基于机器视觉的啤酒瓶口缺陷检测分类方法研究

Research on beer bottle defect classification detection method based on machine vision

针对目前国内玻璃空瓶机器视觉检测系统存在瓶口缺陷分类检测精度不高的问题,提出一种基于机器视觉的可靠的检测方法。首先选取封盖面缺口、外环口崩口、口缘毛刺、口面磨损、内环口崩口、封盖面破裂等6种常见缺陷类型作为分类目标,研究6种常见瓶口缺陷类型图像的表面特征,提出以灰度方差等6种瓶口的缺陷特征构成支持向量机(SVM)分类算法的输入向量,并择优选择径向基(RBF)函数作为SVM分类器的核函数,然后根据瓶口缺陷的分类性质选择多类分类方式中的一类对余类法(OVR)设计相应的SVM。最后,每种缺陷都选取80个样本对所设计SVM分类器进行训练学习与测试。测试结果表明:设计的SVM分类器能较精准地检测出6种常见的瓶口缺陷类型,识别率为91.6%,满足生产企业对机器视觉检测系统缺陷分类识别的要求。

Aiming at the defect classification detection accuracy of existing domestic glass bottles machine vision inspection system is not high issue,a reliable detection method based on machine vision is proposed. First,six common types of defects： cover plane gap,outer mouth chipping,burrs rim,mouth surface wear,inner mouth chipping and cracking are selected as classification targets,and the surface features of six common bottle defects type image are researched. Proposing six bottle features as gray variance constitutes a SVM classification algorithm of input vectors,and the RBF（ radial base function） is selected as a kernel function of SVM classifier. Then,based on the nature of the bottle defect classification we choose OVR multi-class classification to design the corresponding SVM. Finally,80 samples are selected from each of these defects,and the train and test for the designed SVM classifier is carried out. The test results show that the designed SVM classifier can more accurately detect six common types of bottle defects,and the recognition rate is 91. 6%,which meets the producer ＇s requirements for the defect classification recognition of machine vision inspection system.