基于机器视觉的电磁阀表面缺陷检测技术研究

Study on Surface Defect Inspection of an Electromagnetic Valve Based on Machine Vision

摘要: 传统的瑕疵检测多是依靠人力来完成，经由操作人员对成品的检查来判断是否为瑕疵产品，并且记录缺陷种类。本文以机器视觉为基础，设计环境光源并使用CCD取像，针对电磁阀上不同的缺陷，分别于空间域以及频率域，进行电磁阀的缺陷检测。在空间域中使用的图像形态学检测方法，可以快速的检测图像中的轮廓缺陷，但是对于具有方向性、细长的表面刮痕缺陷判断的准确性较低。因此，本文将空间域中较难判定的表面缺陷，利用傅立叶变换将图像从二维的空间域转换到频率域，利用傅立叶频谱突显电磁阀检测表面上的方向性纹路，进行表面缺陷检测。本文的实验结果发现表面有缺陷的图像会在傅立叶频谱中产生较明显的差异，因此利用二值化图像处理将此差异变大，并利用傅立叶逆变换将傅立叶频谱转换成空间域的图像，并比较差异。本文中结合蓝色和白色LED条型灯的照明系统，分别于空间域和频域中进行电磁阀的缺陷检查。

Abstract: Defect detection is mostly done by manpower. In the past, products were inspected to determine whether they were defective products by manual means, and the types of defects were recorded. A machine vision based solenoid valve defects inspection system was proposed in this work. This system is based on capturing images according to the different defects in the specimens, as well as inspection defects and frequency domains by using a CCD camera along with a machine vision algorithm and designs a lighting system. The inspection method first used an erosion and dilation algorithms of image morphology in the spatial domain. Mathematical logic operation was employed to retain the features of the image obtained by different algorithms. Although morphological image detection methods can quickly inspect defect contours, they have difficulty to inspect directional features i.e., thin scratches on specimen surfaces. Therefore, Fourier transform was applied to convert images from two-dimensional spatial domains as the frequency domain to inspect the defects that could not be inspected by the morphological method in the spatial domain. In the frequency domain, standard deviation was used. Therefore, defects were difficult to determine in the spatial domain, this paper uses Fourier transform to transform the image from the two-dimensional space domain to the frequency domain, and uses the Fourier spectrum to highlight the directional grain on the detection surface of the solenoid valve for surface defect detection. The experimental results show that defective images produce more noticeable differences in the Fourier spectrum. This difference is magnified using binary image processing, in which inverse Fourier transform transforms the Fourier spectrum into images in the spatial domain, and these images are compared. In this paper, the lighting system combining blue and white LED strip lights is used to inspect the defect of the solenoid valve in the space and frequency domains respectively.