基于计算机视觉的镁薄板表面缺陷检测系统的研究

发布时间：2018-02-24 04:02

本文关键词： 计算机视觉缺陷检测特征提取实时检测　出处：《辽宁科技大学》2016年硕士论文　论文类型：学位论文

【摘要】：镁是最轻的结构金属材料之一,因其质量轻、刚性好、密度低、散热快等优点被广泛应用于航空、运输、化工等各个领域。在镁合金薄板的轧制过程中,因机械及控制精度原因,易出现边裂、波纹、褶皱等缺陷,如果这个问题不能够被及时解决,将严重影响镁合金薄板成品质量,因此如何对镁合金薄板表面缺陷进行高效、快速地检测将成为镁合金薄板制造业的关键。传统的人工检测方法,效率低下,误检率高,而且耗费了大量的人力资源。现有的检测技术中常用的涡流检测技术、红外检测技术以及漏磁检测技术等由于其原理的局限性,检测速率和可识别的缺陷类型极为有限。近年来,随着计算机、自动化、人工智能、图像识别等技术的发展,以计算机视觉为核心的表面缺陷检测技术已经成为镁合金薄板生产的研究重点。本课题研制了镁合金薄板表面缺陷识别系统,其实时采集薄板表面图像,通过计算机视觉技术自动检测缺陷并利用贝叶斯分类器判定缺陷类型,为后续实现生产过程的无人值守及高度自动化控制奠定了基础。主要完成工作如下:(1)根据实验环境、生产线上的实际状况以及经济预算,确定了本系统的软硬件设计方案,如选择LED光源、确定照明方案、配置操作系统等。(2)由于薄板带表面具有对比度低、易发生光照反射的特点,选择了直方图均衡化、中值滤波等图像预处理的方法。(3)通过研究分析薄板常见的五个缺陷特征,选择了几何特征、纹理特征中的九个特征值组成特征向量。(4)在详实对比了各种缺陷分类器特性的基础上,结合实际生产中对于检测速度的要求,最终确定了具有快速处理多类问题优势的贝叶斯分类器来判断缺陷分属问题的划分。(5)通过研究各种软件平台,最终实现了薄板表面缺陷实时检测系统。本系统在实时检测过程中,识别率为83.6%,平均识别1个样本的时间为16毫秒,满足实际工业化生产需要。
[Abstract]:Magnesium is one of the lightest structural metal materials, because of its advantages of light weight, good rigidity, low density, fast heat dissipation and so on, it is widely used in aviation, transportation, chemical industry and other fields. Because of mechanical and control precision, defects such as edge crack, ripple, fold and so on are easy to appear. If this problem can not be solved in time, it will seriously affect the quality of magnesium alloy sheet, so how to carry on the high efficiency to the magnesium alloy sheet surface defect, Rapid detection will be the key to the manufacturing of magnesium alloy sheet. The traditional manual detection method has low efficiency, high error detection rate, and consumes a lot of human resources. Due to the limitation of the principle of infrared detection and magnetic flux leakage detection, the detection rate and identifiable defect types are very limited. In recent years, with the development of computer, automation, artificial intelligence, image recognition and other technologies, The surface defect detection technology with computer vision as the core has become the research focus of magnesium alloy sheet production. In this paper, the surface defect recognition system of magnesium alloy sheet is developed, which can collect the surface image of magnesium alloy sheet in real time. Automatic detection of defects by computer vision and identification of defect types by Bayesian classifier lay the foundation for unattended and highly automated control of the subsequent production process. The main work accomplished is as follows: 1) according to the experimental environment, The actual situation and economic budget of the production line, the hardware and software design scheme of the system is determined, such as selecting LED light source, determining lighting scheme, configuring operating system, etc.) because of the low contrast on the surface of the thin strip, The method of image preprocessing, such as histogram equalization, median filter and so on, is chosen. By studying and analyzing the five common defect features of thin plate, the geometric features are selected. Based on the detailed comparison of the characteristics of various defect classifiers, combined with the requirements of detection speed in actual production, Finally, the Bayesian classifier which has the advantage of fast processing multi-class problems is determined to judge the partition of defect problem. In the process of real-time detection, the recognition rate of the system is 83.6, and the average time of identifying one sample is 16 milliseconds, which meets the needs of practical industrial production.
【学位授予单位】：辽宁科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TG115.28;TP391.41

【参考文献】