基于机器视觉玻璃缺陷在线检测系统的研究

发布时间：2018-06-01 00:34

本文选题：机器视觉 + 支持向量机　；参考：《山东理工大学》2016年硕士论文

【摘要】：玻璃缺陷在线检测系统已经在日益发展的今天发挥了至关重要的作用,机器视觉也已发展成为玻璃缺陷在线检测系统的中枢。随着计算机技术、信息技术和网络技术的迅速发展,基于机器视觉玻璃缺陷在线检测系统的先进性与卓越性越发的突出,逐渐成为当今自动检测系统的发展主流和研究热点。本文设计了基于机器视觉的玻璃缺陷在线检测系统,提出基于原始动态阈值算法的改进算法与基于支持向量机选择最优核函数的方法和参数优化方法。对玻璃缺陷的特征进行提取及分析,在此基础上对其进行识别与分类。首先,本文以机器视觉结构为指导思想,通过对其硬件以及软件各模块的功能的详细介绍以及特点的分析,设计了适用于玻璃缺陷在线检测系统总体方案。其次,针对玻璃表面缺陷图像的特点,研究了缺陷图像的预处理方法,对缺陷图像的进行了增强;重点研究了玻璃缺陷提取模块,针对传统的缺陷提取算法的不足,研究并提出了改进的动态阈值的玻璃缺陷提取算法。通过实验对算法的时效性与准确性进行验证,实验结果表明改进后的算法能快速将玻璃缺陷的存在检出,准确率高达97%,能够满足玻璃在线质量检测的各种指标。最后,对于缺陷的分类研究采用了基于支持向量机的算法,在支持向量机的基础上对多种分类器进行了设计,并选取了足够数量的缺陷样本进行实验验证,获得了令人满意的分类结果。通过实验验证的结果可以得出,此分类算法比起现有的分类算法具有更高的准确性和效率性。
[Abstract]:On-line glass defect detection system has been playing an important role in the development of today, and machine vision has become the center of on-line glass defect detection system. With the rapid development of computer technology, information technology and network technology, the advance and excellence of the on-line detection system based on machine vision glass defects are becoming more and more prominent, and gradually become the mainstream and research focus of automatic detection system. An on-line glass defect detection system based on machine vision is designed in this paper. The improved algorithm based on original dynamic threshold algorithm and the method of selecting optimal kernel function based on support vector machine and parameter optimization method are proposed. The features of glass defects were extracted and analyzed, and then the glass defects were identified and classified. Firstly, based on the structure of machine vision, this paper introduces the functions of the hardware and software modules in detail and analyzes the characteristics of the system, and designs the overall scheme of the on-line detection system for glass defects. Secondly, according to the characteristics of the glass surface defect image, the pre-processing method of the defect image is studied, and the defect image is enhanced. The focus is on the glass defect extraction module, aiming at the shortcomings of the traditional defect extraction algorithm. An improved dynamic threshold algorithm for glass defect extraction is proposed. The experimental results show that the improved algorithm can detect the defects of glass quickly, and the accuracy is as high as 97%, which can meet the requirements of on-line quality detection of glass. Finally, the algorithm based on support vector machine is used in the research of defect classification. Several classifiers are designed on the basis of support vector machine, and a sufficient number of defect samples are selected for experimental verification. Satisfactory classification results have been obtained. The experimental results show that the proposed algorithm is more accurate and efficient than the existing classification algorithms.
【学位授予单位】：山东理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TQ171.7;TP391.41

【参考文献】