基于机器视觉的光学元件损伤在线检测研究

发布时间：2018-05-19 03:15

本文选题：损伤在线检测 + 机器视觉　；参考：《中国科学院大学(中国科学院光电技术研究所)》2017年硕士论文

【摘要】：激光持续辐照下,容易诱导光学元件发生损伤,极大地增加了系统研制成本,还会限制系统输出能力,因此有必要对光学元件损伤进行在线检测。若发现光学元件损伤,应在损伤急速增长之前停止辐照,避免出现不可逆或炸裂性损伤。本文主要研究基于机器视觉的光学元件损伤在线检测方法,通过提取在线采集的光学元件表面图像的特征,评估光学元件表面损伤情况,并对检测系统中用到的关键图像处理技术进行了深入研究。论文首先基于DirectShow视频采集框架搭建了图像采集和处理系统,对视频采集中用到的主要方法和原理进行详细叙述。设计并搭建了光学元件损伤在线检测系统的图像采集和处理软件框架,主要包括损伤图像采集、处理、配置、通信等功能。其次研究图像相似理论,根据激光辐照前后,光学元件损伤图像相似程度,快速、实时检测光学元件损伤,提出将广泛用于图像质量评价的结构相似测度算法用于光学元件损伤检测,该算法可以准确描述光学元件损伤轻重程度。最后,论文研究光学元件损伤检测中的图像分割算法,利用图像分割算法对损伤区域进行特征提取,统计损伤区域面积、长短轴等信息,定量分析光学元件损伤情况。在实际应用中,损伤图像因为光照不均以及各种复杂多变的损伤效应,严重影响图像分割和检测效果。因此,针对本文实验中在线采集到的损伤图像的噪声特点,研究各种滤波算法。在中值滤波、形态学滤波效果都不理想的情况下,最终选用基于图像退化模型的图像复原方法,利用大气散射模型对退化过程进行近似,仅通过一次均值滤波对退化参数和噪声进行估计,就能有效去除噪声。在此基础上,为了实现快速、准确的分割,论文还进一步对传统的最大类间方差法(Otsu)算法进行改进。Otsu算法实现简单,性能稳定,是应用最广泛的图像分割方法,但是当损伤图像存在微小损伤点时分割性能下降,因此在使用全局Otsu算法确定最佳分割阈值T后,计算每个像素局部信噪比,对0~T范围的像素进行搜索,对信噪比超过设定阈值的像素,再次使用Otsu进行局部阈值分割。对本文提出的算法,采集本文实验系统在运行过程中的实时图像进行实验,实验结果表明本文算法均可有效实现光学元件损伤在线检测。
[Abstract]:The continuous laser irradiation can easily induce the damage of optical elements, which greatly increases the cost of system development and limits the output ability of the system. Therefore, it is necessary to detect the damage of optical elements on line. In this paper, the on-line damage detection method of optical components based on machine vision is studied, and the surface damage of optical components is evaluated by extracting the features of the surface images of optical components collected on line. The key image processing techniques used in the detection system are also studied. Firstly, the image acquisition and processing system is built based on DirectShow video capture frame, and the main methods and principles used in video capture are described in detail. The image acquisition and processing software framework of the online optical component damage detection system is designed and built, which mainly includes the functions of image acquisition, processing, configuration, communication and so on. Secondly, the theory of image similarity is studied. According to the similarity degree of damage image of optical element before and after laser irradiation, the damage of optical element can be detected quickly and in real time. A structural similarity measure algorithm, which is widely used in image quality evaluation, is proposed for damage detection of optical elements. This algorithm can accurately describe the severity of damage of optical elements. Finally, the paper studies the image segmentation algorithm in the damage detection of optical elements, uses the image segmentation algorithm to extract the feature of the damage area, statistics the area of the damage area and the information of the long and short axis, and analyzes the damage situation of the optical element quantitatively. In practical applications, image segmentation and detection are seriously affected by uneven illumination and various complex and varied damage effects. Therefore, according to the noise characteristics of the damage images collected online in this paper, various filtering algorithms are studied. When the median filter and morphological filter are not ideal, the image restoration method based on image degradation model is used to approximate the degradation process by using atmospheric scattering model. The noise can be effectively removed by estimating the degradation parameters and noise only by means of the first-order mean filter. On this basis, in order to achieve fast and accurate segmentation, this paper further improves the traditional Otsuu algorithm, which is simple and stable, and is the most widely used image segmentation method. But the segmentation performance is degraded when there are small damage points in the damaged image, so after using the global Otsu algorithm to determine the optimal segmentation threshold T, the local signal-to-noise ratio of each pixel is calculated, and the pixels in the 0 T range are searched. For pixels whose signal-to-noise ratio exceeds the set threshold, Otsu is used to segment the local threshold again. For the algorithm proposed in this paper, the real-time images of the experimental system in this paper are collected and the experimental results show that the proposed algorithm can effectively realize the on-line damage detection of optical elements.
【学位授予单位】：中国科学院大学(中国科学院光电技术研究所)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH74;TP391.41

【相似文献】