摘要
为实现织物的瑕疵实时性在线监测,给出一项基于机器视觉的在线检测方法。在系统进行离线训练时,通过无瑕疵织物图像的叠加距离函数及极值权重分析的精确求取得到织物图案纹理基元周期,获得标准基元后,构造无瑕疵标准基元偏移序列,形成模糊分类器;在线检测时,根据构造的偏移序列获得待检测的织物图像块特征,使用模糊分类器进行分类,假如有瑕疵,接着创建精确分类器进行分类。实验结果表明,对任一批图案纹理相同的布匹在线检测,该方法检测一帧图像的平均时间为150ms,准确率达96%以上,实时性高且误检率低。
The machine vision testing method was put forward for the purpose of achieving real-time online testing of the fabric flaws.In off-line training,the superposition distance function and the weight analysis of extremum of the pictures of flawless fabric were used to get the precise elementary cycle of the fabric texture pattern.With the obtained standard elements,the offset sequence for flawless standard elements was established and the fuzzy classifier was formed.In on-line testing,the image block of the untested fabric was captured with the established offset sequence and classified using the fuzzy classifier.If the fabric was flawed,the precise classifier was formed to continue the classification.Experimental results show that the average time on testing one frame of any cloth with the same texture pattern on-line is 150 ms with above 99% accuracy.It has high real-time performance and low fall-out ratio.
出处
《计算机工程与设计》
北大核心
2016年第10期2851-2856,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61070121)
关键词
图案布匹
机器视觉
基元周期
瑕疵检测
分类器
patterned fabric
machine vision
elementary cycle
flaw testing
classifiers
