玻璃质量在线检测算法研究与系统实现
|
摘要
随着中国社会经济的迅速发展,对玻璃外观质量的要求越来越高,为了提高国内玻璃行业的生产自动化水平,针对国内玻璃质量在线检测理论研究与系统开发的严重滞后,本文对基于机器视觉的运动图像实时处理算法与理论进行了研究,并在此基础上完成了玻璃质量在线检测的关键技术研究与系统实现。
通过对玻璃检测的技术要求和应用特点进行分析,提出了玻璃质量在线检测系统的总体技术方案,对玻璃缺陷识别原理进行了阐述,提出了系统目标和技术要求,从硬件和软件两个层次对检测系统进行设计,将硬件系统划分为五个模块分别进行介绍,提出了软件系统的总体设计框图,重点阐述了图像处理和分析流程。
设计了一套玻璃图像获取系统,对采用不同光源所获得的玻璃图像进行了比较,采用板卡程序库来实现图像采集程序。通过预处理流程得到增强了的灰度玻璃图像,主要包括线性变换、中值滤波、运动模糊消除几个过程。为了减少环境的影响,提取了玻璃图像模板,然后将实时图像和模板图像进行不同的差影运算,来得到缺陷图像。
将图像进行分区,每个区域选择不同的阈值,提出了一种自适应阈值选取算法;经过基于微分算子的边缘检测算法,可以提取出物体目标边界;基于样品缺陷特征,在对浮法玻璃灰度图像进行二值化时,采用了一种新的求分块阈值的局部阈值分割方法;采用形态学上的开闭运算来消除噪声的影响。缺陷的查找定位算法采用行程长编码算法,本文利用此算法设计了线阵图像的污点(blob)查找算法;根据玻璃图像的规律设计了除灰流程;通过分析缺陷图像,提取了若干特征作为区分缺陷类型的指标。
为了区分不同类型的缺陷,设计了神经网络分类器;在分析了标准BP算法的基础上,提出了一种改进的BP算法,对原有BP算法中存在的缺陷给予改进,并针对两种不同的训练方式进行了讨论,将这种改进的BP算法应用于玻璃缺陷分类和字符识别中。结果表明,改进后的BP算法收敛速度加快,识别精度提高。
最后,研制并开发了玻璃质量在线检测系统,包括硬件和配套处理软件,目前已经试运行于工业现场。
With the rapid development of society and economy, the quality requirement of glass is higher and higher, to increase the automotive level of China glass industry, and aiming at the serious lag of theories study and system development to online quality detection of glass, real-time processing algorithm and theory of motion image that based on machine vision technology is studied in the thesis, and some key technologies of glass quality online detection system are realized.
Total technology scheme of online quality detection system of glass is proposed according the analyses to the technical requirement and application feature of glass detection. The principle of glass defect recognition is explained. The aim of system and technical requirement is proposed. The system of detection is designed from two points of hardware and software. The system of hardware is divided into five modules. The total design frame of software is proposed. The flow of image processing and analyze is expounded as the emphasis.
An acquisition system of glass image is designed. The images are compared which are acquired with different lamp-houses. The grab procedure is realized using the library of grab-board. The flow of preprocess is designed, including linear transform, median filter and the eliminating of eliminate motion blur. The flow results in strengthened gray image of glass. We distill the template of stand image to reduce the influence of environments. The real-time image is subtracted from stand template to get the image of detection.
The image needs to be divided into several areas. Each area choose different threshold. A self-adapted algorithm of threshold selection is proposed. The boundary of target could be distilled by the algorithm of edge detection based on differential coefficient operator. Based on the feature of sample defect, a new method of local threshold-division is adopted to get the threshold of area. The noise could be eliminated using basic algorithm that base on the mathematic morphology. The lookup and orientation of defect adopt RLE algorithm. We design the flow of dust-elimination according to the rhythm of glass image. Several features are abstracted to distinguish the types of defect.
Neural network is designed as sorter. On the base of analyzing standard BP algorithm, in allusion to the limitations that standard BP algorithm have, some improved methods are given. Two different training methods are compared. The improved algorithm is realized in Visual C++ environment. Experiments are done in allusion to pattern classify and character recognize. The results show that this BP algorithm is more efficient on the aspects of convergence rate and the precision of recognize.
