搜索
I木材表面缺陷图像识别的算法研究II摘要随着木材加工业的集约化发展，木材产品的生产量持续大幅度增长。在生产中，对木材表面加工质量高水平的苛求，尤其是一致性的要求，使得传统的人工检测方式已经难以胜任。为此，本论文基于机器视觉理论对木材表面缺陷识别进行了深入研究。结合数字图像处理技术和支持向量机模式识别技术，本论文研究了木材表面缺陷图像预处理、特征提取、模式识别问题，研究并改进了用于检测木材表面缺陷的定位和识别等图像处理算法。图像的预处理是检测的第一步，它对图像缺陷特征的正确提取是非常关键的。本文针对传统滤波算法在抑制噪声的同时，也会对图像的边缘及细节有比较大的损害，使图像的边沿及细节变模糊的问题，提出了加权有向平滑滤波算法。并在图像分割上融合了几种分割方法，提出一种改进的基于双正交小波变换的多分辨率图像融合方法和基于融合技术的小波变换和形态学边缘检测算法，优化了分割效果，为后续特征提取打下了很好的基础。对于木材缺陷的识别，本文从纹理特征（5个灰度共生矩阵参数）和颜色特征（4个颜色矩参数）两个角度来描述缺陷。根据各参数分布情况，选择标准差较小的参数作为分类器输入特征向量；以及采用主分量分析法进行特征提取，降低纹理特征维数，消除模式特征之间的相关性，突出其差异性，满足识别层的输入要求。并采用支持向量机分类器进行缺陷的模式识别，达到较高的识别率。实验结果证明：根据木材表面缺陷图像的纹理特征和颜色特征，运用数字图像处理技术，来解决木材表面缺陷的分割和识别等问题，是行之有效的途径。关键词：数字图像处理技术；图像分割；特征提取；支持向量机IIIAbstractWiththedevelopmentofwoodindustry,themanufactureofwoodproductsisincreasingsignificantly.Thedemandofaconsistenthigh-qualitysurfacewoodproductintroducesautomaticinspectionthatcannotbeeasilysatisfiedbytraditionalmanualinspection.Basedonthetheoryofcomputervision,aresearchondefectdistinguishofthewoodsurfaceismadeinthepaper.Imagepreprocess,featureextractionandpatternrecognitionofwoodsurfacedefectimagesarealsostudiedbymeansofdigitalimageprocessingtechniqueandpatternrecognitiontechnologybasedonSVM(SupportVectorMachines).Imageprocessingalgorithmsarestudiedandimprovedtoorientateandrecognizewoodsurfacedefect.Imagepreprocessisthefirststepfordetection,whichisvitaltothecorrectextractionofthedefectionfeature.Inthefactofatraditionalfilteringalgorithmcansubstantiallydamagetheedgesanddetailsoftheimageandblurtheimage’sedgesanddetails,aweightedanddirectionalsmoothingalgorithmisproposedinthispaper.Mergingseveralimagesegmentationmethod,aimprovedmethodofimagefusionofmulti-resolutionanalysisbasedonbiorthogonalwavelettransformandaedgedetectionalgorithmbasedonthefusiontechnologyofwavelettransformandmorphologicaledgedetectionareproposedinthepaper.Thussegmentationresultisoptimizedandlayingtherootforfeatureextractionoffollowup.Thedefectsaredescribedfromtwoaspectsbasedonimagecharacteristic,thetexturefeatures(fivegraylevelco-occurrencematrixparameters)andcolorfeatures（fourcolormomentparameters）toidentifythewooddefects.Accordingtothedistributionoftheseparameters,theparameterswhichhavesmallstandarddeviationareselectedastheinputeigenvectoroftheclassifiers.Andthefeaturesareextractedbytheprincipalcomponentsanalysiswhichcanreducethetexturedimensionsandeliminatetherelevancebetweenfeaturemodesandhighlighttheirdifferencetosatisfytheinputrequestoftherecognitionlevel.UsingSupportVectorMachinesclassifiertoidentifythedefects,thecorrectratesofpatternrecognitionachievebetterlevel.Theexperimentresultsshowitisaneffectivewaytosolvethesegmentationandidentificationofwoodsurfacedefectsbytexturefeaturesandcolorfeaturesofwoodsurfacedefectimagesaccordingtothedigitalimageprocessingtechnology,.Keyword：digitalimageprocessingtechnique；imagesegmentation；featureextraction；SVM(SupportVectorMachines)I目录第一章绪论...................................................................................................................-1-1.1课题的研究背景和意义.....................................................................................-1-1.1.1课题的研究背景......................................................................................-1-1.1.2课题的研究意义......................................................................................-1-1.2木材表面缺陷检测的研究现状及发展趋势........................................................-2-1.2.1木材缺陷的常用检测方法.......................................................................-2-1.2.2国内外研究现状......................................................................................-3-1.2.3木材检测技术的发展与展望....................................................................-4-1.3木材表面缺陷特征及存在形式..........................................................................-5-1.3.1木材缺陷种类..........................................................................................-5-1.3.2木材缺陷对木材质量的影响....................................................................-8-1.4课题的主要研究内容和创新..............................................................................-8-第二章木材表面缺陷图像的增强预处理......................................................................-11-2.1图像增强概述..................................................................................................-11-2.2木材缺陷图像灰度变换...................................................................................-12-2.2.1木材缺陷图像灰度化处理.....................................................................-12-2.2.2木材缺陷图像灰度变换.........................................................................-13-2.3木材缺陷图像平滑..........................................................................................-16-2.3.1邻域平滑...............................................................................................-16-2.3.2中值滤波...............................................................................................-16-2.3.3加权有向平滑滤波................................................................................-17-2.4图像锐化................................................................