基于多元图像分析的包装罐内壁缺陷检测  

Packaging Cans Inner Surface Inspection System Based on Multivariate Image Analysis

作　　者：王宣银[1] 梁冬泰[1] 

机构地区：[1]浙江大学流体传动及控制国家重点实验室,杭州310027

出　　处：《农业机械学报》2009年第6期222-226,共5页Transactions of the Chinese Society for Agricultural Machinery

基　　金：教育部新世纪优秀人才支持计划资助项目(NCET-04-0545)

摘　　要：为提高包装罐生产线内壁缺陷检测准确性与可靠性,研究了一种采用单摄像机的内壁缺陷检测系统。利用基于形态学的区域提取算法,从罐内图像中分割出内壁检测区域图像。提出基于多元图像分析(MIA)的内壁缺陷检测算法。利用图像融合构成环形合格样本图像,消除罐内焊缝区域的影响,把多个环形合格样本图像与测试样本内壁检测区域图像堆叠起来,用重合区域的图像构造多元测试图像。用基于主成分分析(PCA)的多元图像处理方法获得多元测试图像的主分量表示,将去掉第一主分量和噪声后的Q统计图像作为内壁缺陷特征的检测空间,利用阈值处理检测缺陷,解决了罐体内壁照明困难、亮度不均造成缺陷误检率高的问题,提高了检测系统的准确性和鲁棒性。实验表明对内壁缺陷检测的误检率降低到2%,验证了检测系统的有效性和可靠性。In order to improve the accuracy and reliability of defect detection for packaging cans production lines, an inner surface inspection system with a single camera was studied. By using morphological region extraction algorithms the inspection region of interest image can be obtained from the whole inner image. For defect feature detection an approach based on multivariate image analysis （MIA） was proposed. To cancel the effect of seam regions in the inner images, a method of images fusion was implemented to form the ring-like good sample image without seam region. By stacking both the ringlike good sample images and the test image, the multivariate test images were constructed with their overlapping part. By using MIA technique with principal component analysis （PCA）, the principal component scores of the multivariate test images were obtained. As the feature space for defect detection the Q-statistic image was derived from the residuals which were left after the extraction of the first PC and noise. The surface defects can be effectively detected using an appropriate threshold. The experimental results show that the proposed inspection system has less sensitivity to the inhomogeneous of illumination, and has more robustness and reliability with pseudo reject rate reducing to 2%.

关 键 词：包装罐 机器视觉 缺陷检测 区域提取 多元图像分析 Q统计量 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TS292[自动化与计算机技术—计算机科学与技术]