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。

With the continuous irradiation of a laser, some optical components are easy to be damaged. Induced damage not only greatly increases the cost of the system, but also limits the output ability of the system. Therefore, it is necessary to detect the damage of optical components on line. When the damage of optical components was detected, the system should stop light before the damage increase sharply and deteriorate to avoid the irreversible and disastrous damage. This paper has mainly researched the online inspection method for optical components based on machine vision through extracting the feature from the image acquired by CCD to estimate the damage, and we has deeply researched the key technology of the image processing of the system.

Firstly, this paper has built an image acquisition and processing system based on the Direct-show, and the procedure and method of the video acquisition framework was narrated detailed. The image acquisition and processing software has designed, which includes image acquisition, processing, configuration, communication and other functions.

Second, this paper has studied the image similarity theory, and the damage of optical components could be inspected rapidly and in real time according to the similarity assessment of image before and after laser irradiation. The structural similarity measure algorithm (SSIM) was applied to accurately describe the optics damage degree in this paper, which has been widely used in image quality assessment.

Finally, the image segmentation algorithm has been researched for online damage inspection of optical components. Some features have been extracted from the damaged image by using image segmentation algorithm. Then we can calculate the area and major axis to analysis the optics damage degree quantitatively. However，the uneven illumination and complex damage effects in actual application may affect the result of image segmentation and damage inspection. Therefore, according to the noise characteristics of damaged image, various filtering algorithms have been researched. With the bad performance of median filter and morphology filter, the image restoration technology based on a degradation model is be used. The atmospheric scattering model has been used to approximate the degradation process of image, so the image noise can be reduced effectively by using a mean filter to estimate the degradation parameters and noise. Furthermore, in order to achieve fast and accurate segmentation, the traditional Otsu algorithm was further improved in this paper. It is the most popular image segmentation method with the advantages of easy implement and steady performance. However, its performance was decreased when some tiny object points appear in damaged image. Therefore, after getting the optimal segmentation threshold T by using the Otsu, we compute the local signal-to-noise of each pixel of the image, and search for the pixels whose gray-level is ranged at 0~T. When the local signal-to-noise of searched pixel exceeds the threshold, we use a local Otsu to segment the tiny object again.

We acquired some online damage image of the experiment to test the performance of the algorithms proposed, the experimental results verify the validity of the online damage inspection.