摘要
SMT产品机器视觉检测AOI中,对具体产品样本进行机器学习时,难以得到被检测产品的多种缺陷样本,合格产品样本也需要减少学习样本数量,以提高检测速度,减少人工检测。文中研究了SMT产品的机器视觉图象特征及分布,结合快速分层的并行检测方法,提出了针对大量产品正、负类样本学习建立检测经验参数,针对具体产品负类学习建立模板的学习方法。并在具体产品样本学习中采取动态阈值,以减少人工检测确立学习样本的数量。多个生产线的产品检测实验表明本方法可以速有效地建立AOI检测模板参数。
In automated optical inspection for SMT products, positive Class samples were difficult to find for machine learning and samples number were expected to decreased to improve effect. Based on quick inspection method, this paper proposed machine learning for image features that independent with products types to set up AOI classifier and other features for samples. thresholds were suggested during model setup that refusing rate descended while more datail type product images were learned. Experiments for SMT products from different lines showed that this approach was effective.
出处
《电子质量》
2009年第1期39-41,44,共4页
Electronics Quality
