Woven Fabric Density Measurement by Using Multi-Scale Convolutional Neural Networks
Woven Fabric Density Measurement by Using Multi-Scale Convolutional Neural Networks
复制标题
DOI：
10.1109/access.2019.2922502
复制
发表时间：
2019-06
期刊：
影响因子：
3.9
通讯作者：
Shuo Meng;R. Pan;Weidong Gao;Jian Zhou;Jingan Wang;Wentao He
中科院分区：
文献类型：
--
作者：
Shuo Meng;R. Pan;Weidong Gao;Jian Zhou;Jingan Wang;Wentao He
织物密度测量在织物结构参数分析中起着关键作用。现有的自动测量方法适应性差，在实际应用中效果不佳。为了解决这些问题，我们使用卷积神经网络（CNN）来定位机织物密度测量的经纱和纬纱。首先，我们使用一个便携式无线设备来捕获高分辨率的织物图像，并建立一个新的数据集与标记的纱线位置。基于这个数据集，我们提出了一个有效的多尺度卷积神经网络（MSnet）架构来定位经线和纬线。然后，利用Hough变换和预测纱线位置的图像投影，准确测量织物密度。实验结果表明，该方法在各种图案和织物密度下均达到了较高的识别精度，在准确性和鲁棒性方面上级现有的方法。该方法可以为更多的织物结构参数分析提供新的思路。
Fabric density measurement plays a key role in the analysis of fabric structural parameters. Existing automatic measurement methods lack varieties of adaptability and present poor performance in practical application. In order to solve these problems, we use convolutional neural networks (CNNs) to locate warps and wefts for woven fabric density measurement. First, we use a portable wireless device to capture high-resolution fabric images and set up a new dataset with labeled yarns location. Based on this dataset, we propose an effective multi-scale convolutional neural network (MSnet) architecture to locate warps and wefts. Then, by using Hough transform and image projection of predicted yarns location, the fabric density is measured accurately. The experimental results emphasize that the proposed method has reached high accuracy under various kinds of patterns and densities of the fabrics and is superior to the state-of-the-art methods in terms of its accuracy and robustness. Promisingly, the proposed method can provide novel ideas for more fabric structural parameter analyses.