現(xiàn)時(shí)間較早，目前，已經(jīng)出現(xiàn)了大量成熟有效的檢測算法。基于統(tǒng)計(jì)的方法統(tǒng)計(jì)法通過直方圖ADDINZOTERO_ITEMCSL_CITATION{"citationID":"rrSD6lbl","properties":{"formattedCitation":"\\super[9,10]\\nosupersub{}","plainCitation":"[9,10]","noteIndex":0},"citationItems":[{"id":106,"uris":["/users/local/xIMhsWAM/items/4MQPZXY6"],"uri":["/users/local/xIMhsWAM/items/4MQPZXY6"],"itemData":{"id":106,"type":"article-journal","container-title":"電子學(xué)報(bào)","issue":"8","page":"1778-1784","title":"二維直方圖準(zhǔn)分的Otsu圖像分割及其快速實(shí)現(xiàn)","volume":"39","author":[{"family":"張新明","given":""},{"family":"孫印杰","given":""},{"family":"鄭延斌","given":""}],"issued":{"date-parts":[["2011"]]}},"label":"page"},{"id":92,"uris":["/users/local/xIMhsWAM/items/Q9KZJ9MG"],"uri":["/users/local/xIMhsWAM/items/Q9KZJ9MG"],"itemData":{"id":92,"type":"article-journal","container-title":"光電子．激光","issue":"7","page":"1426-1433","source":"","title":"改進(jìn)二維直方圖區(qū)域劃分的閾值分割方法研究","volume":"24","author":[{"family":"李淼","given":""},{"family":"楊恢先","given":""},{"family":"張建波","given":""},{"family":"周彤彤","given":""},{"family":"譚正華","given":""}],"issued":{"date-parts":[["2013"]]}},"label":"page"}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[9]、形態(tài)學(xué)ADDINZOTERO_ITEMCSL_CITATION{"citationID":"zAOK7K9a","properties":{"formattedCitation":"\\super[12,13]\\nosupersub{}","plainCitation":"[12,13]","noteIndex":0},"citationItems":[{"id":199,"uris":["/users/local/xIMhsWAM/items/7VI2M6A9"],"uri":["/users/local/xIMhsWAM/items/7VI2M6A9"],"itemData":{"id":199,"type":"article-journal","container-title":"ImageandVisionComputing","DOI":"10.1016/j.imavis.2009.03.007","ISSN":"02628856","issue":"10","language":"en","page":"1585-1592","source":"Crossref","title":"Fabricdefectdetectionusingmorphologicalfilters","volume":"27","author":[{"family":"Mak","given":"K.L."},{"family":"Peng","given":"P."},{"family":"Yiu","given":"K.F.C."}],"issued":{"date-parts":[["2009",9]]}},"label":"page"},{"id":109,"uris":["/users/local/xIMhsWAM/items/VKEPCW4K"],"uri":["/users/local/xIMhsWAM/items/VKEPCW4K"],"itemData":{"id":109,"type":"article-journal","container-title":"光電工程","issue":"8","page":"90-95","source":"BaiduScholar","title":"基于小波多尺度積與形態(tài)學(xué)的瑕疵檢測算法","volume":"38","author":[{"literal":"崔玲玲"},{"literal":"盧朝陽"},{"literal":"李靜"},{"literal":"李益紅"}],"issued":{"date-parts":[["2011"]]}},"label":"page"}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[10]、共生矩陣ADDINZOTERO_ITEMCSL_CITATION{"citationID":"wG7u3dOl","properties":{"formattedCitation":"\\super[14]\\nosupersub{}","plainCitation":"[14]","noteIndex":0},"citationItems":[{"id":115,"uris":["/users/local/xIMhsWAM/items/EQ83E4FE"],"uri":["/users/local/xIMhsWAM/items/EQ83E4FE"],"itemData":{"id":115,"type":"article-journal","container-title":"計(jì)算機(jī)研究與發(fā)展","issue":"11","page":"1991-1999","title":"基于多分辨共生矩陣的紋理圖像分類","volume":"48","author":[{"literal":"鐘樺"},{"literal":"楊曉鳴"},{"literal":"焦李成"}],"issued":{"date-parts":[["2011"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[11]等計(jì)算不同紋理特征的差異性來判別背景與瑕疵ADDINZOTERO_ITEMCSL_CITATION{"citationID":"a1lihbnekpb","properties":{"