基于机器视觉的磁控溅射薄膜件的划痕检测

Scratch detection of magnetron-sputtered parts based on machine vision

针对目前磁控溅射薄膜件人工检测效率低的问题,结合企业需求,设计了一种基于机器视觉技术的磁控溅射薄膜件划痕自动检测系统,实现对产品表面各种划痕的自动检测。首先对采集到的图像进行校正,利用标准图像创建模板,通过形状模板匹配提取样本图像中感兴趣的区域。然后将图像分解成低频和高频两部分,对低频部分进行先膨胀后腐蚀的操作,对高频部分进行中值滤波以去除噪声,图像合成后通过取反和灰度缩放生成增强图像;最后采用二维Otsu拟合线阈值分割法对图像进行分割,采用选取指定形状特征和骨架提取法对划痕缺陷进行提取和测量。对100件样品进行多次随机验证的结果表明,本文设计的系统能快速、准确、高效地提取并测量划痕缺陷,每个样品检测的平均耗时仅0.25 s,正检率达98.5%。该方案与人工检测比,速度更快、准确度更高,可以满足工业应用的需求。

Aiming at the problem of low efficiency during manual inspection of the magnetron-sputtered parts,an automatic inspection system based on machine vision technology was designed to achieve the automatic detection of scratches on the surface of magnetron-sputtered part with the needs of a plant.Firstly,the captured images ware corrected,and the region of interest was extracted from the sample image through shape template matching with a template which was created from the standard image.The image was then decomposed into two parts:low-frequency part and high-frequency part.The dilation and erosion were carried out on the low-frequency part,and the median filter denoising operation was performed on the highfrequency part.After the image is synthesized,an enhanced image was generated through inversion and gray scaling.Finally,the image was segmented by two-dimensional Otsu thresholding method,and the scratch defects were extracted and measured by shape feature specifying and skeleton extraction.According to the results of multiple random verifications on 100 samples,the system designed in this paper is able to extract and measure scratch defects quickly,accurately,and efficiently.The average detection time for each sample was 0.25 s,and the positive detection rate was 98.5%.As compared with manual inspection,this solution is faster and more accurate,and meets the needs of industrial applications.