玻璃缺陷图像识别的关键技术研究
|
摘要
在当今社会,玻璃被广泛的应用于人们的日常生活中。此外,玻璃作为一种生产材料在制造业中也起着重要的作用。然而,玻璃的生产流程中受工艺和环境限制,会产生各种缺陷,这些缺陷会破坏玻璃原有的机械性能和热稳定性,从而对人们的日常生活及工业加工造成不好的影响。目前,绝大多数的玻璃生产厂商依靠传统的人工检验方法对缺陷玻璃予以发现和控制,但由于受到工人易于疲劳、鉴别经验不足等主客观因素的束缚,检验效果并不理想。正是在这种研究背景下,设计并实现一种替代人工检验的玻璃缺陷检测技术是很有必要的。
论文重点研究了模式识别技术在玻璃缺陷检测行业里的应用与发展。通过对模式识别相关技术的研究,设计并实现一个检测性能较好的玻璃缺陷检测系统,以便对玻璃缺陷进行自动分类。论文主要工作如下:
(1)针对玻璃图像对比度低,边缘模糊等特点,对目前主流的图像预处理算法进行了比较实验,最终选定了三种适用于玻璃图像的预处理方法,包括:灰度化处理、空间线性变换以及二值化方法。
(2)研究了小波、小波包变换的原理与编码方法,并将不变矩技术(包括Hu's矩、Zernike's矩、小波矩)引入到玻璃图像的特征提取中。此外,针对当前小波矩算法误差较大的缺点,利用两次坐标变换设计并实现了一种精确的小波矩算法(E-WMI),取得了较好的实验效果。
(3)研究了支持向量机的基础理论与实现方法。针对当前支持向量机中增量学习算法的主要缺陷,设计了一种改进的对称增量学习算法(S-ISVM),并应用到论文的玻璃缺陷识别中。
(4)用VC与MATLAB混合编程技术完成了玻璃缺陷检测系统中模式识别模块的核心代码编写与实验平台搭建,并通过大量的实验数据证明:在相同识别技术的条件下,用论文提出的E-WMI对图像进行特征提取所得到的识别精度最高。此外,使用论文提出的S-ISVM方法对新增样本进行训练可以在保证识别精度的基础上,有效缩短训练时间。
论文在分析比较了目前主流的图像预处理算法、特征提取算法与识别分类方法基础上,针对玻璃缺陷图像的特点,设计了相对适用的图像预处理、特征提取及分类识别技术,并搭建了玻璃缺陷检测平台。
Glass is an important and widely used material in producing and living of the contemporary society.Limited by the crafts and environment,its manufacturing process is apt to generate defect.The defect not only undermines the glass's mechanical performance but the thermal stability.So,it does a bad affection for human's daily life and the industrial production.Recently,most glass producers depend on traditional manual method for monitoring and controlling the defective glass.Because of both the subjective and objective factors,the recognition result is bad.In this background,it becomes necessary to design and realize an automatic recognition method for defective glass image.This thesis researches related Pattern Recognition technologies in the glass image field.In conclusion,it has important significance and research value to realize an automatic glass defective recognition system.The main work of this thesis contains as follow:
(1) According to the characteristic of the glass images,a lot of experiments have been done to determine the advisable image pre-processing algorithm such as gray processing,space linear transformation and image binaryzation
(2) Researched the referred theories on wavelet and wavelet packet,and used the moments invariants(including Hu's moment,Zernike's moments and wavelet moment)to the glass image feature extracting.Besides,designed a new exact wavelet moments algorithm(E-WMI) to solve the computing errors in the present algorithms.
(3) Researched the basic theories on SVM and designed a classifier based on it to recognize the defective glass images.According to the disadvantage of present increment training methods,designed a new Symmetrical increment method(S-ISVM) to train the increment samples.
(4) Finished the mainly sub-modules with VC++ and MATLAB for the glass defective recognition system and collected these modules to a testing platform.After lots of experiment,it proved that using the methods designed in this thesis has the better performance in the glass defective recognition.The new exact wavelet moments have the better recognition rate compared with other feature extracting methods and the S-ISVM can improve the recognition efficiency.
The thesis compares kinds of image pre-processing,feature extracting arithmetic in terms of theory,coding,and results and designs the optimal methods regarding image analysis in glass defect.