At last, the online quality detection system of glass is realized, including hardware and software. Now the system is running on the field.
通过对玻璃检测的技术要求和应用特点进行分析,提出了玻璃质量在线检测系统的总体技术方案,对玻璃缺陷识别原理进行了阐述,提出了系统目标和技术要求,从硬件和软件两个层次对检测系统进行设计,将硬件系统划分为五个模块分别进行介绍,提出了软件系统的总体设计框图,重点阐述了图像处理和分析流程。
设计了一套玻璃图像获取系统,对采用不同光源所获得的玻璃图像进行了比较,采用板卡程序库来实现图像采集程序。通过预处理流程得到增强了的灰度玻璃图像,主要包括线性变换、中值滤波、运动模糊消除几个过程。为了减少环境的影响,提取了玻璃图像模板,然后将实时图像和模板图像进行不同的差影运算,来得到缺陷图像。
将图像进行分区,每个区域选择不同的阈值,提出了一种自适应阈值选取算法;经过基于微分算子的边缘检测算法,可以提取出物体目标边界;基于样品缺陷特征,在对浮法玻璃灰度图像进行二值化时,采用了一种新的求分块阈值的局部阈值分割方法;采用形态学上的开闭运算来消除噪声的影响。缺陷的查找定位算法采用行程长编码算法,本文利用此算法设计了线阵图像的污点(blob)查找算法;根据玻璃图像的规律设计了除灰流程;通过分析缺陷图像,提取了若干特征作为区分缺陷类型的指标。
为了区分不同类型的缺陷,设计了神经网络分类器;在分析了标准BP算法的基础上,提出了一种改进的BP算法,对原有BP算法中存在的缺陷给予改进,并针对两种不同的训练方式进行了讨论,将这种改进的BP算法应用于玻璃缺陷分类和字符识别中。结果表明,改进后的BP算法收敛速度加快,识别精度提高。
最后,研制并开发了玻璃质量在线检测系统,包括硬件和配套处理软件,目前已经试运行于工业现场。
With the rapid development of society and economy, the quality requirement of glass is higher and higher, to increase the automotive level of China glass industry, and aiming at the serious lag of theories study and system development to online quality detection of glass, real-time processing algorithm and theory of motion image that based on machine vision technology is studied in the thesis, and some key technologies of glass quality online detection system are realized.
Total technology scheme of online quality detection system of glass is proposed according the analyses to the technical requirement and application feature of glass detection. The principle of glass defect recognition is explained. The aim of system and technical requirement is proposed. The system of detection is designed from two points of hardware and software. The system of hardware is divided into five modules. The total design frame of software is proposed. The flow of image processing and analyze is expounded as the emphasis.
An acquisition system of glass image is designed. The images are compared which are acquired with different lamp-houses. The grab procedure is realized using the library of grab-board. The flow of preprocess is designed, including linear transform, median filter and the eliminating of eliminate motion blur. The flow results in strengthened gray image of glass. We distill the template of stand image to reduce the influence of environments. The real-time image is subtracted from stand template to get the image of detection.
The image needs to be divided into several areas. Each area choose different threshold. A self-adapted algorithm of threshold selection is proposed. The boundary of target could be distilled by the algorithm of edge detection based on differential coefficient operator. Based on the feature of sample defect, a new method of local threshold-division is adopted to get the threshold of area. The noise could be eliminated using basic algorithm that base on the mathematic morphology. The lookup and orientation of defect adopt RLE algorithm. We design the flow of dust-elimination according to the rhythm of glass image. Several features are abstracted to distinguish the types of defect.
Neural network is designed as sorter. On the base of analyzing standard BP algorithm, in allusion to the limitations that standard BP algorithm have, some improved methods are given. Two different training methods are compared. The improved algorithm is realized in Visual C++ environment. Experiments are done in allusion to pattern classify and character recognize. The results show that this BP algorithm is more efficient on the aspects of convergence rate and the precision of recognize.
At last, the online quality detection system of glass is realized, including hardware and software. Now the system is running on the field.
