摘要
针对形状复杂、型式多样的触点类元件自动检测问题,设计了一种具有自学习功能的智能视觉检测系统.该系统分为自学习和检测2个过程,通过对一定数量的学习样本提取包括几何、纹理及结构特征在内的大量基础特征矢量,使用信息论进行特征选择和变换,形成一组能够使类间尽可能分开的新的特征矢量—典型变量,并将这些特征应用支持向量机(SVM)来训练识别算法.实验结果表明,该系统能够适应触点类复杂零件检测,且检测准确率高,能够满足智能检测和工程实时性要求.
An intelligent vision inspection system with self-learning function was designed for automatic inspection of contact components with complex shapes and different patterns. The system is divided into selflearning process and inspection process. By learning a large number of samples＇ basic feature vectors, in- cluding geometric, texture and structural features, the system uses the information theory for feature selection and transformation, and forms a group of canonical variants to make the classes as far as possible. Support vector machine（SVM） trains these new features to accomplish the recognition algorithm. The experimental results show that the system can adapt to the inspection of complex contact components with high detection accuracy, it is powerful to process the intelligent inspection and can meet the real-time requirements of project as well.
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
2009年第12期1936-1940,共5页
Journal of Shanghai Jiaotong University
关键词
视觉检测
触点
自学习
信息论
支持向量机
vision inspection
contact component
self-learning
information theory
support vectormachine（SVM）
