基于ANN集成的贴片安装机器视觉检测  

ANN ENSEMBLES BASED MACHINE VISION INSPECTION FOR SOLDER JOINTS

作　　者：罗兵[1] 陈恒法[1] 吴锐珍[1] 

机构地区：[1]华南师范大学南海学院,广东南海528225

出　　处：《华南师范大学学报（自然科学版）》2008年第2期43-49,共7页Journal of South China Normal University(Natural Science Edition)

摘　　要：在贴片安装产品机器视觉检测中,由于图像数据量大、变化复杂、样本分布和错误代价不平衡及检测实时性的要求,引脚的焊接缺陷检测难度最大.从图像中选择和提取多个特征,分别用于各个结构简单的ANN检测分类器,再将多个ANN输出进行线性组合来得到最后的检测结果.各ANN学习时,样本初始权重考虑样本的不平衡性,学习中再用提升算法来调整样本权重和集成系数;用遗传算法来学习确定ANN,用代权重的检测正确率和最小正确分隔边缘作为适应值函数,两类边缘与代价成正比.ANN集成后在精度提高的同时保持了良好的泛化性能.实验结果表明:本方案准确率高、泛化性好、速度快.In automatic machine vision inspection for SMT assembly products,components solder joints inspection is a meaningful but hard work suffered from various complicated mass image data and unbalanced samples.This study tries to solve it by ANN ensembles,each of which was inputted by different features extracted from the image.Every simple ANN was trained by the genetic algorithm whose fitness function was extracted from the loss-weighted minimal margin.AdaBoost algorithm provided ANN linear combination coefficients and training samples＇ weights were adjusted according to other ANN＇s training results.Experimental results showed that this approach had high accuracy,well generalization performance and low calculation cost.

关 键 词：ANN集成 自适应提升 机器视觉 模式识别 遗传算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]