基于FNBC分类器的PCB产品视觉检测中缺陷分类研究

Research of defect identification classification algorithm based on FNBC for PCB produducts vision inspection

针对PCB产品视觉检测中图像缺陷细微、形状复杂、特征难于提取、易受噪声影响的问题,本文把Fisher分的属性转换方法和朴素贝叶斯分类器相结合,把Fisher分的属性转换方法应用朴素贝叶斯分类器上提出一种新的分类器—Fisher分朴素Bayes分类器(Fisher Naive Bayes Classifier,FNBC)。并将Fisher分朴素Bayes分类器应用到PCB产品视觉检测中缺陷分类中。实验表明,该方法六类缺陷混合识别率达到95.6%,高于BP神经网络的最优识别率91.8%和基于区域方法的81.3%,而且训练和分类时间短,具有重要的应用价值。.

Considering the problem that fine image defect＇s fineness, complex shape, difficultly to extraction feature, and effected by noise easily on PCB products machine vision inspection system, presented defect identification classification algorithm based on FNBC. It resolved the problem that fine and complex defect is difficult to classify. It resolved the problem that the structure of binary tree affected the accuracy of classifier, and upgraded defect classification accuracy finally. The e^ts show that six defects discrimination of this method is up to 95.6%, higher than BP network＇s best discrimination 91.8% and 81.3% by method based on region. And the training and inspecting time is few. Verified this method efficiency from theory and experiment, and it has great value for research and usage.