基于形状知识的不合格品自动检测算法  

Automatic detection algorithm for defection products based on shape information

作　　者：程俊[1,2] 李嘉翊[1] 郑志刚[1] 徐梦杰[1] 

机构地区：[1]中国科学院合肥智能机械研究所,合肥230031 [2]中国科学技术大学自动化系,合肥230027

出　　处：《电子测量技术》2011年第1期119-123,共5页Electronic Measurement Technology

摘　　要：实现了一种基于形状知识的不合格品自动检测系统的视觉方案。该方案首先利用长球销的形状特点得到球头圆心位置,根据工件的物理参数定位待识别部位的图像坐标,从而有效降低了后续的识别难度。对于特征明显并且不随放置位置变化的部位,设计相应的直接特征提取算法,利用特征检测识别该部位是否加工。而对于特征不明显或者特征可能变化的部位,利用模拟人类视皮层中物体识别机制的机器学习算法自动提取部位特征并利用SVM算法识别。系统的实际运行情况表明,该方案可以快速有效的识别工件是否合格,错误率约为千分之五,基本上达到了人眼的识别率。Have Implemented a machine vision solution of automatic detection system of defection products based on shape information.The solution first gets the centre of the circle on the bulb based on the long ball studs＇s shape information,then use the physical parameters of the work piece to position the image coordinate of the place which needs to be recognized,so it efficiently lowers the recognition difficulty later on.The places with obvious characters that not to be influenced by where they placed,are designed homologous direct feature extraction algorithm to recognize whether processed;While the places without obvious characters or characters that may be changed,we use a machine learning algorithm that simulating object recognition mechanism of human visual cortex to automatic extact characters and use the SVM algorithm to recognize.It proves to be that the solution can fast and efficiently recognize whether the work pieces are acceptable with the running conditions of the system,and it almost reaches the discrimination of human eyes with a error rate of only about 0.5%.

关 键 词：图像测量 模式识别 特征提取 机器学习 R&P模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]