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.
出处
《计算机应用研究》
CSCD
北大核心
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)
关键词
自动光学检测
统计学习
贝叶斯决策
缺陷检测
automatic optical inspection （AOI）
statistical learning
Bayesian decision
defects inspection
