基于机器学习的光学元件表面疵病检测-计算机应用技术专业论文.docxVIP

西南科技大学硕士研究生学位论文 西南科技大学硕士研究生学位论文 第1页 摘 要 疵病检测是精密光学元件质量检测的主要方面。本文在国家自然科学基 金项目“强激光光学元件表面疵病检测技术研究的资助下，对大口径光学 元件表面疵病检测展开了研究。 在暗室环境下采用侧照明方式对大口径光学元件进行照明，以获取大小 两个视场不同精度暗背景下的亮疵病图像。分析了疵病图像的噪声来源以及 光照对疵病成像的影响，设计了一种有效去除疵病噪声的算法以及光照非均 匀性校正的快速算法。根据疵病小且分布不均的特点选取了合适的分割算法。 根据现有的图像几何特征提取算法总是针对某一个或几个参数进行设计，不 具通用性，且很多算法对单像素宽的非环区域不能检测的缺点，提出了一种 基于顶点链码与离散格林相结合的图像几何参数提取统一计算框架，该计算 框架能同时计算多个几何参数，能把对目标区域的曲面积分变为曲线积分， 其运算速度得到了极大的提高。通过对比研究大小两个视场下的疵病尺寸来 获取疵病的物理尺寸。使用机器学习的方式对疵病进行分类，并取得了良好 的效果，分类识别率达到了90％左右。 关键词： 疵病滤波顶点链码离散格林机器学习 AbstractDefects Abstract Defects detection is an important aspect of quality control for precision optical component．In this paper，Funded by the Nmional Natural Science Fund Project”Strong laser optics Surface Defects Detection，”this paper gives some research to the large—diameter optical components for surface defect detection，as follows． Applying side lighting method on large—diameter optical components in darkroom environment，two bright defects images with different views and different precision in dark background are obtained．After the analysis of noise sources of Defects imaging process and the impact of the light，an effective and fast algorithm is design for the removal of defects noise and light non．uniformity． The appropriate segmentation algorithm is chosen，according to the characteristics of small defect disease and uneven distribution．The current image geometric feature extraction algorithms are always designed for one or several parameters；SO they are non—universal．Many algorithms can not detect the non．ring region of single。pixel wide．To solve these problems，this paper proposes a fast algorithm for calculating geometric Feature Parameters of the image．Combined the direction of Vertex chain code and the Discrete Green Theorem，surface integral is turned into line integral，SO that the calculation of the geometric information is unified in this framework．Experiments show that the new method can calculate various geometric parameters，and its computational speed has