论文重点研究了模式识别技术在玻璃缺陷检测行业里的应用与发展。通过对模式识别相关技术的研究,设计并实现一个检测性能较好的玻璃缺陷检测系统,以便对玻璃缺陷进行自动分类。论文主要工作如下:
(1)针对玻璃图像对比度低,边缘模糊等特点,对目前主流的图像预处理算法进行了比较实验,最终选定了三种适用于玻璃图像的预处理方法,包括:灰度化处理、空间线性变换以及二值化方法。
(2)研究了小波、小波包变换的原理与编码方法,并将不变矩技术(包括Hu's矩、Zernike's矩、小波矩)引入到玻璃图像的特征提取中。此外,针对当前小波矩算法误差较大的缺点,利用两次坐标变换设计并实现了一种精确的小波矩算法(E-WMI),取得了较好的实验效果。
(3)研究了支持向量机的基础理论与实现方法。针对当前支持向量机中增量学习算法的主要缺陷,设计了一种改进的对称增量学习算法(S-ISVM),并应用到论文的玻璃缺陷识别中。
(4)用VC与MATLAB混合编程技术完成了玻璃缺陷检测系统中模式识别模块的核心代码编写与实验平台搭建,并通过大量的实验数据证明:在相同识别技术的条件下,用论文提出的E-WMI对图像进行特征提取所得到的识别精度最高。此外,使用论文提出的S-ISVM方法对新增样本进行训练可以在保证识别精度的基础上,有效缩短训练时间。
论文在分析比较了目前主流的图像预处理算法、特征提取算法与识别分类方法基础上,针对玻璃缺陷图像的特点,设计了相对适用的图像预处理、特征提取及分类识别技术,并搭建了玻璃缺陷检测平台。
Glass is an important and widely used material in producing and living of the contemporary society.Limited by the crafts and environment,its manufacturing process is apt to generate defect.The defect not only undermines the glass's mechanical performance but the thermal stability.So,it does a bad affection for human's daily life and the industrial production.Recently,most glass producers depend on traditional manual method for monitoring and controlling the defective glass.Because of both the subjective and objective factors,the recognition result is bad.In this background,it becomes necessary to design and realize an automatic recognition method for defective glass image.This thesis researches related Pattern Recognition technologies in the glass image field.In conclusion,it has important significance and research value to realize an automatic glass defective recognition system.The main work of this thesis contains as follow:
(1) According to the characteristic of the glass images,a lot of experiments have been done to determine the advisable image pre-processing algorithm such as gray processing,space linear transformation and image binaryzation
(2) Researched the referred theories on wavelet and wavelet packet,and used the moments invariants(including Hu's moment,Zernike's moments and wavelet moment)to the glass image feature extracting.Besides,designed a new exact wavelet moments algorithm(E-WMI) to solve the computing errors in the present algorithms.
(3) Researched the basic theories on SVM and designed a classifier based on it to recognize the defective glass images.According to the disadvantage of present increment training methods,designed a new Symmetrical increment method(S-ISVM) to train the increment samples.
(4) Finished the mainly sub-modules with VC++ and MATLAB for the glass defective recognition system and collected these modules to a testing platform.After lots of experiment,it proved that using the methods designed in this thesis has the better performance in the glass defective recognition.The new exact wavelet moments have the better recognition rate compared with other feature extracting methods and the S-ISVM can improve the recognition efficiency.
The thesis compares kinds of image pre-processing,feature extracting arithmetic in terms of theory,coding,and results and designs the optimal methods regarding image analysis in glass defect.
