摘要
为实现经编机织布过程中布匹瑕疵的实时检测,提出了一种基于机器视觉的实时检测方法。离线训练时分别学习有瑕疵和无瑕疵纹理布匹图像,自动求取纹理基元周期和纹理方向,用以构建实用的两方向Gabor滤波器组,进而提取有和无瑕疵图像特征。在线检测时,以离线所构建的Gabor滤波器组分解图像,以离线所求取的参数窗口化Gabor子图,进而提取子图特征并采用特征变化率来代替原始特征的方法以消除光照不均影响。实验表明,该方法可以适应不同纹理布匹检测需求,消除光照影响,布匹检测准确率高达99%,检测一帧(54 pixel×600 pixel)的平均时间为100 ms,实时性和准确性高,可实现经编机布匹瑕疵的在线实时检测。
To achieve real-time detection of fabric defects on warp knitting machine in the process of weaving, a machine vision-based rapid real-time detection method is proposed. In off-time case, it extracts the texture parameters for the estab-lishment of practical Gabor filter banks. In real-time case, it decomposes the images using the Gabor filter banks, partitions the Gabor sub-graph by the extracted texture cycle and extracts the features of every partition, then proposes a discriminant method of defects feature to solve the problem of uneven illumination. Experiments demonstrate that the method can adapt to the detection needs of fabrics with different textures and is insensitive to the effects of light;the accuracy rate is above 99%;the average time of detection is about 100 ms（based on images of 54 pixel ＆#215; 600 pixel）;the method is with high instantaneity and high accuracy and can achieve real-time detection of fabric defects of warp knitting machine.
出处
《计算机工程与应用》
CSCD
北大核心
2015年第9期185-190,共6页
Computer Engineering and Applications
基金
江苏高校优势学科建设工程资助项目(PAPD)
江苏省产学研前瞻性联合研究项目(No.BY2012056)