引文
[1] 王丽萍, 蓝天. 玻璃缺陷在线自动检测设备性能及实践. 玻璃,2003(5):46-47
[2] 王平顺. 图像处理技术在玻璃缺陷检测中的应用研究. 燕山大学硕士学位论文,2005
[3] CCD&CMOS 图像和机器视觉产品手册. 凌云光视数字图像公司, 2005
[4] 宋培华. 在线检测中机器视觉的 CCD 方案. 计算机工程,2001,27(11):23-25
[5] 陈元琰. 基于计算机视觉的玻璃瓶裂纹在线检测系统. 计算机应用, 2001, 21(11): 48-49
[6] 于红斌, 李志能, 陈抗生. 一种运动模糊图像的快速恢复算法. 浙江大学学报(工学版),1999, 33 (5):564-568
[7] 何斌, 马天予, 王运坚. Visual C++数字图像处理. 北京:人民邮电出版社,2001
[8] 蔡岷, 韩力群. 计算机皮革图像检测与分类系统. 自动化仪表,1999,20(5):25-27
[9] 冯燕. 基于图像处理的玻璃均匀性检测系统. 计算机应用, 2004, 24(4):64-66
[10] 章毓晋. 图像处理和分析. 北京:清华大学出版社,1999
[11] 章毓晋. 图像分割. 北京:北京科学出版社,2001
[12] 张爱华, 余胜生, 周敬利. 一种基于边缘检测的局部阈值分割算法. 小型微型计算机系统,2003,24(4):661-663
[13] 戴 青 云 . 数 学 形 态 学 在 图 像 处 理 中 的 应 用 进 展 . 控 制 理 论 与 应 用 ,2001,18(4):439-444
[14] J Yang, X Liu, W Guo. A Recognition of Vehicle plate with Mathematical Morphology and Neural Networks. Journal of Wuhan University of Technology (Transportation Science & Engineering),2001,25(3): 371-374
[15] W X Wang, L Y Wang. Froth image segmentation algorithms and their validation. Proceedings of ICSP2000, 2000IEEE:2042-2044
[16] Anil.K.Jain, sushil Bhattacharjee. Text Segmentation Using Gabor Filters For Automatic Document Processing. Journal of Machine Vision and application, 1992(5):169-184
[17] Knopf G.K., Gupta, M.M.. A multipurpose neural processor for machine vision systems. IEEE Transactions on Neural Networks, 1993, 4 (5):762-777
[18] C Vasantha Lakshmi, C Patvardhan . A Multi-font OCR System for Printed Telugu Text. Proceedings of the Language Engineering Conference,2002:7-11
[19] 卢桂馥. 非均匀背景目标图像的分割方法. 计算机工程, 2004,30(16):141-142
[20] Rafael C. Gonzalez. Digital Image Processing(Second Edition).Publishing House ofElectronics Industry,2003
[21] 石争浩. 基于机器视觉的铸件表面缺陷人工神经网络检测方法. 重型机械,2004(2):51-54
[22] Thomopoulos S.C.A., Bougoulias D.K.. A self-organizing neural network for automatic pattern recognition and classification, 1991 IEEE International Joint Conference on Neural Networks, 1991(3): 2683 -2692
[23] B Chen, L L Hoberock. Machine vision fuzzy object recognition and inspection using a new fuzzy neural network. Proceedings of the IEEE International Symposium on Intelligent Control, 1996:206 -211
[24] 江虹, 曾立波, 胡继明. 优化的 BP 神经网络分类器的设计与实现. 计算机工程与应用,2001(5):122-125
[25] 韩皖贞.计算机视觉玻壳缺陷检测方法的研究.天津轻工学院硕士生学位论文,2002:1-5
[26] 王永强,于德敏,徐增扑.机器视觉玻壳缺陷检测方法的研究.天津轻工学院学报,2001(2):9-12
[27] 罗大庸. 一种基于神经网络的分类系统. 长沙铁道学院学报,1995, 13(1):11-15
[28] 张宏林. 数字图像模式识别技术及工程实践. 北京:人民邮电出版社,2003
[29] 谢华. 微晶玻璃缺陷检测和质量分级的研究. 华南理工大学硕士学位论文,2002
[30] 叶世伟,史忠值(译).神经网络原理. 北京:机械工业出版社. 2003
[31] Martin T. Hagan. Neural Network Design. PWS Publishing Company. 2002
[32] 姜立方,刘泊,施莲辉. 一种改进的 BP 算法及其在模式识别中的应用. 哈尔滨理工大学学报,2003,8(3):90-93
[33] 张晓光. 基于神经网络的焊缝缺陷识别专家系统. 计算机工程,2003, 29(17):22-23