引文
[1]G.Mallat.A theory of multi-resolution signal decomposition:the wavelet representation.IEEE Transactions on Pattern Analysis and Machine Intelligence.1989,11(7):674-693
[2]Chui C K,Wang J Z.A Cardinal Spline Approach to Wavelets.American Mathematical Society.1991,113(3):785-793
[3]Hu M.Visual Pattern Recognition by Moment Invariant.IRE Transactions on Information Theory.1961,49:179-187
[4]Khotanzad.Invariant Image Recognition by Zernike's Moments.IEEE Transactions on Pattern Analysis and Machine Intelligence.1990,12(5):489-497
[5]Shen,Horace H S Ip.Discriminative Wavelet Shape Descriptors for Recognition of 2-D Patterns.Pattern Recognition.1999,32:151-165
【6]范晓峰,施泽生.基于小波矩的新型图形识别算法.计算机工程与应用.2001,7:47-49
【7]王其聪.基于小波分析的矩特征和神经网络的图像识别【硕士学位论文].浙江工业大学,2004
[8]Hong Pan,Liangzheng Xia.Exact and Fast Algorithm for Two-Dimensional Image Wavelet Moments via the Projection Transform Pattern Recognition.2005,38:395-402
[9]Siu-Yeung Cho.Genetic Evolution Processing of Data Structures for Image Classification.IEEE Transactions on Knowledge and Data Engineering,2005,17(2)
【10]边肇祺,张学工.模式识别(第二版).北京:清华大学出版社.ISBN:730201059
【11]严红平.模式识别简述.自动化博览.2006,2:22-26
【12]李海啸.基于矩和BP神经网络的纹理分割【硕士学位论文].东南大学,2005
【13]彭淑敏.神经网络图像识别技术研究与实现【硕士学位论文].西安电子科技大学,2005
[14]Le Dung,Makoto Mizukawa.A Pattern Recognition Neural Network Using Many Sets of Weights and Biases.Proceedings of the 2007 IEEE International Symposium on Computational Intelligence in Robotics and Automation Jacksonville,FL,USA,June 20-23,2007:285-290
[15]Vladimir N.Vapnik V.The Nature of Statistical Learning Theory.New York:Springer-Verlag,Inc.1995 ISBN:0-387-94559-8
[16]Colin Campbell.Algorithmic Approaches to Training Support Vector.Department of Engineering Mathematics,Bristol University
【17]陆波,尉询楷.支持向量机在分类中的应用.中国图像图形学报.2005,10(8):1029-1035
【18]王立国,张晔.支持向量机多类目标分类器的结构简化研究.中国图像图形学报.2005,10(5):571-574
【19]顾磊,吴慧中.基于新的决策规则的球形支持向量机分类算法.系统仿真学报.2008,20(11):2901-2904
[20]Syed N,Liu H,Sung K.Incremental learning with support vector machines.Proceedings of the Workshop on Support Vector Machines at the International Joint Conference on Artificial Intelligence.Stockholm,Sweden:Morgan Kaufmann,1999:876-892.
【21]萧嵘,王继成,孙正兴,张福炎.一种SVM增量学习算法α-ISVM.软件学报.2001,12(12):1818-1824
【22]孔锐,张冰.一种快速支持向量机增量学习算法.控制与决策.2005,20(10):1129-1132
【23]毛建洋,黄道.一种新的支持向量机增量算法.华东理工大学学报(自然科学版).2006,32(8):989-991
[24]王晓丹,郑春颖.一种新的SVM对等增量学习算法.计算机应用.2006,26(10):2440-2443
[25]Mitra S K,Parker J M.Expert System for Identifying Defects:Stones and Cord in Glass.proceeding of the Sixth European Conference on Artificial Intelligence.1984.:3625-3637
[26]Sehmiedl.R.Automation of Flaw Detection on the Float Glass Ribbon.Glass.1985,62(11):408-409
[27]Latham.V,Nixon.M,Mayars.C.Automatic Optical Inspection of Table Glass.Glass Technology,1986,27(6):188-194
[28]Kysztof.J.Cios,RobertE.Tjia,NingLiuete.Study of Continuous ID3 and Radial Basis Function Algorithms for the Recognition of Glass Defects.0-7803-0164-1/91/0000-0049$01.0001991 IEEE.49-54
【29]李巧先.关于玻璃缺陷在线检测技术的研究.上海计量测试.1999,4:26-28
【30]马竹梧,白凤双.平板玻璃生产监控及质量缺陷诊断专家系统.基础自动化.2001,12:1-27
【31]刘刚.计算机图像处理技术在玻璃按等级分检中的应用【硕士学位论文].河南科技大学,2003
【32]冯燕,何明一,俞华憬.基于图像处理的玻璃均匀性检测系统.计算机应用.1999,4:64-66
[33]刘力双,王宝光,张挑.镀膜玻璃的瑕疵在线检测系统实验研究.制造业自动化.2003, 25(10):52-55
【34]陈作炳,连迅,马荣海.基于ARM与DSP玻璃缺陷在线检测系统.电子测量技术.2006,4:55-60
【35]肖燕.基于数字图像识别技术的玻璃缺陷在线检测实验系统研究【硕士学位论文].郑州大学,2007
[36]杨淑莹.VC++图像处理程序设计.北京:清华大学出版社.ISBN:7-81082-163-6
【37]张同燕.基于图像处理技术的玻璃缺陷识别方法研究【硕士学位论文].郑州大学,2007
【38]王平顺.图像处理技术在玻璃缺陷检测中的应用研究【硕士学位论文].燕山大学,2005
【39]贾沛.特征选择技术研究【硕士学位论文].华中科技大学,2003
[40]Sangho Yoon,Robert M Gray.Feature Selection Based on Maximizing Separability in Gauss Mixture Model and its Application to Image Classification.0-7803-9134-9/05/$20.00(?)2005 IEEE.487-491
【41]杨蕊红.小波矩算法在图像识别中的应用研究【硕士学位论文].西北工业大学,2003
【42]秦磊.基于矩的图像分析和快速算法【硕士学位论文].东南大学,2003
[43]RuiHong Yang.The Application of Wavelet Invariant Moments to Image Recognition.1-4244-0060-0/06/$20.00 2006 IEEE.3243-3247
[44]Vapnik V,Levin E,Le Cun Y.Measuring the VC Dimension of a Learning Machine.Neural Computation.1994,8(5):51-76.
