基于统计分析的PCB组装缺陷特征学习方法  

Features learning method for PCB assembling defects inspection based on statistical analysis

作　　者：邝泳聪[1] 欧阳高飞[1] 谢宏威[1] 洪始良[1] 杨锦荣[1] 

机构地区：[1]华南理工大学机械与汽车工程学院,广州510640

出　　处：《计算机应用研究》2010年第2期775-777,783,共4页Application Research of Computers

基　　金：国家杰出青年科学基金资助项目(50825504);广东省科技攻关重点项目(2008A010300002);粤港关键领域重点突破招标项目(东莞专项20081628)

摘　　要：为了减少自动光学检测系统对用户经验的依赖,提出了一种基于统计分析的PCB组装缺陷特征学习方法。该方法通过对良品和不良品样本图像的统计学习优选出分类能力强的特征,再采用最小风险贝叶斯决策得到特征分类参数。实验结果表明,该算法有效地简化了用户检测程序的编程和调试,提高了AOI的使用效率和准确率。In order to reduce the experience-dependence of automatic optical inspection （AOI）, proposed a Bayesian-based features learning for PCB assembling defects inspection. By statistical learning images of good product sample and defective product sample, selected features with better ability of classification capacity, and based on the risk minimization of Bayesian, worked out the decision parameters for feature classing. Experimental results show that the proposed method effectively simplifies the programming and debugging of user inspection application, and greatly improves the efficiency and accuracy of AOI.

关 键 词：自动光学检测 统计学习 贝叶斯决策 缺陷检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]