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

Abstract:

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%.

Key words: cheese yarn, image processing, defect detecting, edge extraction, multi-scale analysis, template convolution