【45]田园.支持向量机训练算法的研究与优化【硕士学位论文].西安理工大学,2008
[46]Bugres C J C.A Tutorial on Support Vector Machines for Pattern Recognition.Data Mining and Knowledge Discovery.1998 2(2):121-167
【47]肖靓.基于支持向量机的图像分类研究【硕士学位论文].同济大学,2006
【48]王炜,郭小明.关于核函数选取的方法.辽宁师范大学学报(自然科学版).2008,31(1):1-4
【49]黄琼英.支持向量机对分类算法的应用研究【硕士学位论文].河北工业大学,2005
[50]Platt J C,Cristianini N,Shawe Taylor J.Large Margin DAGs for Mutticlass Classification.Advances in Neural Information Processing Systems.2000,12(12).:547-553
[51]Pairwise.Classification and Support Vector Machines.Cambridge:The MIT Press,1999.
[52]萧嵘,王继成.一种SVM增量学习算法南京大学学报(自然科学).2002,38(2):152-157
[53]王建华,宋永胜.基于边界向量的支持向量机增量算法.计算机工程与应用.2007年43(33):173-177
[2]Chui C K,Wang J Z.A Cardinal Spline Approach to Wavelets.American Mathematical Society.1991,113(3):785-793
[3]Hu M.Visual Pattern Recognition by Moment Invariant.IRE Transactions on Information Theory.1961,49:179-187
[4]Khotanzad.Invariant Image Recognition by Zernike's Moments.IEEE Transactions on Pattern Analysis and Machine Intelligence.1990,12(5):489-497
[5]Shen,Horace H S Ip.Discriminative Wavelet Shape Descriptors for Recognition of 2-D Patterns.Pattern Recognition.1999,32:151-165
【6]范晓峰,施泽生.基于小波矩的新型图形识别算法.计算机工程与应用.2001,7:47-49
【7]王其聪.基于小波分析的矩特征和神经网络的图像识别【硕士学位论文].浙江工业大学,2004
[8]Hong Pan,Liangzheng Xia.Exact and Fast Algorithm for Two-Dimensional Image Wavelet Moments via the Projection Transform Pattern Recognition.2005,38:395-402
[9]Siu-Yeung Cho.Genetic Evolution Processing of Data Structures for Image Classification.IEEE Transactions on Knowledge and Data Engineering,2005,17(2)
【10]边肇祺,张学工.模式识别(第二版).北京:清华大学出版社.ISBN:730201059
【11]严红平.模式识别简述.自动化博览.2006,2:22-26
【12]李海啸.基于矩和BP神经网络的纹理分割【硕士学位论文].东南大学,2005
【13]彭淑敏.神经网络图像识别技术研究与实现【硕士学位论文].西安电子科技大学,2005
[14]Le Dung,Makoto Mizukawa.A Pattern Recognition Neural Network Using Many Sets of Weights and Biases.Proceedings of the 2007 IEEE International Symposium on Computational Intelligence in Robotics and Automation Jacksonville,FL,USA,June 20-23,2007:285-290
[15]Vladimir N.Vapnik V.The Nature of Statistical Learning Theory.New York:Springer-Verlag,Inc.1995 ISBN:0-387-94559-8
[16]Colin Campbell.Algorithmic Approaches to Training Support Vector.Department of Engineering Mathematics,Bristol University
【17]陆波,尉询楷.支持向量机在分类中的应用.中国图像图形学报.2005,10(8):1029-1035
【18]王立国,张晔.支持向量机多类目标分类器的结构简化研究.中国图像图形学报.2005,10(5):571-574
【19]顾磊,吴慧中.基于新的决策规则的球形支持向量机分类算法.系统仿真学报.2008,20(11):2901-2904
[20]Syed N,Liu H,Sung K.Incremental learning with support vector machines.Proceedings of the Workshop on Support Vector Machines at the International Joint Conference on Artificial Intelligence.Stockholm,Sweden:Morgan Kaufmann,1999:876-892.
