摘要
随着印刷电路板组装技术向高密度化和"零缺陷"方向发展,市场对自动光学检测系统的要求也向高准确率、智能化发展.而传统自动光学检测系统的检测算法不仅需要进行复杂的设置,而且需要大量不同类型的样本进行训练以提高系统的泛化性能,但在电路板组装过程中,有缺陷的样本难以获得.针对此类问题,提出了一种适应少缺陷样本的智能检测方法.首先对焊点图像的一系列特征进行了提取;然后介绍了一种基于统计方法的自动阈值设置方法;最后建立了用于进行焊点分类的BP神经网络.结果表明,方法具有较高的准确率.
As electronic components get smaller and the board densities become more compact,it is necessary for automatic inspection in electronic manufacturing.The automatic optical inspection(AOI) system is demanded more precise and intelligent.The traditional inspection methods require large quantity samples of all types to train the inspector,or do some complicated setting.To overcome the disadvantages,an intelligent method was proposed.Firstly,a series of features of soldered joints were defined.Then,an automatic boundary setting method based on statistic was introduced.Finally,the neural network was established to classify the soldered joints.The performance of the method was verified by the experiment.
出处
《焊接学报》
EI
CAS
CSCD
北大核心
2009年第5期57-60,共4页
Transactions of The China Welding Institution
基金
国家杰出青年科学基金资助项目(50822504)
关键词
焊点
神经网络
机器视觉
检测
solder joint
neural networks
machine vision
inspection
