应用多尺度多方向模板卷积的筒子纱缺陷检测  

Defect detection of cheese yarn based on multi-scale multi-direction template convolution

作　　者：蔡逸超 周晓[1] 宋明峰 牟新刚[1] CAI Yichao;ZHOU Xiao;SONG Mingfeng;MOU Xin′gang(School of Mechanical and Electronic Engineering,Wuhan University of Technology, Wuhan,Hubei430070,China)

机构地区：[1]武汉理工大学机电工程学院,湖北武汉430070

出　　处：《纺织学报》2019年第4期152-157,共6页Journal of Textile Research

基　　金：国家自然科学基金项目(61701357);中央高校基本科研业务费专项资金资助项目(183204007)

摘　　要：针对筒子纱不规则的纹理基元增加缺陷检测难度的问题,提出一种基于多尺度多方向模板卷积的筒子纱表面网纱缺陷检测算法。首先通过系统采集筒子纱的顶面纹理图像;其次对原图像进行OTSU阈值处理,并采用椭圆拟合确定纹理位置,利用极坐标变换将其展开;然后利用水平方向一维高斯差分算子提取边缘,通过改变高斯核的标准差获取多个尺度的边缘图像;进而将0°～180°角度区间量化成多个方向模板与各尺度的边缘图像进行卷积,并投票记录每个像素位置在各尺度下的多方向模板卷积结果的最大值;最后,选取经验阈值对最强卷积响应进行分割得到缺陷图像。实验结果表明,该方法可有效检测筒子纱表面的网纱缺陷,识别准确率达0. 96。Aiming at the hard defect detection due to coarse and irregular texture element of cheese yarn,a novel cheese yarn surface defects detection algorithm based on multiscale multi-direction template convolution was proposed.Firstly,the image of top textile of cheese yarn was acquired by the system.Secondly,the OTSU threshold method and the ellipse circle fitting method were used to determine the position of the textile region of cheese yarn.Then,the polar coordinate transformation was used to expand the ring-shape surface to a rectangle.After that,the one-dimensional difference of Gaussian operator in horizontal direction was used to extract the edge.Multiple scale edge images were extracted by varying the standard deviation of Gauss kernel.Furthermore,the angle range of 0° to 180° was quantified into several directional template images used to convolute with edge images,then the maximum convolution result of each pixel in among all scales was elected by voting method.Finally,the defects were segmented out from the maximum correlation image by an empirical threshold according to prior information.The experiment results show that the method can effectively detect the net-yarn surface defects,and the recognition accuracy reaches 96%.

关 键 词：筒子纱 图像处理 缺陷检测 边缘提取 多尺度分析 模板卷积 

分 类 号：TS103.7[轻工技术与工程—纺织工程] TP274.3[轻工技术与工程—纺织科学与工程]