QQ客服 官方微博 反馈留言

题名	基于机器视觉的大口径光学元件表面损伤检测技术研究
作者	田玉婷
文献子类	硕士
导师	邬融
关键词	大口径光学元件 Large aperture optics 损伤检测 Defect detection 全内反射 Total internal reflection 暗场成像 Dark-field imaging 局部信噪比 Local area signal-to-noise ratio
其他题名	Research on Surface Damage of Large Aperture Optics Based on Machine Vision
英文摘要	高功率激光装置包含数量众多的终端大口径光学元件，由于光束自聚焦等原因光学元件在高通量激光辐照下会产生激光诱导损伤。光学元件表面疵病将对透过元件的光束造成局部光强调制，降低输出光束质量并对下游光学元件造成破坏。为避免疵病尺寸增长到一定程度导致不可修复问题，需要在疵病初步形成阶段就将其检测出来并进行下架维修。高功率激光系统对光学元件表面质量要求很高，使用前后都需对元件表面疵病进行检测，对元件疵病面积进行统计，为光学元件上线或者下线维修退库提供判据，存储并跟踪疵病坏点增长情况，故实现疵病的低漏检率检测具有重要意义。本文从照明方案配置与疵病识别图像算法展开了研究分析，主要研究内容如下： 针对疵病检测的照明方案配置问题，对全内反射暗场侧向照明的光耦合效率进行了理论分析和定量计算，结合有无散射体等两种方式，引入了散射系数概念，提出了耦合效率相关的照度均匀限定条件，为均匀照明光场的构建提供了理论指导。分析了各种耦合方案形成的照明光场均匀性，同时实验验证了其疵病图像的照度效果。散射系数的大小取决于光线折返次数，增大光源发散角，并在光学元件侧边配置散射体，能大幅提高照明均匀性，由此设计了适合光学元件表面疵病检测的散射式全内反射侧向照明的技术方案。 针对单一全局阈值导致的疵病尺寸误差等问题，为提高疵病的像素尺寸精度，设计了一种多级联合的自适应局部阈值分割算法。首先采用顶帽变换算法抑制高亮复杂非均匀背景，再采用一维最大熵算法筛选并定位出候选疵病，最后再递归地在每个定位区域内分别进行最大类间方差法分割。算法的检测分辨力结合显微镜标定为~150 μm。 为模仿人类视觉机制的优异疵病检测性能从而实现疵病的低漏检率识别，提出了改进的局部信号强度比算法。改进后的局部信号强度比算法对信号图像进行自适应局部增强，可大幅提升损伤目标信号强度值，并抑制非损伤区域信号强度值，显著增强信噪比。改进后的算法具有更强的识别微小低信噪比损伤目标的能力，在全内反射暗场侧向照明方案配置下，可检测出约30 μm的微小损伤。与以往的局部信号强度比算法相比漏检率得到了明显降低，可获得50 μm以上的损伤点小于0.4%的漏检率。; High power laser system involves massive final large aperture optics. Due to self-focusing, optical elements will generate laser-induced damage under high flux laser irradiation. Surface damage will produce local light intensity modulation, which will reduce the output beam quality and cause damage to downstream optics. In order to avoid the size of damage reaching irreversible threshold, it is necessary to detect and repair damage spots at their preliminary formation stage. High power laser system has a high requirement for surface quality of optics. It is necessary to detect damage before and after use, count damage area, provide online and offline criteria, and store the growth of damage. Thus it is very important to recognize damage with low missing detection rate. In this paper, the research about lighting scheme configuration and damage recognition algorithm were carried out. Main research contents of this article are as follows: To solve lighting scheme configuration problem, coupling efficiency with/without scatterer under total internal reflection dark field edge illumination was analyzed and quantitatively calculated. Scattering coefficient and uniform illumination condition were proposed, which can provide theoretical guidance for the construction of uniform illumination. The illumination uniformity formed by various coupling schemes was analyzed and verified experimentally. Scattering coefficient depends on backtracks of beams. Increasing divergent angle of light source and placing scatterers around optical element side can substantially improve illumination uniformity. The technical scheme for total internal reflection edge illumination with scatterers suitable for surface flaws detection was designed. In order to improve the pixel size accuracy of damage, a multilevel joint algorithm based on adaptive local threshold segmentation was designed to solve damage size error caused by a single global threshold. First Top-hat algorithm was used to suppress complex background. Then the one-dimensional maximum entropy algorithm was used to screen and locate candidate defects. Finally, Otsu algorithm was performed within each located region recursively. According to the measurement results of microscope, the detection resolution of this algorithm is ~150μm. In order to imitate the excellent damage detection performance of human vision mechanism and recognize damage with low missing detection rate, improved local area signal strength ratio algorithm was proposed. The improved algorithm can achieve adaptive local enhancement of signal image. Target signal enhancement and non-target signal suppression can be achieved simultaneously. In particular, the signal-to-noise ratio can be improved significantly. Our improved algorithm has stronger ability to identify small target with low signal-to-noise ratio and can recognize damage with about 30μm under total internal reflection lateral illumination in dark field. The missing detection rate is lower than that of the existing local area signal strength algorithm. The results show that the missing rate of our method is below 0.4% for damage points with diameter above 50μm.
学科主题	光学工程
内容类型	学位论文
源URL	[http://ir.siom.ac.cn/handle/181231/31067]
专题	中国科学院上海光学精密机械研究所
作者单位	中国科学院上海光学精密机械研究所
推荐引用方式 GB/T 7714	田玉婷. 基于机器视觉的大口径光学元件表面损伤检测技术研究[D].

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。