formattedCitation":"\\super[15]\\nosupersub{}","plainCitation":"[15]","noteIndex":0},"citationItems":[{"id":436,"uris":["/users/local/xIMhsWAM/items/5SHW5PUK"],"uri":["/users/local/xIMhsWAM/items/5SHW5PUK"],"itemData":{"id":436,"type":"article-journal","container-title":"TheJournalofTheTextileInstitute","issue":"2","note":"publisher:Taylor&Francis","page":"233–241","title":"Detectionofchemicalfabricdefectsonthebasisofmorphologicalprocessing","volume":"107","author":[{"family":"Hu","given":"Jiajuan"},{"family":"He","given":"Zhiyong"},{"family":"Weng","given":"Guirong"},{"family":"Sun","given":"Lining"},{"family":"Zuo","given":"Baoqi"},{"family":"Wang","given":"Chen"}],"issued":{"date-parts":[["2016"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[12]。Rebhi等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"a1lihbnekpb","properties":{"formattedCitation":"\\super[15]\\nosupersub{}","plainCitation":"[15]","noteIndex":0},"citationItems":[{"id":436,"uris":["/users/local/xIMhsWAM/items/5SHW5PUK"],"uri":["/users/local/xIMhsWAM/items/5SHW5PUK"],"itemData":{"id":436,"type":"article-journal","container-title":"TheJournalofTheTextileInstitute","issue":"2","note":"publisher:Taylor&Francis","page":"233–241","title":"Detectionofchemicalfabricdefectsonthebasisofmorphologicalprocessing","volume":"107","author":[{"family":"Hu","given":"Jiajuan"},{"family":"He","given":"Zhiyong"},{"family":"Weng","given":"Guirong"},{"family":"Sun","given":"Lining"},{"family":"Zuo","given":"Baoqi"},{"family":"Wang","given":"Chen"}],"issued":{"date-parts":[["2016"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[13]提出了一種基于局部同質(zhì)性和數(shù)學(xué)形態(tài)學(xué)的紡織品檢測新方法，仿真結(jié)果顯示，該方法能準(zhǔn)確檢測出瑕疵區(qū)域。Jing等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"a1lihbnekpb","properties":{"formattedCitation":"\\super[15]\\nosupersub{}","plainCitation":"[15]","noteIndex":0},"citationItems":[{"id":436,"uris":["/users/local/xIMhsWAM/items/5SHW5PUK"],"uri":["/users/local/xIMhsWAM/items/5SHW5PUK"],"itemData":{"id":436,"type":"article-journal","container-title":"TheJournalofTheTextileInstitute","issue":"2","note":"publisher:Taylor&Francis","page":"233–241","title":"Detectionofchemicalfabricdefectsonthebasisofmorphologicalprocessing","volume":"107","author":[{"family":"Hu","given":"Jiajuan"},{"family":"He","given":"Zhiyong"},{"family":"Weng","given":"Guirong"},{"family":"Sun","given":"Lining"},{"family":"Zuo","given":"Baoqi"},{"family":"Wang","given":"Chen"}],"issued":{"date-parts":[["2016"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[14]提出了基于局部二值模式(LBP)和灰度共生矩陣的紡織品瑕疵分類算法，取得了99.33%的分類效果。Zhou等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"a1lihbnekpb","properties":{"formattedCitation":"\\super[15]\\nosupersub{}","plainCitation":"[15]","noteIndex":0},"citationItems":[{"id":436,"uris":["/users/local/xIMhsWAM/items/5SHW5PUK"],"uri":["/users/local/xIMhsWAM/items/5SHW5PUK"],"itemData":{"id":436,"type":"article-journal","container-title":"TheJournalofTheTextileInstitute","issue":"2","note":"publisher:Taylor&Francis","page":"233–241","title":"Detectionofchemicalfabricdefectsonthebasisofmorphologicalprocessing","volume":"107","author":[{"family":"Hu","given":"Jiajuan"},{"family":"He","given":"Zhiyong"},{"family":"Weng","given":"Guirong"},{"family":"Sun","given":"Lining"},{"family":"Zuo","given":"Baoqi"},{"family":"Wang","given":"Chen"}],"issued":{"date-parts":[["2016"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[15]將五個(gè)分形特征與模糊C均值相結(jié)合，在標(biāo)準(zhǔn)數(shù)據(jù)庫上，取得了較低的誤檢率。