Automatic discrimination between dusts and digs for large fine optics
-
摘要: 在大口径光学元件表面疵病初检时,灰尘和麻点由于形态类似,不易区分。针对该问题,提出了一种基于模式识别理论的灰尘麻点判别方法。该判别方法以既有的疵病检测系统为基础,根据灰尘麻点的暗场成像特点,选取了合适的特征并根据因子分析理论对特征进行变换,最后基于贝叶斯判别原理对灰尘麻点进行分类。采用自制定标板建立了灰尘麻点训练样本库,并进行多组实验,选取了合适的判别函数,最后进行了对未知样品表面灰尘麻点的区分。实验结果表明,该判别方法的正确率可以达到95%以上。目前此判别方法已经用于惯性约束聚变系统中大口径光学元件表面灰尘与麻点自动区分。Abstract: In surface defects evaluation of large fine optics, dusts and digs are very difficult to be distinguished from each other. In this paper, a pattern-recognition-based automatic discrimination method between dusts and digs is proposed to solve the problem. Based on the existing surface defects evaluation system(SDES), the process of feature selection and feature extraction is described on the basis of factor analysis. The dusts and the digs are classified according to the principle of Bayes discrimination. With the images of the calibration board, a data set that can be used for training and identifying dusts and digs is prepared. In order to obtain the suitable classifier, several comparison experiments are performed. The classifier has also been employed to the unlabeled data to verify the theory. The result shows that the accuracy is above 95%, which greatly reduces the number of false alarm that would otherwise result in remanufacturing. This method has already been applied to classifying dusts and digs for large fine optics in the inertial confinement fusion (ICF) system.
-
Key words:
- surface defect /
- dusts /
- digs /
- pattern recognition /
- factor analysis /
- Bayes discrimination
期刊类型引用(7)
1. 侯劲尧,刘卫国,周顺,高爱华,葛少博,肖相国. 基于卷积神经网络的光学元件表面缺陷图像分类. 应用光学. 2023(03): 677-683 . 百度学术
2. 赵博,史迎馨. 卷积神经网络的高精密光学元件表面缺陷检测. 激光杂志. 2021(11): 185-189 . 百度学术
3. 翁建文,袁银麟,郑小兵,康晴,涂碧海,夏茂鹏,洪津. 多角度偏振成像仪偏振通道响应非一致性测量方法. 光学学报. 2020(08): 177-188 . 百度学术
4. 陆敏,王治乐,高萍萍,郭继锴. 光学元件的疵病检测及现状. 光学仪器. 2020(03): 88-94 . 百度学术
5. 王贵林,朱俊辉,李嘉祥,李治斌. 大口径光学元件表面疵病在位检测与评价研究. 应用光学. 2019(06): 1167-1173 . 百度学术
6. 冯凯萍,吕笑文,张丽文,杨耀雨,陈良奇,江晓亮. 一种新的光学元件表面划痕检测算法. 电脑知识与技术. 2018(01): 230-232 . 百度学术
7. 孙力,刘晨,姚红兵. 基于机器视觉的树脂镜片水印疵病检测. 江苏大学学报(自然科学版). 2018(04): 425-430 . 百度学术
其他类型引用(12)
-

计量
- 文章访问数: 1390
- HTML全文浏览量: 290
- PDF下载量: 435
- 被引次数: 19