[34] 张新峰. 模式识别及其在图像处理中的应用. 测控技术,2004,23(5):28-32
[35] C S Won. Improved block-based image segmentation. Proceedings of International Conference on Image Processing, 1999(1):329-332
[36] Z Q Wen, T Yang. Building a rule-based machine-vision system for defect inspection on apple sorting and packing lines. Expert System with Application,1999(16):307-313
[37] 曹凯. 基于面向对象模糊特征类库的特征识别模型的研究. 天津工业大学学报, 2004,23(2):55-57
[38] 胡金滨,唐旭清. 人工神经网络的 BP 算法及其应用. 信息技术,2004,28(4):1-4
[39] 唐胜,周经野,钱跃良,李锦涛. 一种基于模拟退火的自适应算法. 计算机工程与应用,2002(18):43-46
[40] Jonathan D. Holmes. Inspection of float glass using a novel retro-reflective laserscanning system. SPIE,Vol 3131:180-190
[41] www.glassinfo.com.cn
[42] Klaus Bange, Hartmut M uller. Characterization of Defects in Glasses and Coatings on Glasses by Microanalytical Techniques. Mikrochim. Acta,2000(132):493-503
[43] Hazem Raafat, Salem Taboun. An integrated robotic and machine vision system for surface flaw detection and classification. Computers ind,1996,30(1):27-40
[44] Hui-min Ma, Guang-da Su, Jun-yan Wang, Zheng Ni. A glass bottle defect detection system without touching. Proceedings of the First International Conference on Machine Learning and Cybernetics, 2002:628-632
[45] Seungkoo Lee, George Vachtsevanos. An application of rough set theory to defect detection of automotive glass. Mathematics and Computers in Simulation, 2002(60): 225-231
[2] 王平顺. 图像处理技术在玻璃缺陷检测中的应用研究. 燕山大学硕士学位论文,2005
[3] CCD&CMOS 图像和机器视觉产品手册. 凌云光视数字图像公司, 2005
[4] 宋培华. 在线检测中机器视觉的 CCD 方案. 计算机工程,2001,27(11):23-25
[5] 陈元琰. 基于计算机视觉的玻璃瓶裂纹在线检测系统. 计算机应用, 2001, 21(11): 48-49
[6] 于红斌, 李志能, 陈抗生. 一种运动模糊图像的快速恢复算法. 浙江大学学报(工学版),1999, 33 (5):564-568
[7] 何斌, 马天予, 王运坚. Visual C++数字图像处理. 北京:人民邮电出版社,2001
[8] 蔡岷, 韩力群. 计算机皮革图像检测与分类系统. 自动化仪表,1999,20(5):25-27
[9] 冯燕. 基于图像处理的玻璃均匀性检测系统. 计算机应用, 2004, 24(4):64-66
[10] 章毓晋. 图像处理和分析. 北京:清华大学出版社,1999
[11] 章毓晋. 图像分割. 北京:北京科学出版社,2001
[12] 张爱华, 余胜生, 周敬利. 一种基于边缘检测的局部阈值分割算法. 小型微型计算机系统,2003,24(4):661-663
[13] 戴 青 云 . 数 学 形 态 学 在 图 像 处 理 中 的 应 用 进 展 . 控 制 理 论 与 应 用 ,2001,18(4):439-444
[14] J Yang, X Liu, W Guo. A Recognition of Vehicle plate with Mathematical Morphology and Neural Networks. Journal of Wuhan University of Technology (Transportation Science & Engineering),2001,25(3): 371-374
[15] W X Wang, L Y Wang. Froth image segmentation algorithms and their validation. Proceedings of ICSP2000, 2000IEEE:2042-2044
[16] Anil.K.Jain, sushil Bhattacharjee. Text Segmentation Using Gabor Filters For Automatic Document Processing. Journal of Machine Vision and application, 1992(5):169-184
[17] Knopf G.K., Gupta, M.M.. A multipurpose neural processor for machine vision systems. IEEE Transactions on Neural Networks, 1993, 4 (5):762-777
