摘要
针对具有一致性或存在特定模式的产品图像中可能出现的偏移数目未知的问题,提出将图像划分为不可重叠、大小相同的小区域,计算各区域的灰度均值,对所有区域灰度均值的广义似然比统计量求和来构建控制图,从而实现了对该类产品图像的实时监控.利用仿真试验对不同区域大小下控制图的性能进行了对比分析.结果表明,当划分的区域大小与目标偏移的范围一致时,不论图像中出现单个或多个偏移,控制图的检测效果最优.
In order to solve the unknown number of faults problem in monitoring of images with uniformity or a specific pattern, a new control charting method is proposed. Each image is divided into non-overlapping,regular regions of equal size first. Then the mean intensity for each region is calculated. The control chart is built based on the sum of the generalized likelihood ratio for each mean intensity. Performance of the control chart under different region sizes is analyzed and compared through computer simulations. It is shown that the chart achieves its best detection performance for both single and multiple faults when the region size is set to approximately the size of target fault to be detected.
出处
《系统工程学报》
CSCD
北大核心
2016年第1期127-134,共8页
Journal of Systems Engineering
基金
国家自然科学基金杰出青年资助项目(71225006)
关键词
图像数据
广义似然比
过程控制
image data
generalized likelihood ratio
statistical process control
