基于机器视觉的磁环尺寸及表面缺陷检测系统研究与开发

发布时间：2018-03-26 18:24

本文选题：机器视觉　切入点：磁环　出处：《浙江理工大学》2017年硕士论文

【摘要】：为了满足高效率、高精度和智能化的要求,作为非接触式检测手段之一,以数字图像处理技术为基础的视觉检测技术正越来越多的应用于工业生产的各个领域。基于此背景,本文对磁环尺寸及表面缺陷检测方法进行了研究,提出了一套基于数字图像处理技术的磁环尺寸及表面缺陷检测算法并将其应用于磁环在线检测系统中。根据检测要求和生产环境,设计了一套磁环缺陷在线检测系统,该系统包括机械部分、视觉部分、控制部分和应用软件。机械部分完成上料、传送、翻转等任务;视觉部分完成原始图片的采集;控制部分接收分捡信号剔除次品磁环;应用软件负责对原始图像进行处理和识别并向控制部分发送分捡指令。首先对采集的磁环图像,研究了空域滤波、形态学处理、阈值分割等预处理方法。对常用的空域滤波方法进行了对比,采用中值滤波对原始图像进行处理。基于二值化和形态学轮廓提取算法提取磁环的轮廓序列,采用最小旋转矩形和一种新的Hough椭圆拟合算法分别拟合磁环的侧面轮廓和正面轮廓,在对相机标定的基础上,通过拟合几何图形可测量磁环尺寸。基于拟合椭圆尺寸信息采用区域掩膜法提取出磁环前景区域。将前景图像做次峰灰度级到背景灰度级的映射,在此基础上,用多种自适应阈值分割方法对磁环缺陷进行阈值分割并做了比较。针对磁环表面纹理对缺陷分割造成干扰的问题,本文提出了基于小波分析纹理弱化和自适应Canny的边缘分割方法。利用小波变换的多尺度分析思想,对磁环图像进行小波变换并提取近似系数;在研究磁环表面的基于灰度共生矩阵和基于灰度直方图的统计特征时发现,磁环表面的平均纹理熵值和Canny算子参数间存在一定的函数关系,通过此函数关系自适应确定Canny算子的高低阈值;在小波域中对图像近似部分作边缘检测,检出磁环和缺陷的边缘,对边缘作一定的形态学和修边处理,并和上述经形态学处理过的轮廓作异或运算得到磁环缺陷边缘,对上述缺陷边缘判断闭合并填充即可得到磁环的缺陷区域。不同光照下的实验表明本文算法光照鲁棒性较强,检测率实验表明,缺陷综合正检率达94.7%。基于VS2013开发环境,在Sapera LT软件库和OpenCV计算机视觉库的基础上开发了磁环表面缺陷检测应用软件,以MFC为基础搭建应用程序界面。在上位机软件和分捡气缸等控制部分的配合下,能够对生产线上的磁环缺陷进行精确判定和自动分捡,实现了磁环尺寸及表面缺陷的智能检测。
[Abstract]:In order to meet the requirements of high efficiency, high precision and intelligence, as one of the non-contact detection methods, the visual detection technology based on digital image processing technology is more and more used in various fields of industrial production. In this paper, the size of magnetic ring and the detection method of surface defects are studied. Based on digital image processing technology, a magnetic ring size and surface defect detection algorithm is proposed and applied to the magnetic ring on-line detection system. According to the detection requirements and production environment, a magnetic ring defect online detection system is designed. The system includes a mechanical part, a visual part, a control part and an application software. The mechanical part completes the tasks of feeding, transmitting and flipping, the visual part completes the collection of the original picture, the control part receives and picks up the signal and removes the defective magnetic ring. The application software is responsible for processing and recognizing the original image and sending the sorting instruction to the control part. Firstly, the spatial filtering and morphological processing are studied for the collected magnetic ring image. Threshold segmentation and other preprocessing methods. The common spatial filtering methods are compared, the median filter is used to process the original image, and the contour sequence of the magnetic ring is extracted based on binary and morphological contour extraction algorithms. The minimum rotation rectangle and a new Hough ellipse fitting algorithm are used to fit the profile and the front profile of the magnetic ring respectively. The size of magnetic ring can be measured by fitting geometry. The region mask method is used to extract the foreground region of magnetic ring based on the fitting ellipse dimension information. The foreground image is mapped from sub-peak gray level to background gray level. In this paper, several adaptive threshold segmentation methods are used to segment the defect of magnetic ring and compared. Aiming at the problem that the surface texture of magnetic ring interferes with the defect segmentation, In this paper, an edge segmentation method based on wavelet analysis texture weakening and adaptive Canny is proposed. It is found that there is a functional relationship between the average texture entropy of the magnetic ring surface and the parameters of the Canny operator when studying the statistical characteristics of the magnetic ring surface based on the gray level co-occurrence matrix and the gray histogram. According to the function relation, the threshold of Canny operator is determined adaptively, the edge of the approximate part of the image is detected in the wavelet domain, the magnetic ring and the edge of the defect are detected, and the edge is treated with morphology and edge-trimming. The defect edge of the magnetic ring can be obtained by using the morphological contour and the defect region of the magnetic ring can be obtained when the defect edge is closed and filled. The experiments under different illumination conditions show that the algorithm is robust to illumination. The test results show that the comprehensive positive rate of defect detection is 94.70.Based on the VS2013 development environment, the application software of magnetic ring surface defect detection is developed on the basis of Sapera LT software library and OpenCV computer vision library. The application program interface is built based on MFC. With the cooperation of the upper computer software and the separation cylinder, the magnetic ring defects in the production line can be accurately determined and automatically picked up, and the intelligent detection of the magnetic ring size and surface defects can be realized.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】