[18] C Vasantha Lakshmi, C Patvardhan . A Multi-font OCR System for Printed Telugu Text. Proceedings of the Language Engineering Conference,2002:7-11
[19] 卢桂馥. 非均匀背景目标图像的分割方法. 计算机工程, 2004,30(16):141-142
[20] Rafael C. Gonzalez. Digital Image Processing(Second Edition).Publishing House ofElectronics Industry,2003
[21] 石争浩. 基于机器视觉的铸件表面缺陷人工神经网络检测方法. 重型机械,2004(2):51-54
[22] Thomopoulos S.C.A., Bougoulias D.K.. A self-organizing neural network for automatic pattern recognition and classification, 1991 IEEE International Joint Conference on Neural Networks, 1991(3): 2683 -2692
[23] B Chen, L L Hoberock. Machine vision fuzzy object recognition and inspection using a new fuzzy neural network. Proceedings of the IEEE International Symposium on Intelligent Control, 1996:206 -211
[24] 江虹, 曾立波, 胡继明. 优化的 BP 神经网络分类器的设计与实现. 计算机工程与应用,2001(5):122-125
[25] 韩皖贞.计算机视觉玻壳缺陷检测方法的研究.天津轻工学院硕士生学位论文,2002:1-5
[26] 王永强,于德敏,徐增扑.机器视觉玻壳缺陷检测方法的研究.天津轻工学院学报,2001(2):9-12
[27] 罗大庸. 一种基于神经网络的分类系统. 长沙铁道学院学报,1995, 13(1):11-15
[28] 张宏林. 数字图像模式识别技术及工程实践. 北京:人民邮电出版社,2003
[29] 谢华. 微晶玻璃缺陷检测和质量分级的研究. 华南理工大学硕士学位论文,2002
[30] 叶世伟,史忠值(译).神经网络原理. 北京:机械工业出版社. 2003
[31] Martin T. Hagan. Neural Network Design. PWS Publishing Company. 2002
[32] 姜立方,刘泊,施莲辉. 一种改进的 BP 算法及其在模式识别中的应用. 哈尔滨理工大学学报,2003,8(3):90-93
[33] 张晓光. 基于神经网络的焊缝缺陷识别专家系统. 计算机工程,2003, 29(17):22-23
[34] 张新峰. 模式识别及其在图像处理中的应用. 测控技术,2004,23(5):28-32
[35] C S Won. Improved block-based image segmentation. Proceedings of International Conference on Image Processing, 1999(1):329-332
[36] Z Q Wen, T Yang. Building a rule-based machine-vision system for defect inspection on apple sorting and packing lines. Expert System with Application,1999(16):307-313
[37] 曹凯. 基于面向对象模糊特征类库的特征识别模型的研究. 天津工业大学学报, 2004,23(2):55-57
[38] 胡金滨,唐旭清. 人工神经网络的 BP 算法及其应用. 信息技术,2004,28(4):1-4
[39] 唐胜,周经野,钱跃良,李锦涛. 一种基于模拟退火的自适应算法. 计算机工程与应用,2002(18):43-46
[40] Jonathan D. Holmes. Inspection of float glass using a novel retro-reflective laserscanning system. SPIE,Vol 3131:180-190
[41] www.glassinfo.com.cn
[42] Klaus Bange, Hartmut M uller. Characterization of Defects in Glasses and Coatings on Glasses by Microanalytical Techniques. Mikrochim. Acta,2000(132):493-503
[43] Hazem Raafat, Salem Taboun. An integrated robotic and machine vision system for surface flaw detection and classification. Computers ind,1996,30(1):27-40
[44] Hui-min Ma, Guang-da Su, Jun-yan Wang, Zheng Ni. A glass bottle defect detection system without touching. Proceedings of the First International Conference on Machine Learning and Cybernetics, 2002:628-632
[45] Seungkoo Lee, George Vachtsevanos. An application of rough set theory to defect detection of automotive glass. Mathematics and Computers in Simulation, 2002(60): 225-231