作者：吴俊芳¹, 刘桂雄², 付梦瑶², 王小辉³

作者单位：1. 华南理工大学理学院, 广东 广州 510640;
2. 华南理工大学机械与汽车工程学院, 广东 广州 510640;
3. 广州市光机电计算研究院, 广东 广州 510663

关键词：滤光片;表面缺陷;分类识别;支持向量机

摘要：

针对目前滤光片表面缺陷识别普遍采用人工方式,成本高、无法满足实时性等问题,提出一种基于有向无环图支持向量机(DAG-SVM)的滤光片表面缺陷识别方法。该方法结合滤光片常见缺陷的特点,设计出包含3个结构简单、性能优良的二分类器的滤光片表面缺陷识别方法,克服多分类器算法复杂、难以保证分类正确率的问题。实验结果表明:该方法对滤光片的点缺陷、印子缺陷、划痕缺陷及崩缺陷的识别正确率为100%。

Surface defects classification for optical filters based on support vector machine

WU Junfang¹, LIU Guixiong², FU Mengyao², WANG Xiaohui³

1. School of Science, South China University of Technology, Guangzhou 510640, China;
2. School of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou 510640, China;
3. Guangzhou Research Institute of Optics-Mechanics-Electricity Technology, Guangzhou 510663, China

Abstract: As the surface defects of optical filters are commonly identified by artificial manner which is uneconomical and hysteretic, a new surface defects classification methods for optical filter based on directed acyclic graph support vector machine (DAG-SVM) is proposed. The proposed method takes the characters of filter surface defects into account and comprises three two-classifiers which is simple and performs well. It solves such problems as complex algorithm and lower classification accuracy which occur in multi-classifiers. The experimental result indicates that the proposed method can classify the four common types of filters defects, including point, mark, scratch and broken, with the accuracy of 100%.

Keywords: optical filter;surface defects;classification;support vector machine

2016, 42(2): 92-95144  收稿日期: 2015-6-7;收到修改稿日期: 2015-8-11

基金项目: 广东省产学研结合引导项目(粤财教[2012]393) 中央高校基本科研业务费资助项目(2014ZM0077)

作者简介: 吴俊芳(1977-),女,讲师,博士,主要从事智能传感与检测技术研究。

参考文献

[1] 刁智华,王会丹,魏伟. 机器视觉在农业生产中的应用研究[J]. 农机化研究,2014(3):206-211.
[2] 侯海飞. 基于机器视觉的胶囊表面缺陷的识别与分拣技术研究[D]. 杭州:浙江大学,2013.
[3] 秦玲. 基于机器视觉的无人车道路识别[D]. 西安:西安工业大学,2013.
[4] BENJAMIM X C, GOMES R B. Visual identification of medicine boxes using features matching[C]∥Proceedings of IEEE International Conference on Virtual Environments,2012:43-47.
[5] 光学零件表面疵病:GB/T 1185——2006[S]. 北京:中国标准出版社,2006.
[6] 刘怀广. 浮法玻璃缺陷在线识别算法的研究及系统实现[D].武汉:华中科技大学,2011.
[7] 汪云云. 结合先验知识的分类器设计研究[D]. 南京:南京航空航天大学,2011.
[8] FERENCZ A, LEARNED-Miller E G. Building a classification cascade for visual identification from one example[C]∥Proceedings of the Tenth IEEE International Conference on Computer Vision,2005.
[9] 范劲松. SVM理论及其应用的研究[D]. 合肥:中国科学技术大学,2000.
[10] DAMIANO V, SIMONE D F, DINUZZO. Finding potential support vectors in separable classification problems[J]. IEEE Transactions on Neural Networks and Learning Systems,2013,24(11):799-1813.
[11] CHEN L J, MAO X, XUE Y L. Speech emotion recognition:features and classification models[J]. Digital Signal Processing,2013,22(6):1154-1160.
[12] CHEN C H, GAN Z L Action Recognition from a different view[J]. China Communications,2013,10(12):139-148.
[13] 朱波,刘飞,李顺江. 基于优化有向无环图支持向量机的多变量过程均值异常识别[J]. 计算机集成制造系统,2013(3):559-568.