Conci等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"dwM7fQvp","properties":{"formattedCitation":"\\super[16]\\nosupersub{}","plainCitation":"[16]","noteIndex":0},"citationItems":[{"id":316,"uris":["/users/local/xIMhsWAM/items/54UCNA2U"],"uri":["/users/local/xIMhsWAM/items/54UCNA2U"],"itemData":{"id":316,"type":"article-journal","abstract":"Industrialvisionsystemsmustoperateinreal-time,producealowfalsealarmrateandbeflexiblesoastoaccommodatechangesinthemanufacturingprocesseasily.Thisworkpresentsasystemforfabricmanufacturinginspection.Thisenvironment,likepaperandbirchwoodboardindustries,hasparticularcharacteristicsinwhichmorphologicalfeatureextractionforautomatedvisualinspectioncannotbeused.Theutilizationoffractaldimensionisinvestigatedfordiscriminatingdefectiveareas.Theefficiencyofthisapproachisillustratedintextileimagesfordefectrecognition(withoverall96%accuracy).Whilethismaysoundcomplex,themethodisinfactsimpleenoughtobesuitableforPCimplementation,asdemonstratedinthepresentwork,andutilizationacrosstheWordWideWeb","container-title":"ComputerNetworks&IsdnSystems","DOI":"10.1016/s0169-7552(98)00211-6","issue":"20-21","page":"1887-1895","source":"BaiduScholar","title":"Afractalimageanalysissystemforfabricinspectionbasedonabox-countingmethod","volume":"30","author":[{"family":"Conci","given":"Aura"},{"literal":"ClaudiaBelmiroProen?a"}],"issued":{"date-parts":[["1998"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[16]采用分形的思想對紡織品進(jìn)行檢測，并假設(shè)當(dāng)分形的結(jié)構(gòu)發(fā)生變化時(shí)含有瑕疵?；陬l譜分析的方法頻譜分析法將紡織品圖像從空間域變換至變換域，通過在變換域更好地突出瑕疵區(qū)域。頻譜分析法比較有代表性的理論有傅里葉變換、Gabor變換及小波變換等。Sarma等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"a1ak63c3db4","properties":{"formattedCitation":"\\super[17]\\nosupersub{}","plainCitation":"[17]","noteIndex":0},"citationItems":[{"id":437,"uris":["/users/local/xIMhsWAM/items/EJR898CP"],"uri":["/users/local/xIMhsWAM/items/EJR898CP"],"itemData":{"id":437,"type":"paper-conference","container-title":"2013InternationalConferenceonAdvancedElectronicSystems(ICAES)","page":"157–159","publisher":"IEEE","title":"Detectingthesurfacedefectsonhotrolledsteelsheetsusingtextureanalysis","author":[{"family":"Sarma","given":"ASadaSiva"},{"family":"Janani","given":"R"},{"family":"Sarma","given":"ASV"}],"issued":{"date-parts":[["2013"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[17]采用三層二維小波變換來提取紡織品的紋理特征，并通過訓(xùn)練人工神經(jīng)網(wǎng)絡(luò)(ANN)分類器來檢測熱軋鋼表面的瑕疵?；谟?xùn)練的方法訓(xùn)練法多采用神經(jīng)網(wǎng)絡(luò)算法，通過構(gòu)建多層的網(wǎng)絡(luò)來訓(xùn)練參數(shù)，在將訓(xùn)練出的參數(shù)作為閾值來判斷是否為瑕疵塊。