【作者】 刘勇；

【导师】 吴援明； 彭真明；

【作者基本信息】 电子科技大学 ， 信号与信息处理， 2017， 硕士

【摘要】 工业缺陷检测是工业生产必不可少的环节,是提高产品质量、提升品牌形象的重要途径。中国是世界最大的制造业国,然而如今的缺陷检测手段仍然以人工为主,与发达国家有很大差距。随着人口红利的消失,工业缺陷检测自动化将是未来的必然趋势。镜面表观裂纹作为常见缺陷的一种,因为受到光照不均,只在特定视角可见等原因,使得常规方法难以检测。本课题针对镜面表面常见缺陷,设计了一套基于图像处理的检测方法,切实可用的对产品进行等级分类,实现了对产品的高速、高效检测。本文工作主要包括四个部分:(1)设计了一种基于软件指导硬件照明设计的方案。通过Phong光照模型的分析,模拟出可能的打光条件,得出针对镜面表面的合适光照系统,即同轴光照垂直照明可得到均一稳定的背景,并可以突出缺陷位置,解决了光照不均,以及在特定视角可见等问题。(2)研究了一种基于边缘检测与形态学结合的工件提取方法。通过高分辨率工业CCD采集到的图像,对图像进行划痕、孔洞、黑点等缺陷进行检测。算法首先通过光照补偿解决光照轻微不均问题,然后通过Canny边缘提取结合形态学处理的方法分割出工件所在位置,锁定检测区域。(3)采用了基于几何特征的缺陷检测方法。对于工件区域提取Canny边缘,形态学膨胀算法,形态学提骨架算法提取完整缺陷形态。然后应用形态学腐蚀,连通分量提取统计,得出疑似缺陷。(4)设计了针对无纹理缺陷的形态学缺陷分类方法。针对疑似缺陷,首先根据几何参数,粗分类为线状缺陷与孔状缺陷,然后应用产品等级分类标准表,每类缺陷再进行细分,最终划分出产品等级,决定产品是否合格,不合格输出不合格原因。更多还原

【Abstract】 Industrial defect detection is a key part in industrial production,which improves product quality,increasing brand reputation greatly.China is the world’s largest manufacturing country,but defect detection work is still mainly artificial at present,and there is still a big gap between the developed countries.With the rise of human resources cost,automatic industrial defect detection will be an inevitable trend in the future.As one of the common defects,the specular surface defects are difficult to detect because of non-uniform illumination,and sometimes it is visible only at a certain angle of view.Aiming at detecting defects on specular surface,this paper designs a set of practical system which can classify the products based on the image processing method.It realizes the high-speed and high-efficiency detection of the products.This thesis is divided into four parts:(1)Illumination design.Through Phong model,simulations of lighting are carried out for better images.In trials and errors,finally we obtained an almost perfect solution,which is vertical illumination with coaxial light,providing uniform and stable background and highlighting the defects regardless of different angles.(2)Product area extraction.Through the high-resolution industrial CCD,the algorithm first solves the problem of light unevenness by illumination compensation.Then get the product location by Otsu,which shrinks the detection area.(3)Defect detection.Canny edge detection,morphological dilation algorithm and morphological skeleton algorithm are used to extract the complete defect target.Then the morphological corrosion and connectivity component extraction statistics were used to obtain the suspected defects.(4)Defect recognition.According to the geometric parameters,rough classification is classified as line defect and hole defect,and then the product grade classification standard table is used to subdivide each kind of defect,and the product grade is finally divided to decide whether the product is qualified,and output the failure reason.更多还原