基于小波变换和SSVME的PCB产品视觉检测中缺陷分类研究  

DEFECT CLASSIFICATION RESEARCH IN WAVELET TRANSFORM AND SSVME BASED PCB PRODUCT VISION INSPECTION

作　　者：任斌[1] 程良伦[2] 

机构地区：[1]东莞理工学院电子工程学院,广东东莞523808 [2]广东工业大学自动化学院,广东广州510006

出　　处：《计算机应用与软件》2012年第6期167-171,共5页Computer Applications and Software

基　　金：广东省自然科学基金(S2011010002144);省部产学研结合项目(2010B090400457;2011B090400269;2011A091 000028);东莞理工学院自然科学青年基金(2010ZQ04)

摘　　要：针对PCB产品视觉检测中图像缺陷细微、形状复杂、特征难于提取、易受噪声影响的问题,提出基于小波变换和光滑支持向量机集成SSVME(Smooth Support Vector Machine Ensemble)的多分类方法,有效解决了细微、复杂缺陷难以识别分类的问题。实验表明,该方法六类缺陷混合识别率达到95.26%,高于BP神经网络的最优识别率90.35%和基于区域方法的80.67%,而且训练和分类时间短。从理论和实验中验证了该方法的有效性,是PCB产品视觉检测领域中缺陷识别分类的新方法,具有重要的应用价值。Considering problems that exist in PCB product vision inspection such as slim image defect,complex shape,hard to extract feature,vulnerable to noise and so on,the thesis presents a wavelet transform and smooth support vector machine ensemble（SSVME） based multiple classification algorithm,which effectively resolves problems like hard to recognize or classify slim or complex defects.Experiments show that,with this method,the discrimination ratio of the mixture of six kinds of defects reaches 95.26%,higher than both the highest discrimination ratio of BP neural network,which is 90.35%,and that of region based method,which is 80.67%.Moreover,its training and classification time is short.The method is verified for its effectiveness both theoretically and experimentally.Therefore it is significantly valuable for applications.

关 键 词：PCB 缺陷检测 小波变换和光滑支持向量机集成算法 分类 

分 类 号：TP302.7[自动化与计算机技术—计算机系统结构]