Zhang等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"WLzxsJ27","properties":{"formattedCitation":"\\super[18]\\nosupersub{}","plainCitation":"[18]","noteIndex":0},"citationItems":[{"id":356,"uris":["/users/local/xIMhsWAM/items/GVDB2NPB"],"uri":["/users/local/xIMhsWAM/items/GVDB2NPB"],"itemData":{"id":356,"type":"article-journal","abstract":"Inthispaper,anewapproachforfabricdefectclassificationusingradialbasisfunction(RBF)networkimprovedbyGaussianmixturemodel(GMM)isinvestigated.First,thegraylevelarrangementintheneighborhoodofeachpixelisextractedasthefeature.Thisrawfeatureissubjecttoprincipalcomponentanalysis(PCA)whichadoptsthebetweenclassscattermatrixasthegenerationmatrixtoeliminatethevariancewithinthesameclass.Second,theRBFnetworkwithGaussiankernelisusedastheclassifierbecauseofthenonlineardiscriminationabilityandsupportformulti-output.Totraintheclassifier,GMMisintroducedtoclusterthefeaturesetandpreciselyestimatetheparameterinGaussianRBF,inwhicheachclusterstrictlyconformstoamulti-varianceGaussiandistribution.ThustheparameterofeachkernelfunctioninRBFnetworkcanbeacquiredfromacorrespondingcluster.Theproposedalgorithmisexperimentedonfabricdefectimageswithnineclassesandachievessuperiorperformance,whichprovesitsutilityinpractice.","collection-title":"Meta-heuristicIntelligenceBasedImageProcessing","container-title":"PatternRecognitionLetters","DOI":"10.1016/j.patrec.2010.05.030","ISSN":"0167-8655","issue":"13","journalAbbreviation":"PatternRecognitionLetters","language":"en","page":"2033-2042","source":"ScienceDirect","title":"Fabricdefectclassificationusingradialbasisfunctionnetwork","volume":"31","author":[{"family":"Zhang","given":"Yu"},{"family":"Lu","given":"Zhaoyang"},{"family":"Li","given":"Jing"}],"issued":{"date-parts":[["2010",10,1]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[18]研究了一種利用高斯混合模型(GMM)改進(jìn)的徑向基函數(shù)(RBF)網(wǎng)絡(luò)進(jìn)行紡織品瑕疵點(diǎn)分類的新方法?；谀Ｐ偷姆椒ɑ谀Ｐ偷姆椒ㄖ饕R爾科夫隨機(jī)場模型ADDINZOTERO_ITEMCSL_CITATION{"citationID":"a1qae9pcbeq","properties":{"formattedCitation":"\\super[19]\\nosupersub{}","plainCitation":"[19]","noteIndex":0},"citationItems":[{"id":200,"uris":["/users/local/xIMhsWAM/items/VB6LRANG"],"uri":["/users/local/xIMhsWAM/items/VB6LRANG"],"itemData":{"id":200,"type":"article-journal","container-title":"IEEETransactionsonPatternAnalysisandMachineIntelligence","DOI":"10.1109/TPAMI.2004.30","ISSN":"0162-8828","issue":"7","language":"en","page":"951-955","source":"Crossref","title":"GaussianMRFrotation-invariantfeaturesforimageclassification","volume":"26","author":[{"literal":"HuawuDeng"},{"family":"Clausi","given":"D.A."}],"issued":{"date-parts":[["2004",7]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[19]，高斯混合模型及自回歸模型。Ozdemir等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"NovxlewI","properties":{"formattedCitation":"\\super[20]\\nosupersub{}","plainCitation":"[20]","noteIndex":0},"citationItems":[{"id":333,"uris":["/users/local/xIMhsWAM/items/GRY8HV6L"],"uri":["/users/local/xIMhsWAM/items/GRY8HV6L"],"itemData":{"id":333,"type":"article-journal","container-title":"紡織學(xué)報(bào)","issue":"4","page":"137-142","source":"","title":"基于GMRF模型的統(tǒng)計(jì)特征畸變織物疵點(diǎn)識(shí)別","author":[{"family":"楊曉波","given":""}],"issued":{"date-parts":[["2013"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[20]以馬爾科夫隨機(jī)場模型為基礎(chǔ)，結(jié)合KL變換進(jìn)行紡織品的瑕疵檢測，大大提高了檢測效率。