基于支持向量机和机器视觉的磁瓦缺陷检测的研究

Magnet's defect detection based on machine vision and LS- SVM

提出了一种基于支持向量机(SVM)和机器视觉的磁瓦缺陷的检测方法,通过图像增强、边缘提取的方法提取磁瓦缺陷后,提取磁瓦各种缺陷类型的形态特征参数,其中选择了面积、周长、紧凑性、离心率、一阶中心矩、二阶中心矩、三阶中心矩、四阶中心矩、五阶中心矩以及六阶中心矩等。将总共90张磁瓦侧面图片的漏磨、开裂和正常等3种情况的特征参数提取,在每种类型随机选取18个样本,即共54个样本作为LS-SVM的建模量,其余36个样本作为LS-SVM的预测样本来检验模型的优劣,获得缺陷类型识别准确率达91.67%,该结果表明通过支持向量机和机器视觉可实现较高精度和效率的磁瓦缺陷检测。

This paper presents a method for magnet's defect detection based on machine vision and support vector machine( SVM). Through the method of image enhancement,edge extraction to obtain the defect part, then extract the characteristic parameters,the morphological characteristic parameters we chose are the area,perimeter,compactness,and the first to sixth central moments. A total of 90 piece of magnet's picture with three kinds of defects including the missing block,leakage of grinding cracking,and normal are studied,by extracting characteristic parameters to be the inputs of LS-SVM and choosing a set of 54 training samples and remaining 36 test samples to charge the LS-SVM model,the result showee the defect recognition accuracy of 91. 67%,which shows that support vector machine and machine vision used in the magnet's defect detection can achieve higher accuracy and efficiency.