【21]萧嵘,王继成,孙正兴,张福炎.一种SVM增量学习算法α-ISVM.软件学报.2001,12(12):1818-1824
【22]孔锐,张冰.一种快速支持向量机增量学习算法.控制与决策.2005,20(10):1129-1132
【23]毛建洋,黄道.一种新的支持向量机增量算法.华东理工大学学报(自然科学版).2006,32(8):989-991
[24]王晓丹,郑春颖.一种新的SVM对等增量学习算法.计算机应用.2006,26(10):2440-2443
[25]Mitra S K,Parker J M.Expert System for Identifying Defects:Stones and Cord in Glass.proceeding of the Sixth European Conference on Artificial Intelligence.1984.:3625-3637
[26]Sehmiedl.R.Automation of Flaw Detection on the Float Glass Ribbon.Glass.1985,62(11):408-409
[27]Latham.V,Nixon.M,Mayars.C.Automatic Optical Inspection of Table Glass.Glass Technology,1986,27(6):188-194
[28]Kysztof.J.Cios,RobertE.Tjia,NingLiuete.Study of Continuous ID3 and Radial Basis Function Algorithms for the Recognition of Glass Defects.0-7803-0164-1/91/0000-0049$01.0001991 IEEE.49-54
【29]李巧先.关于玻璃缺陷在线检测技术的研究.上海计量测试.1999,4:26-28
【30]马竹梧,白凤双.平板玻璃生产监控及质量缺陷诊断专家系统.基础自动化.2001,12:1-27
【31]刘刚.计算机图像处理技术在玻璃按等级分检中的应用【硕士学位论文].河南科技大学,2003
【32]冯燕,何明一,俞华憬.基于图像处理的玻璃均匀性检测系统.计算机应用.1999,4:64-66
[33]刘力双,王宝光,张挑.镀膜玻璃的瑕疵在线检测系统实验研究.制造业自动化.2003, 25(10):52-55
【34]陈作炳,连迅,马荣海.基于ARM与DSP玻璃缺陷在线检测系统.电子测量技术.2006,4:55-60
【35]肖燕.基于数字图像识别技术的玻璃缺陷在线检测实验系统研究【硕士学位论文].郑州大学,2007
[36]杨淑莹.VC++图像处理程序设计.北京:清华大学出版社.ISBN:7-81082-163-6
【37]张同燕.基于图像处理技术的玻璃缺陷识别方法研究【硕士学位论文].郑州大学,2007
【38]王平顺.图像处理技术在玻璃缺陷检测中的应用研究【硕士学位论文].燕山大学,2005
【39]贾沛.特征选择技术研究【硕士学位论文].华中科技大学,2003
[40]Sangho Yoon,Robert M Gray.Feature Selection Based on Maximizing Separability in Gauss Mixture Model and its Application to Image Classification.0-7803-9134-9/05/$20.00(?)2005 IEEE.487-491
【41]杨蕊红.小波矩算法在图像识别中的应用研究【硕士学位论文].西北工业大学,2003
【42]秦磊.基于矩的图像分析和快速算法【硕士学位论文].东南大学,2003
[43]RuiHong Yang.The Application of Wavelet Invariant Moments to Image Recognition.1-4244-0060-0/06/$20.00 2006 IEEE.3243-3247
[44]Vapnik V,Levin E,Le Cun Y.Measuring the VC Dimension of a Learning Machine.Neural Computation.1994,8(5):51-76.
【45]田园.支持向量机训练算法的研究与优化【硕士学位论文].西安理工大学,2008
[46]Bugres C J C.A Tutorial on Support Vector Machines for Pattern Recognition.Data Mining and Knowledge Discovery.1998 2(2):121-167
【47]肖靓.基于支持向量机的图像分类研究【硕士学位论文].同济大学,2006
【48]王炜,郭小明.关于核函数选取的方法.辽宁师范大学学报(自然科学版).2008,31(1):1-4
【49]黄琼英.支持向量机对分类算法的应用研究【硕士学位论文].河北工业大学,2005
[50]Platt J C,Cristianini N,Shawe Taylor J.Large Margin DAGs for Mutticlass Classification.Advances in Neural Information Processing Systems.2000,12(12).:547-553
[51]Pairwise.Classification and Support Vector Machines.Cambridge:The MIT Press,1999.
[52]萧嵘,王继成.一种SVM增量学习算法南京大学学报(自然科学).2002,38(2):152-157
[53]王建华,宋永胜.基于边界向量的支持向量机增量算法.计算机工程与应用.2007年43(33):173-177