Basu等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"ReIUZ7oN","properties":{"formattedCitation":"\\super[21]\\nosupersub{}","plainCitation":"[21]","noteIndex":0},"citationItems":[{"id":408,"uris":["/users/local/xIMhsWAM/items/PEQK3GGC"],"uri":["/users/local/xIMhsWAM/items/PEQK3GGC"],"itemData":{"id":408,"type":"paper-conference","abstract":"Considersaspecificclassoftextureswhicharestochastic,possiblyperiodic,two-dimensionalsignalsdisplayingfractal-likeorself-similarcharacteristics.Mostnaturaltexturesbelongtothisclass.Theauthorsexploretheuseofautoregressive(AR)processesontreesastexturemodel.ThistheorywasproposedbyBassevilleetal.(1992)formultiscalesignalanalysis.ThegeneralizedlatticestructuresusedforparametrizationofARprocessesontreesmakecomputerimplementationfastandefficient.Theauthorshavedoneextensiveexperimentsontexturegenerationandstudytheeffectofreflectioncoefficientsandmodelorderonthequalityofgeneratedtextures","container-title":"IaprInternationalConferenceonPatternRecognitionVolIIIConferenceC:Image","DOI":"10.1109/ICPR.1992.202013","source":"BaiduScholar","title":"Multi-scalemodelingoftextures","URL":"/publication/3513942_Multi-scale_modeling_of_textures","author":[{"family":"Basu","given":"M."},{"family":"Lin","given":"Z."}],"accessed":{"date-parts":[["2020",1,4]]},"issued":{"date-parts":[["1992"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[21]提出了一種多尺度自回歸模型，考慮一類特殊的紋理，這些紋理是隨機(jī)的，可能是周期性的，顯示分形或自相似特征的二維信號(hào)，在花紋類圖案上取得了一定的效果。1.3.2周期性紡織品瑕疵檢測算法上述的瑕疵檢測算法雖然在一定程度上取得了較好的檢測結(jié)果，但基本上都是針對簡單的平紋、斜紋類紡織品。第二類周期性紡織品因其紋理的復(fù)雜，成熟的檢測算法較少，但經(jīng)過近幾年的快速發(fā)展，已經(jīng)形成了多種檢測方法，可以分為模板法、統(tǒng)計(jì)法和圖像分解法等。模板法在模板法中，傳統(tǒng)的圖像相減法(Traditionalimagesubtraction，TIS)是一種常用的方法。Ngan等人ADDINZOTERO_ITEMCSL_CITATION{"citationID":"a2q362fn5au","properties":{"formattedCitation":"\\super[22]\\nosupersub{}","plainCitation":"[22]","noteIndex":0},"citationItems":[{"id":439,"uris":["/users/local/xIMhsWAM/items/VTZIRF99"],"uri":["/users/local/xIMhsWAM/items/VTZIRF99"],"itemData":{"id":439,"type":"paper-conference","container-title":"32ndAppliedImageryPatternRecognitionWorkshop,2003.Proceedings.","page":"163–168","publisher":"IEEE","title":"Defectdetectiononpatternedjacquardfabric","author":[{"family":"Ngan","given":"HenryYT"},{"family":"Pang","given":"GranthamKH"},{"family":"Yung","given":"SP"},{"family":"Ng","given":"MichaelK"}],"issued":{"date-parts":[["2003"]]}}}],"schema":"/citation-style-language/schema/raw/master/csl-citation.json"}[22]在TIS的基礎(chǔ)上提出了一種黃金圖像相減法(Goldenimagesubtraction，GIS)，其中一個(gè)黃金塊就稱為一個(gè)模板。然后又在GIS的基礎(chǔ)上，提出一種基于小波預(yù)處理的黃金圖像相減法(WGIS)ADDINZOTERO_ITEMCSL_CITATION{"citationID":"kpLSAucQ","properties":{"formattedCitation":"\\super[23]\\nosupersub{}","plainCitation":"[23]","noteIndex":0},"citationItems":[{"id":124,"uris":["/users/local/xIMhsWAM/items/58P4KFWD"],"uri":["/users/local/xIMhsWAM/items/58P4KFWD"],"itemData":{"id":124,"type":"article-journal","container-title":"PatternRecognition","issue":"4","page":"559-576","title":"Waveletbasedmethodsonpatternedfabricdefectdete